Final Report Summary - ACOMIN (Advanced Computing for Innovation)
Executive Summary:
The general objective of AComIn is to strengthen the research and innovation capacity of the Institute of Information and Communication Technologies (IICT), Bulgarian Academy of Sciences (BAS) by increasing the knowledge and skills of IICT researchers in advanced areas, by purchasing modern research infrastructure as well as by development of innovation capacity. AComIn should help the institute to successfully accomplish its strategic mission: by 2016, i.e. 5 years after its creation, IICT has to become a leading RTD Centre in Eastern Europe, providing facilities and working conditions comparable to the average standards of the EU Centres of Excellence in ICT. The institute will support the sustainable regional and national growth and employment by providing RTD results to advanced industrial organisations and users from the public sector.
In terms of human capacity, IICT managed to improve its human resources. AComIn project attracted 16 incoming post-doctoral researchers, who worked in IICT with long term contracts, and eight visiting professors who were recruited in IICT with short term employments. Four of the post-doctoral researchers remained in the institute after AComIn end. As for the staff of the institute with permanent contracts, comparing the figures before and after AComIn (2011 vs. 2016), overall the number of permanent staff remains the same but the average age is now between 40 and 50 compared to 50-60 in 2011. Many female researchers obtained scientific degrees and habilitations; new young researchers and doctoral students were attracted. Moreover, IICT offers continued research education to obtain a PhD with independent preparation or paid PhD study to young experts coming from the industry. This illustrates the capacity building and multiplicator role of IICT.
In terms of publication activity, all research teams in AComIn produced internationally acknowledged results. Several research directions (not envisioned in the original proposal) emerged as result of new ideas brought by the new postdocs. In the category “papers with IF and SJR with DOI” AComIn scientists published 161 papers. The project actually implied for 3 years more than doubled indicators as follows: Total IF, Average IF, and the SJR SCImago Journal & Country Rank. Another effect is the strong reduction of the number of publications in domestic Volumes.
The equipment acquired for the SmartLab is the most visible and understandable asset of IICT. It clearly complements the existing high performance computing core and extends significantly the research tasks that can be tackled using the new devices. The activities triggered by the SmartLab acquisition highlight that this is a perfect mechanism to attract additional interest and to generate industrial projects and collaborations. SmartLab reinforces the existing research with complementary activities. This infrastructure opens a clear perspective for cooperation with the industry and makes IICT one of the key actors in advanced computing in South East Europe.
The innovation activities of IICT are considerably extended in 2012-2015. There are six projects in AComIn-related topics, completed within the Competitiveness Operational Programme, coordinated by Bulgarian SMEs which collaborate with IICT as a research partner providing innovation. Based on AComIn results, IICT performed eight projects for contracted research with seven companies and one client from the public sector, in the areas of material analysis, audio processing and 3D modelling and simulation. Fifteen collaboration project for joint research and development of pilot prototypes were accomplished with academic organisations, users from the public sector and companies. Eleven patent applications of six inventions were submitted to the Bulgarian Patent Office, the World Intellectual Property Organisation and the European Patent Office. Certificates for one Registered design and one Utility model were issued by the Bulgarian Patent Office.
The dissemination activities within AComIn (16 Scientific Forums in the project research areas, twenty Technology Transfer events, yearly Doors Open Days, numerous promotional materials including three movies and various media reactions) made the project visible not only in the research community but in the society at large.
In this way the AComIn project had a definite, direct and clearly visible positive impact through significant strengthening of human potential across all units involved in it. This impact concerns not only personnel directly involved in the project, but all other researchers of the IICT BAS. The indirect effect comes, among others, from strengthening the international reputation of the IICT BAS as a prime research institution, and making it a key player at the regional level.
Project Context and Objectives:
The Institute of Information and Communication Technologies (IICT), Bulgarian Academy of Sciences (BAS) (http://www.iict.bas.bg) is an independent legal entity established on 1 July 2010. It arose after the Academy has been reorganised, as recommended in its International Evaluation performed by the European Science Foundation/ALLEA (November 2009). As a part of the reorganisation, involving extensive personal attestation and structural optimisation, the IICT was founded with the aim to integrate the research teams from three former Institutes of the Academy: the Institute of Parallel Processing (IPP), Institute for Information Technologies (IIT) and Institute for Computer and Communication Systems (ICCS).
The IICT is a self-governing permanent research unit of BAS established to perform the following major tasks: (i) to carry out basic and applied scientific research; (ii) to train students, on a contractual basis with the higher schools, granting them a possibility to participate in the scientific research process; (iii) to provide postgraduate tutorship to the PhD students and professional researchers. The IICT governing bodies are: (i) the Assembly of Research Scientists, (ii) the Scientific Council and (iii) the Director. The legal basis of the IICT-BAS is legally grounded by the (i) Law of Bulgarian Academy of Sciences and (ii) Statutes of the Bulgarian Academy of Sciences.
IICT-BAS has permanent staff of about 100 scientists and carries out research in different areas. The Institute is leader of two national infrastructures: the supercomputing one and the CLARIN-BG infrastructure for development of language resources and language technologies for Bulgarian. The AComIn project is generally focused on Advanced Computing, Language and Semantic Technologies, Image and Signal Processing as well as Optimisation and Intelligent Control. Nine of the 13 IICT departments worked in the four AComIn topics as follows:
- In Advanced Computing: Department of Parallel Algorithms, Department of Scientific Computations and Department of Grid Technologies and Applications;
- In Language and Semantic Technologies: Department of Linguistic Modelling and Knowledge Processing;
- In Image and Signal Processing: Department of Mathematical Methods for Sensor Data Processing;
- In Optimization and Intelligent Control: Department of Embedded Intelligent Technologies, Department of Intelligent Systems, Department of Hierarchical Systems and Department of Modelling and Optimization.
The general objective of AComIn is to strengthen the IICT research and innovation capacity by increasing the knowledge and skills of IICT researchers in emerging areas as well as by purchasing modern research infrastructure. AComIn should help the institute to successfully accomplish its strategic mission: by 2016, i.e. 5 years after its creation, IICT has to become a leading RTD Centre in Eastern Europe, providing facilities and working conditions comparable to the average standards of the EU Centres of Excellence in ICT. The institute will support the sustainable regional and national growth and employment by providing RTD results to advanced industrial organisations; it will be a focal point of high-quality research training in South-East Europe.
More specifically, the project objectives are as follows:
- Strengthening the IICT Human Potential by recruitment of incoming experienced researchers, foreigners and Bulgarian nationals who have left the country, including long-term employments of 7 post-docs;
- Purchase of modern equipment for setting up a Smart Periphery Lab that enables to fully exploit the potential of the existing IICT core Grid infrastructure and the Bulgarian Supercomputer. Organisation and Training of User Communities with company representatives who need deeper expertise and innovative RTD approach in their areas of activities;
- Networking with the partnering organisations and knowledge exchange with established ICT experts as well as participation of IICT researchers in prestigious international scientific events and exhibitions;
- Building an IICT Innovation Strategy and development of innovation potential and IPR management skills. Patent applications to the European Patent Office and the Bulgarian Patent Office are foreseen as well;
- Organisation of various dissemination activities to spread out the project results and the IICT excellence: scientific Conferences and Workshops, Technology Transfer events oriented to User Communities and innovation-absorbing Bulgarian companies, Information Days and Doors Open Days, a non-scientific Stakeholder Meeting and a Round Table, as well as publishing books and monographs, leaflets and Newsletters, posters and other promotional materials, making three movies for dissemination via TV channels, YouTube, Facebook etc.
- Evaluation of IICT by international EC-selected reviewers, scheduled for months 35-42. Four independent reviewers assess the IICT-BAS overall research quality and capacity in the post-evaluation facility in order to assign a "quality label" to the institute.
AComIn is focused on research topics which are fundamental for ICT and enable radical progress and development of novel applications. The state-of-the-art frontiers of these technologies tackle computational tasks with high complexity which underpin the challenges of Horizon 2020. IICT has proven record of excellence in these synergistically connected areas and it is expected that the project will strengthen further the horizontal collaboration within the institute. Advanced computing becomes a broadly-used technology relatively recently because it requires substantial infrastructural developments; thus the available IICT capacity in advanced computing is a chance for the other IICT departments to quickly switch to most modern computational paradigms and to generate state-of-the-art results and corresponding innovation. More precisely, the scientific focus of the project is defined by the following research fields:
- Area 1: Advanced computing and Finite Elements applications including multiscale and multiphysics simulations of strongly heterogeneous media with strongly nonlinear and/or anisotropic behaviour as well as high-performance computing in engineering and environmental problems.
- Area 2: Monte Carlo methods, algorithms and distributed computing including sensitivity analysis of large mathematical models as well as advanced applications of Monte Carlo simulations in computational physics and environmental sciences.
- Area 3: Multimodal enrichment of voice communication by investigation a new generation of intelligent voice-activated technologies that focus not only on the precise recognition of spoken words but on interpretation of meaning and context to deliver more accurate speech recognition results.
- Area 4: Large-scale approach to multilingual terminology available in public sites by integration of separate resources, taking into account the ambiguity and the complex links between terms, and producing dynamic reference collections that will be publicly available at the AComIn site.
- Area 5: 3D modelling and recognition in biometrics applied to 3D face recognition based on large DB of face images, lips dynamics processing to help speech segmentation and recognition, iris dynamics and colors to enrich current eye tracking techniques and/or iris diagnostics, etc.
- Area 6: Digital preservation of cultural heritage for research and education by integration of facilities provided by Smart Lab devices and advanced computing solutions.
- Area 7: High spatial resolution based on near-field focalisation by using knowledge (delivered by the Smart Lab acoustic holography) about sources of noisy signals and any sound field descriptor such as sound pressure, sound intensity, or particle velocity as a function of position.
- Area 8: Hierarchical optimization in real time applications related to transportation systems, in particular for solving problems of optimal control, real time decision making, and on-line resource allocation in communication and transportation networks.
- Area 9: Energy efficient production technologies by using the Smart Lab devices and related IT methods for the precise study of dynamical behaviour of complex industrial systems and processes, further development of energy efficient technologies and improving the production quality.
- Area 10: Maintenance of industrial facilities operating in aggressive environment by creation of IT models for diagnostics and predictive maintenance of facilities operating under high risk of incidents. This includes modelling the dynamic behaviour of complex industrial systems and processes using the Smart Lab devices.
As for the selected equipment, which upgrades the computational facilities of the institute, the AComIn objective is to purchase devices that ensure the "data autonomy" of IICT, i.e. devices that enable flow of real-world data for the IICT research tasks. The core computational infrastructure of IICT integrates GRID clusters and the Bulgarian supercomputer purchased with considerable national funding within various programmes. However, insufficient advanced ′periphery′ (in a broad sense) to the HPC core was available. IICT needed modern devices to acquire data about objects and processes that might be of interest for the Bulgarian high-tech industry. Therefore AComIn plans to strengthen the IICT research infrastructure by purchasing SmartLab: a set of complementary high-tech devices that enable 3D input (by measuring shapes, microstructure, sound, temperature), 3D output (vizualisation, 3D printing), study of system dynamics, a speech processing studio, as well as an integrating server with software for traffic simulation and modelling. These devices enable highly innovative scientific activities related to material sciences, energy, health, industrial control and optimisation etc.
Project Results:
I. Scientific Foreground
AComIn foreground is defined in terms of new knowledge, generated in modern IT areas as result of the project research activities and technological achievements. The lists of most representative publications and exploitable knowledge items, presented in this Final report, support the claims for innovation and originality.
1. Wigner Monte Carlo Algorithms for Quantum Transport in Nanoelectronics
Efficient Monte Carlo Methods (MCM) are of great importance for the analysis of large-scale computer models. Because of their robustness, MCM is the only viable method for a wide range of high-dimensional problems ranging from Atomic Physics to Finance. MCM is a powerful tool for sensitivity analysis of large and very large mathematical models.
Our research in the frame of AComIn is focused on ground-breaking applications like studying of special nano-structures considered as potential candidates for future quantum computers, where we have considered technologically produced structures. The miniaturisation of devices did bring the developers in the realm of quantum transport. While technologically important quantum effects were only seen as perturbations in the past, today they are so relevant that specialists are exploiting them as principal effects to make new innovative devices work. The MCM has been extremely successful in the description of such devices and effects. The outcomes could be of great impact in the semiconductor community giving the creation of the first fully quantum simulator able to take into account also scattering effects. That could lead to the simulation of 3D devices like FinFETs, nanowires, multi-gate FETs, i.e. devices that are the most likely to be the candidates to substitute the MOSFET technology that is suffering from the miniaturisation effects. Those devices are already considered by industries such as Intel and AMD, as publicly advertised. The code that implements our MC algorithm is parallelised using the MPI library. This allows us to maintain the portability among different operating systems and/or architectures including available computational infrastructure at IICT-BAS.
The applications we are sealing produce advancements of three species: Development of physical insights in the field of Silicon based quantum computing devices; Development of a completely new numerical technique for the time dependent simulation of chemical systems; Development of mathematical advancements and insights in the theories of quantum transport. The following research activities were performed: (i) The research has been focused on the development of a time-dependent MC algorithm. Our MCM represents the first successful attempt in the world to simulate the Wigner equation time-dependently and multi-dimensionally. Groups at the University of Antwerp (Belgium) and in TU-Wien (Austria) are now using this new method on a daily basis. (ii) The two-dimensional Wigner MCM (ballistic regime) has been applied to the study of ordered and disordered arrays of dopants in order to explain the experimental results of various groups (in particular the Shinada’s group from ASMeW, Japan). The results are able to explain the experimentally observed enhanced transport characteristics obtained by ordered arrays. (iii) A sensitivity analysis of the coherence length in the Wigner MCM has been carried with success using the concepts of L1, L2 norms and the cosine similarity. The aim of this work is to show the efficiency of sensitivity studies of the design of semiconductor devices. (iv) A modification to Density Functional Theory to include the one-body Wigner MCM has been developed and validated in the cases of Hydrogen molecules in different geometrical configurations, Boron and Lithium. This is a very important achievement since the use of single body Wigner MCM allows for the first time the use of the Wigner formalism in the study of complex chemical systems. Let us mention that the many-body quantum problem was considered (before appearing the Wigner MCM) as unsolvable because of the non-polynomial growth of the computational complexity when other methods are used.
2. Robust Finite Element Methods and Algorithms for Advanced Computer Simulations
The modern Finite Element Methods (FEM) provide a computational technology for solving important classes of multiscale and multiphysics problems in science and engineering. The obtained results concern such challenging topics as novel FEM discretization techniques and robust scalability for problems with extremely heterogeneous and anisotropic coefficients.
The FEM discretization is based on conforming or nonconforming approximations. In both cases, the FEM basis functions should have a local support, ensuring sparsity of the matrices of FEM linear systems. The obtained new discretization/approximation results include: (i) interpolation, cubature rules, and least squares fitting of harmonic functions based on Radon projections; (ii) mixed finite element methods based on weighted Hdiv bilinear forms and equivalence between mixed and nonconforming FEM discretization. The robustness of mixed FEM discretization is studied in terms of uniform boundlessness of the constant in the inf-sub condition and the related LBB (Ladyzenskaja-Babuška-Brezzi) inequality.
Many of the nowadays real-life applications involve large-scale parallel processing of FEM linear systems with millions or even billions of degrees of freedom (unknowns). Such huge computational problems are beyond the scope of commercial software tools. The obtained new results are in the field of scalable preconditioned conjugate gradients (PCG) iterative solution methods. They include new or significantly improved multilevel, multigrid and domain decomposition methods, namely: robust algebraic multilevel iteration (AMLI) and semi-coarsening AMLI methods for strongly anisotropic problems; auxiliary space multigrid method based on additive Schur complement approximation; spectral analysis of geometric multigrid methods and full multigrid schemes for isogeometric analysis; finite element tearing and interconnect (FETI) solvers for non-standard finite element equations based on boundary integral operators.
The obtained results allow for advanced computer simulations and studies of complex processes and phenomena in science and high technologies. Among others, the robust solvers are tested on world-wide recognized SPE (society for petroleum engineering) benchmarks. The applications staying beyond the above theoretical achievements include industrial flows in porous media, as well as advanced civil engineering, biomedical engineering, environmental engineering, etc. simulations.
Some algorithms based on the numerical investigation for the seismic analysis of tall reinforced concrete (RC) Civil Engineering structures, which have been degradated due to extreme environmental actions and are strengthened by cable elements are developed. The effects of multiple earthquakes on such RC building frames are computed. Damage indices are estimated in order to compare the seismic response of the structures before and after the retrofit by cable element strengthening, and so to elect the optimum strengthening version.
A new computational model is developed for the mathematically rigorous analysis of Civil Engineering structures, which have been environmentally damaged and subsequently strengthened by cable-elements. The problem is treated as an inequality one, where the governing conditions are equalities as well as inequalities. The cable behavior is considered as nonconvex and non-monotone and is described by generalized subdifferential relations including loosening, elastoplastic-fracturing and other effects. Using piece-wise linearization for the cable behavior, a linear complementarity problem, with a reduced number of unknowns, is solved by optimization algorithms.
3. Advanced Computing in Dynamical Analysis of Elastic Structures
Understanding in details the dynamical behavior of elastic structures provides valuable information for the engineers, which can be used in the design, maintenance and health monitoring. The parametric study in frequency domain by the concepts of nonlinear normal modes (NNM) and nonlinear frequency-response function (NFRF) presents such knowledge that describes completely the dynamical characteristics of the structure. The necessity of obtaining reliable and accurate results requires consideration of nonlinear physical models and development of proper mesh discretization methods, e.g. finite element methods. As a consequence, the whole process of computing the NNM and NFRF becomes computationally expensive and cumbersome. Its effective parallel implementation on high performance computers is essential and unavoidable for the future development of new technologies for design, optimization and health monitoring of engineering structures.
New parallel algorithms for computing numerically the NFRF and NNM of dynamical systems are developed. The numerical methods are applied to large-scale dynamical systems that arise from space discretization of real-life elastic structures. New results, which show the optimal scalability of the proposed parallel implementation, are obtained.
The proposed numerical approach for dynamical analysis presents an iterative process for computing periodic responses of nonlinear systems by varying a parameter (such as frequency of vibration), determination of stability of the periodic solution, establishment of bifurcation points and following the secondary branches of solutions that arise from the bifurcation points. The computation of the NNM or NFRF is achieved by combination of finite element method, shooting and continuation methods. The application of the finite element method to elastic structures, considering geometrical type of nonlinearity, results into nonlinear system of second order ordinary differential equations. The shooting method computes the initial conditions that lead to steady-state periodic response of the dynamical system while the continuation method defines a prediction for the next point from the bifurcation diagram and enables to pass turning points (cyclic-fold bifurcation points). Additionally, the shooting method requires a time integration scheme, achieved by Newmark's method and solution of nonlinear algebraic system, achieved by Newton-Raphson's method.
New parallel realization of the shooting method is developed. The parallel process involves simultaneously efficient algorithms and basic matrix operations for sparse and dense matrices. Effective implementation on high performance computers considering equal distribution of computations and memory among available processors is obtained. New results with optimal parallel speedup and efficiency of the shooting method applied to nonlinear dynamical systems are achieved.
The proposed new technology gives the opportunity to academia and industry to compute efficiently the dynamical properties of real-life applications on high performance computers.
4. Advanced Methods, Algorithms and Innovations Based on 3D Digitisation and Prototyping
A key topic in this area is the voxel data processing based on computed tomography (CT) images of microstructures. The addressed advanced applications vary from fabless technologies for design of new composite and/or porous materials, trough quality control of technological processes and nondestructive defectoscopy, to high-tech innovations in biomedical engineering.
Accurate segmentation of 3D CT data of porous media is crucial for the numerical simulations and the computation of the material/object’s macro characteristics at the next stage. Due to the highly irregular structure of the segmentation phases and the presence of noise in the image, the classical methods are not reliable and the results of different standard algorithms may differ drastically (even on up to 50% of the data). A completely new direction of image segmentation is investigated, where some physical properties of the scanned specimen are incorporated in the segmentation process as constraints. In particular, the volume of the solid phase can be determined from the material’s density and weight measurements and can be a priori prescribed (mass conservation), while the solid phase itself, should be connected whenever the specimen is a single material piece.
Two different families of 2-phase image segmentation methods are proposed, analyzed and implemented. The conducted numerical experiments are promising in terms of both improved accuracy and decreased computational time. The computational efficiency is further optimized. MPI parallel implementations of the new algorithms on the high performance cluster at IICT-BAS are developed. It is important to notice, that experimental measurements of the macro characteristics of porous materials could be rather expensive and not always possible. In such cases, virtual material design is a modern research direction that significantly speeds up the analysis and the derivation of new porous materials with a priori given physical properties.
New methods, algorithms and software (including parallel) implementations for numerical upscaling are developed. The computed anisotropic tensors of material properties are beyond the nowadays abilities of the lab equipment for measurements and testing. The reported applications include: (i) numerical homogenization (upscaling) of heterogeneous anisotropic linear elastic materials; the studied specimens vary from innovative porous metals (lighted aluminums) to trabecular bone tissues, where the impact of osteoporosis is inspected; (ii) calibration of biomechanical characteristics of hepatic tissues and related parameters for radio-frequency tumor ablation simulation.
A completely new innovative methodology of 3D digitalization and prototyping is developed. Here, we are able to report the obtained extremely promising results concerning a lightweight 3D printed polymer prototype antenna, metallization and experimental test at 14-18GHz.
5. Language and Semantic Technologies
Automatic Processing of Image Annotations in Large-Scale Image Databases. The main challenge in the emerging area “Language and Vision” is how to produce (automatically) high quality annotations that describe the image semantics, with least effort and minimal costs. Automatic assignment of keywords to images is a rather difficult task and current solutions based on machine learning are unsatisfactory.
We proposed an approach for post-editing of keywords which are assigned to images. The suggested post-editing tackles "noisy" tags: mistakes, inflexions, doublets, normalization of keywords-names, abbreviations, and various linguistic and punctuation phenomena that occur at token level. Further we suggested employing linguistic resources that define English lexical semantics (WordNet) for consolidation of tags by removing “unnecessary” annotations.
The emotional classification of images depends on the individual opinion of each person but we proposed and investigated an idea how to compute image sentiment scores using external resources. We presented an approach for analysis of sentiments and emotions in image tagging using SentiWordNet as an external linguistic resource of emotional words. Calculating a "sentiment score" for each image, the system classifies images into three classes (positive, negative, and neutral). Our approach works with 63.53% precision, 58.7% recall and 61.02% F-measure which is coherent with the results reported for sentiment analysis in general.
Automatic tag sense disambiguation is another hot issue in the “Language and Vision” field. In general automatic tag disambiguation is a fundamental issue for modern management of digital resources because content objects with identical tags can be linked together allowing users to search for similar or related content in blogs, videos, image collections, learning objects or categories of web-pages. Tags ambiguity leads to inaccuracy and misunderstandings. We proposed an integrated method for tag disambiguation of images and showed its effectiveness for arbitrary tags in English with nominal and verbal senses using WordNet as an external resource defining tag senses. The main advantage of the method is the combination of known approaches which ensures some non-zero result for all annotation keywords belonging to WordNet. The evaluation is done on a corpus of about 5,600 tags with some 10,000 senses and showed that the correctly disambiguated tags are 95% for professional images and 87% for social images (that are often annotated manually).
Application of Educational Data Mining for Analysis of the eLearning Portal UCHA.SE. Educational Data Mining allows the discovery of new knowledge based on learners’ usage data in order to help validate and/or evaluate educational systems, to potentially improve some aspects of the quality of education and to lay the groundwork for a more effective learning process. Its potential explains the significant interest it raised in the educational community and the developers of online learning environments. A pilot research projects was carried out in the frame of AComIn. We applied educational data mining to the most popular Bulgarian educational site, UCHA.SE which offers instructional videos and practice exercises in Bulgarian language. The site can be used to support formal and informal education, that is, in schools and for self-learning. Currently it offers more than 4,300 video lessons in 17 subjects, including the basic K-12 subjects, as well as Introductory level English, German, French, and Spanish, and Introduction to Programming. As of September 2015, the site has more than 400,000 registered users - students (including college students), teachers, and parents. The lessons are seen more than 13 mln times.
The goal of the project was to improve the quality of educational services and the subscription rate for the site by leveraging educational data mining. In order to achieve these goals we extracted a “learner model” by applying BITool (business intelligence software supporting a multidimensional data model) to UCHA.SE relational database in which system logs and students’ performance data are stored. As a result the constructed “learner model” contained 30 attributes that reflect four type of information:
- general information about the user (e.g. age, gender, etc.),
- information related to user activities (e.g. “Percent of the total number of all taken exercises”, “Average length watched videos per day”, “Consecutive days the user has accessed the site”, etc),
- information related to user knowledge and skills (e.g. “Percentage of all attempts of these exercises, for which the total score exceeds the threshold of 75 points”, “Percent of the watched videos, which the user also commented”, etc, and
- information about the use of some gamification elements proposed by the site (e.g. “leadership level”, “charisma”, “influence”, etc.).
The goal for increasing the site subscription rate was reformulated as a classification task for predicting whether the user will renew his current subscription in the period of 3 month after its expiration. The classification task was solved by means of a machine learning approach. The training data reflects the user model and was constructed as of a concrete moment in time (end of February 2015). Since the learning data was highly unbalanced (only 18% of all data represented “positive” examples, i.e. the users who renewed their subscription in due time). The learning dataset was improved by combination of subsampling and undersampling methods. The predictive model was created as a combination of rules learned by two machine learning algorithms - JRIP and CN2, as JRIP rules generalized “positive” examples, and CN2 rules – the “negative” ones.
The predictive quality of this easily understandable to the end-users rule-based model, has classification accuracy 86.1% and F-measure 68%, respectively, which is close to the predictive quality of the model produced by one of best classification algorithms - SVM (87% and 68%). However, the SVM model is not understandable by the end-user.
News Media Analysis and Creation of Language Resources. The contemporary Natural Language Processing (NLP) applications, such as the analysis and extraction of relevant information from huge streams of data as well as context-aware Machine Translation, require high quality knowledge-rich processing. The Web provides an extensive volume of relevant knowledge, but most of it is in a textual or semi-structured format. Also, in the last decades the language and culturally specific information constantly increase. This calls for the development of language technology and resources (LTR) for national and local languages. These LTR need to reflect the language and culture specific features of the raw data. In order to support the interaction with rest of the world they are aligned by rule with LTRs for other languages.
Together with partners from Vrije Universiteit, Amsterdam (Computational Lexicology and Terminology Lab (CLTL)) we have been adapting the IICT-BAS in-house NLP processor for Bulgarian to analyse the events and stories in big news media data.
The NLP processor performs the following analyses: tokenization, POS and morphosyntactic tagging, lemmatisation, syntactic parsing, named entity identification and semantic analysis. Our efforts have been invested mainly in the last step as the most complex component of the processing chain. The semantic tagging includes the following submodules: (i) assigning the correct sense from the BTB WordNet on verbs, nouns, adjectives and adverbs; (ii) assigning the correct semantic role.
The first submodule uses the BTB WordNet, which was developed and is being extended within ACOMIN. It has been made freely available at the following link: http://compling.hss.ntu.edu.sg/omw/
The second submodule was developed through the WordNet verb semantic classes that were mapped onto the Bulgarian valency frames. The initial general semantic role labels have been distributed over BulTreeBank (BTB), and now the specialization process is in progress. Additionally, the first automatic transfer of Predicate Matrix (http://adimen.si.ehu.es/web/PredicateMatrix) was performed on Bulgarian data from the Setimes parallel corpora.
For the purposes of semantic annotation and disambiguation, the knowledge-graph tool UKB, developed by the IXA group, Basque country, was used. The experiments have been done with Bulgarian (BulTreeBank) and English data (SemCor). The results show improvement when syntactic relations are added for both languages.
Language Technologies Applied for Generation of a Diabetes Register. This pilot project for large-scale application of language technologies was performed jointly with the University Specialised Hospital for Active Treatment of Endocrinology ″Acad. Ivan Penchev″ (USHATE), Medical University – Sofia. USHATE is authorised by the Bulgarian Ministry of Health to host an anonymous Register of diabetic patients in Bulgaria. This Register contains 28 indicators of diabetic patients including age, sex, codes of diagnoses of diabetes and its complications, diabetes duration, risk factors, data about compensation, laboratory results, hospitalisations and prescribed medication. The AComIn experienced researchers helped for the automatic construction of the Register using language technologies and business analytics tools.
The register is automatically generated from a Repository of more than 112 million pseudonymised reimbursement requests (Outpatient records, ORs) submitted to the National Health Insurance Fund (NHIF) in 2012-2014 for more than 5 million citizens, including 436,000 diabetic patients. ORs are semi-structured in XML format; in each file some tags contain free-text fields with important explanations about the patient: “Anamnesis”, “Status”, “Clinical examinations” and “Therapy”. In AComIn, pilot prototypes for large scale processing of Bulgarian clinical texts were developed that extract automatically most essential data.
We developed a drug extractor using regular expressions to describe linguistic patterns. There are more than 80 different patterns for matching text units to ATC drug names/codes and NHIF drug codes. This extractor identifies the medication name, dosage, frequency and route of admission. Currently, it handles 2,239 drug names included in the NHIF nomenclatures. For treatment effect assessment we provide a comparative study of patients with insulin therapy, incretine-based therapy and oral medications therapy.
For extraction of clinical examination we designed a Numeric value extractor that processes lab and test results. Using collocation extractor we generate patterns for clinical exam data. After analysis of the positive and negative examples for such data we define the clinical exam data language L by a context-free grammar. For parsing clinical examination data we propose a hybrid method inspired by dependency grammars, constituents and Government and binding theory. The obtained result is a list of attribute-value pairs. For monitoring the compensation of Diabetes Type 2 and risk factors we extract values for body mass index (BMI), weight (W), blood pressure (BP), glycated haemoglobin (HbA1c), blood glucose (GLU), etc.
To identify complex relations between Diabetes Mellitus Type 2, its complications and other chronic diseases we defined a Data mining module for finding (i) frequent patterns and (ii) frequent temporal sequences of chronic diseases. For task (i) we apply modification of the classical frequent itemsets mining algorithms dEclat and association rules generation. We proposed a cascade method that reduces the search space for task (ii). In task (ii) several experiments were done – with no limitations for the distance between events (only the order matters), and with different window limitations between events – 1 to 3 months, etc.
Speech Processing. Modern state-of-the art speech recognition systems crucially use a beam search method in huge graph that combines acoustic model probabilities and language model probabilities. A disadvantage of all beam search methods is that they find a suboptimal, but not the best, path in the search space. This leads to propagation of the recognition errors. We have introduced a novel method for real-time lattice rescoring in speech recognition. It improves the performance of speech recognition systems by providing the option to partially compile the word lattice into a deterministic finite-state automaton, making it suitable for the rescoring step in the speech recognition process. In contrast to the widely used n-best approach our method permits the consideration of significantly larger number of alternatives within the same time-constraint thus providing better recognition results. A description of the new method is presented together with empirical evaluation of its performance in comparison with the n-best method. The achieved WER reduction is up to 3.77% at a p-value below 3%. An important advantage of our method is its applicability for real-time speech recognition. A pilot prototype of speech-to-text system for Bulgarian is demonstrated.
AComIn supported also the development of a novel effective method for searching similar audio segments in large audio collections. This method is based on original ideas for similarity search and ensures high reliability in noisy environments and exclusively high speed of searching for similar audio records. A pilot prototype for reliable automatic annotation of subtitled audio signals at phonemic level is demonstrated as well.
Annotated corpus of Bulgarian speech. Using the Speech Lab (purchased in AComIn) a multi-signal corpus of Bulgarian speech was developed. The corpus incorporates sounds, laringograph signals and physiological signals for some of the readers. The sentences in the corpus are segmented automatically by original software with manual correction of the accuracy. The resource is available at http://lml.bas.bg/BulPhonC.
6. Signal and Image Processing
Neuro-fuzzy Approach. In the considered period a theoretical investigation of effect of Intrinsic Plasticity (IP) improvement of Echo State Networks (ESN) reservoirs were summarised and commented. It was observed that IP training “captures” input data structure into the reservoir steady state in a way that could be useful for clustering purposes. In search of explanation of these results the Memory Capacity of Reservoir Equilibrium State (MCRES) was introduced. Achieved due to IP tuning MCRES of ESN reservoirs with different sizes was investigated using three benchmark artificial data sets. Two applications of neuro-fuzzy approach were realised. The first one visualises 3D acoustic waves propagation based on multi-dimensional data clustering. The second one implements adaptive critic designs for optimisation of fuzzy rule base parameters, designed to adapt the measurement noise covariance matrix. The proposed approach assures fast tuning of Kalman filter and improvement of the quality of its estimates in changing working conditions of the MEMS in real time application.
Enhancement of Acoustic Noise Source Localization and Identification. Results in several directions are achieved:
- Generation of the acoustic images in range - cross range - elevation coordinates. The approach enables to determine automatically the range to source (Acoustic camera software require measuring it in advance). The approach enables to localize the source with frequency bandwidth about 10 kHz, which is processed simultaneously;
- Enhancement of angular resolution of the acoustic images up to 2.72 (2.56) times in respect to "delay and sum" beamforming (built-in standard beamforming). The resolution enhancement was obtained for center frequencies from 1 kHz to 18 kHz, at bandwidth 10% and 23% of the center frequency. It was obtained based-on modified Capon`s approach, at range span comparable to dimensions of microphone array of the Acoustic Camera;
- A software tool for bearing noise estimation is prepared. It was applied for development of new tools for non-destructive testing of the bearings;
- Improvement of resolution of SONAH method in high frequency range, based on virtual microphones. The small number of microphones in Acoustic Camera limits SONAH method resolution. Virtual microphones are defined by using autoregressive relation for signal interpolation. In this way measurement surface density and/or area are numerically enlarged and resolution improvement of SONAH has been achieved;
- Fast direction of arrival estimation. A single frequency can be calculated on the basis of at least three data samples. Having a set of local frequencies one can calculate a set of direction angles and draw several rays from corresponding points. Estimation of the shift parameter of the distribution of these points gives the source coordinates;
- Multichannel frequency estimation. The task considers estimation of the power spectral density matrix, which consists of elements related to auto-spectrums and cross-spectrums for each pair of channels. By using the maximum likelihood approach the autoregressive parameter estimator has been synthesized. It has been shown that in the single channel case it is equivalent to the known modified covariance method obtained by a least squares approach. The simulation results demonstrate that the new estimator has similar precision in the middle frequency range to a simple one averaged by multiple channels, but the proposed method becomes much better at low frequencies. Additional attention has been paid to influence of the impulsive noise. The proposed solution assumes removing of noise impulses by the separation of the input sample and interpolation of the detected corrupted values. It has been shown that the proposed approach has advantage over the conventional filtering techniques. Statistical model of the signal sample, description of the detection criterion, flowchart of the filtering procedure and comparison results are published.
Research results achieved in Biometric Authentication. Ear Biometrics is a young topic that is intensively investigated recently. Even if ear has a uniform distribution of color, human external ear characteristics are considered enough unique to each individual and permanent during the lifetime of an adult, almost like fingerprint and/or iris data, but ear data can be gathered without any discomfort for people. A new method representing most essential data of human ear has been proposed, namely the matching potential for ear discrimination was tested via Extended Gaussian Image (EGI) representation and experimentally proved using 3D ear data gathered via the 3D Scanner of the AComIn SmartLab. Another approach to Biometrics Authentication by Ear has been proposed based on an original CBIR (Content Based Image Retrieval) method of IICT-BAS for rapid and reliable recognition of 3D objects using a database of precedents. Each object of interest for recognition is presented in the DB through a sufficient number of 2D projections (images), each from a different view point. Input images for recognition are obtained from conventional 2D cameras, and the most appropriate ear data are retrieved from the DB trough the CBIR method to access that is fast enough and noise resistant. We have already applied successfully this appearance-based approach two times: (i) for recognition of palm signs from a sign language alphabet and (ii) for human face recognition. The effective application for human ears became possible thanks to the 3D scanning technology available in the SmartLab of IICT-BAS (in the frames of AComIn project).
Speech Detection. A new robust feature for contour-based speech endpoint detection is proposed. It is called Group Delay Mean Delta (GDMD) feature and combines the properties of the known Modified Group Delay Spectrum (MGDS) and the original Mean Delta (MD) approach. The effectiveness of proposed feature is experimentally evaluated in the fixed-text Dynamic Time Warping (DTW) - based speaker verification task with short phrases of telephone speech. In comparison with three well know features - Modified Teager Energy (MTE), Energy-Entropy (EE) and MD feature - the experiments have shown that the GDMD feature demonstrates the best performance in endpoint detection tests in terms of verification rate.
Signature verification. A combined off-line signature verification approach is proposed using Neural networks, where global and grid signature features are combined to generate specific feature sets, one for each person represented in the verification system.
Video Stabilisation. A method and robust algorithm for fast 2D video stabilization for handheld devices in real-time is proposed. The BSC (Boundary Signal Computation) chip of TI (Texas Instruments) is essentially used (or emulated herein) for searching of correlations between the 1D integral projections, horizontal and vertical ones, by a SAD (Sums of Absolute Differences) approach. The proposed method is based on an accurate vector model allowing interpretations of increasing complexity for the transformations among frames. Experiments, conducted on testing video clips, are very promising for the future R&D of the method. This method has been extended for the problem of video stabilisation ‘in a point’, related to providing a static background for correct measurements with high-speed industrial cameras, like the one available in the SmartLab of AComIn. This new approach is called “3x3OF9x9” and combines (fuses) the basic method we use with the most lightweight version of the known Optical Flow approach, applying simple Otsu segmentation for eliminating the influence of moving objects in the video. The obtained results show better stabilisation (low that 1% instability for frame size of 640x480) in comparison with commercial software packets, like Warp Stabilizer of Adobe After Effects CS6.
7. Optimisation and Intelligent Control
A New Type of Chemical Nickel Coatings Including Nano Elements. The research on this topic includes a comprehensive study of abrasive wear and starting friction of nickel chemical coatings containing nanosized particles of silicon carbide (SiC) and aluminum oxide (Al2O3) of various sizes. It was carried out with the purpose of replacing old and non-ecological industrial technologies for laying a chrome based coating on different working surfaces. As a result, an advanced technology is developed and tested for renovation of extruding shafts used in industrial manufacturing of sheet materials, i.e. PVC, Plexiglas, other plastics, etc. The commonly used renovation technologies in this area are known to be exclusively harmful due to the waste cyanide products. Our innovative approach overcomes this drawback by applying advanced nanotechnology for nonelectric chemical laying of a nickel coating with included nanoparticles. Enhanced tribological characteristics are obtained due to the embedded micro- and nanosized particles. The technology is environmentally harmless and the new coating features an increased hardness and wear resistance. A special mechanical manipulator is developed as a part of a robotized systems for inspections of coatings with high mechanical wear-resistance and surface-smoothness.
Intelligent Methods for Technical Diagnostics. Technical diagnostics is a classical interdisciplinary area where results from narrow scientific sciences and specific trends are applied: probability and statistics, pattern recognition, decision making, mathematical logic. Modern industry requires efficient fault discovering and isolation solutions in process equipment service which is a real-world problem of typically ill-defined systems, hard to model, with large-scale solution spaces. Design of precise models is impractical, too expensive, or often non-existent. To cope with this problem, methods and tools are developed for intelligent diagnostics, monitoring and decision making for real detection of potential abnormalities in technological processes without applying expensive diagnostic apparatus. Especially the usage of AI-based methodologies enables us to deal with imprecise, uncertain data and incomplete domain knowledge typically encountered in practical applications. Several types of diagnostic approaches with different structures are applied to a mill fan device in the Bulgarian Maritsa East 2 Thermal Power Plant being the largest thermal power plant on the Balkan Peninsula. The possibility to predict eventual damage or wearing out without switching off the device turns out to be of great importance for providing faultless and reliable work of the plant.
Traffic Optimization in Communication Networks. In communication networks, a crossbar switch node routes traffic from the input to output where a message packet is transmitted from the source to the destination. The randomly incoming traffic must be controlled and scheduled to eliminate conflict at the crossbar switch. The goal of the traffic-scheduling for the crossbar switches is to maximize the throughput of packet through a switch and to minimize packet blocking probability and packet waiting time. In this area of research, new models of the incoming traffic in crossbar switch nodes are proposed including several families of input traffic patterns. As compared to the modeling methods and techniques available in the current literature, an important advantage of our models is that the generation of the traffic patterns does not depend on the type of hardware and software tools. Simulation studies and evaluations for the throughput of a switch node by the proposed family of patterns are performed using the grid-structure of the IICT-BAS. A numerical procedure for computation of the upper bound of the throughput is suggested which enables us to estimate the maximal throughput of the switch for different traffic scheduling algorithms.
Target Detection and Parameter Estimation. The research on this topic includes a thorough study of the latest trends in the design of highly efficient and fully automated systems for processing radar data in terms of a priori uncertainty about the targets and disturbances. Parameter estimation and high-speed small targets detection in randomly arriving impulse interference environment is a challenging task in radar systems design. Conventional radar detection architectures like spectral discrimination and non-coherent integration have been considered and used but with limited success. Within this framework, an improved target detection architecture is developed making use of highly efficient Hough velocity estimation technique. The proposed algorithm is based on a Track-Before-Detect processing method, which allows for the previously collected data to be used in the target detection process and parameter estimation. The presented technique has a lot of advantages compared to traditional ones. The obtained results can be successfully applied for radar target parameter estimation as well as in the existing communication network receivers making use of pulse signals.
Mechatronic Systems for Industrial Automation and Robotics. The investigation on this topic is aimed at developing new mechatronic systems incorporating mechanical constructions, electronic modules and intelligent software for automation of technological processes and operations. Main subjects of study are robots, dedicated for multisensory environment perception and exploration, measurements and samples taking, discovering and putting a mark on the objects as well as environment interactions like transportation, carrying in and out of equipment and objects. An important result obtained in this area is an advanced modular system developed for the mechanical construction of the mobile robots. It includes special programmable logical controllers for robot control and electronic modules for the wireless communication. New methods, means and algorithms for adaptive environment behaviour and group control of mobile robots are also examined. An innovative application of the international standard BDS ISO 10825 in mechanical transmissions is also proposed.
Intelligent Transportation Systems. The work on Autonomic transportation aims at meeting the challenge of engineering autonomic behavior in Intelligent Transportation Systems. The research integrates approaches from the disciplines of traffic engineering and autonomic computing. Autonomic computing is inspired by a biological example of body’s autonomic nervous system and it allows for a more efficient management of heterogeneous distributed computing systems. This type of systems are endowed with a number of properties that are generally referred as “self” properties, including self-configuration, self-healing, self-optimization, self-protection or in general term self-management. In this domain, an advanced approach is developed providing a new formal description of the traffic control policy. More complicated formalization is applied for the control of urban traffic systems using bi-level optimization. Such an approach gives potential for increase of the control space of the control problem. The new formalization of the control policy allows for the traffic flows to be controlled in urban networks not only by the green light duration of traffic lights as it is the common manner, but simultaneously by changing the duration of the traffic lights cycle obtained as solution of optimal control problem. This innovation benefits minimization of the waiting time of the vehicles in the urban network, maximization of the traffic outflows, decrease of traffic congestions, minimizing time for travel.
Numerically Effective Kalman Estimator Algorithm for Urban Transportation Network. The research on this topic addresses a problem of vehicles amount estimation in queues, which appear in front of controlled junctions in a congested Urban Transportation Network. The innovative result of the research is an approach, which leads to an effective numerical algorithm of building steady-state Kalman estimator for the store-and-forward urban transportation network model. The evaluation of the numerical algorithm shows its efficiency and applicability for on-line computations in the urban intelligent transportation system. The research presents an original approach, which uses a special structure of the optimization problem data and expands method of resolvent for constructing a stabilizing solution of discrete algebraic Riccati equation.
According to the AComIn Reviewers (their Review is publicly available at http://iict.bas.bg/acomin/docs/deliverables/D6_2.pdf) "all research teams produced internationally acknowledged results. Several research directions (not envisioned in the original proposal) emerged as result of new ideas brought by the new postdocs". It is also a positive effect that the new research topics are in line with merging activities at an international level, so they may grow and extend the institute research potential.
II. Technological Results
In terms of technological and innovation achievements, the results of AComIn show the high potential of the project team to transfer know-how to Users. Here we briefly summarize the figures illustrating the scale of applied research performed in AComIn.
1. Completed projects in the Competitiveness Operational Programme, coordinated by Bulgarian SMEs which collaborate with IICT as a research partner providing innovation. There are six projects tackling AComIn topics that were performed in 2013-2015:
- Project BG161PO003-1.1.06-0001-C0001 "Innovative technology for efficiency evaluation of ERP systems in small and medium enterprises", with beneficiary ERP Bulgaria Ltd. with R&D activities related to the development of software for efficient management of business processes;
- Project BG161PO003-1.1.06-0004-C0001 "Innovative technology solutions for radiofrequency thermoablation" with beneficiary AMET Ltd. Sofia, Bulgaria, with R&D activities related to the development, modern manufacturing and distribution of electronic medical equipment and modules, mechanical parts and units for incorporation;
- Project BG161PO003-1.1.06-0023-C0001 “Analysis and identification of inexplicit relations in large scale numerical data: applications in economics and technological analysis” with beneficiary ADISS Lab Ltd, with R&D activities related to the development of integrated information systems and business analytics tools;
- Project BG161PO003-1.1.06-0023-C0001 "Distributed Information System for Group Control, Distance Diagnostic and Service of Specialised Industrial Robots" with beneficiary SPESIMA GMBH, Sofia, Bulgaria - part of the FRECH Holding, Germany, with R&D activities related to the development of automation systems;
- Project BG161PO003-1.1.06-0023-C0001 "Experimental Development of Software Libraries for Improvement of Image Quality and Stabilisation using Inertial Sensors" with beneficiary MM Solutions AD, Sofia; with R&D activities in the area of complete imaging solutions for mobile camera devices;
- Project BG161PO003-1.1.06-0023-C0001 "Industrial Research for New Technologies in Image Stabilization and Image Quality Enhancement by Implementation of Inertial Sensors" with beneficiary MM Solutions AD, Sofia; with R&D activities in the area of complete imaging solutions for mobile camera devices.
2. Contracted Research
There are eight projects performed for external clients based on results of AComIn research activities, as follows:
- Research and development of models for the distribution of particles with a size of 0.01 - 2000 μm in samples of bulk materials - various types of metal powders and/or nanoparticles by using the laser nano particle sizer Analysette 22 Nano Tech + for measuring the size and distribution of particles. Performed for RETEL JSK, a company dealing with the delivery of complete equipment for the mining industry regarding processes of milling and enrichment of ore materials;
- Research for the development of 3D models with the mobile 3D scanner VIUScan, parameterization of 3D models, processing of the received data, followed byre search, analysis, design and visualization of 3D models of industrial, municipal and archaeological sites with the aim to find innovative solutions and testing new models in the field of 3D printing by using the professional 3D Printer Pro Jet 460+. Performed for ECO BIO ENGINEERING, a company engaged in large-scale modelling and engineering of industrial, archaeological and urban installations;
- Development of models for the distribution of particle with a size of 0.01 - 2000 μm in samples of solutions of bulk materials - various solutions, emulsions and mixtures and/or the contained nanoparticles after filtration and percolation using the laser nano particle sizer Analysette 22 Nano Tech + for measuring the size and distribution of particles. Performed for Diamond Properties, a company performing studies of particle distribution in powders as well as their mixtures and solutions for industrial users;
- Analysis and visualization of the size and distribution of particles in the concentrates of fruit juices and emulsions using the laser nano particle sizer Analysette 22 Nano Tech +, performed for Interaroma - a producer of different clear fruit juices and nectar;
- Construction of 3D models of different types of badges and medals and 3D printing in order to find the proper size. This application is done for B&W Consultant - a company engaged in design and production of promotional materials;
- Research and development of models concerning the distribution of particles with a size of 0.01 - 2000 μm in soil samples - various types of clays using the Laser Nano particle sizer Analysette 22 Nano Tech + for measuring the size and distribution of particles. Performed for the National Institute of Geophysics, Geodesy and Geography – Bulgarian Academy of Sciences;
- CT scanning and microstructure analysis of large series of bone tissue samples, done for an International Research Team;
- Development of a real-life application of an effective method for searching similar audio segments in large audio collections, performed for the Bulgarian TV-related company H-Tech.
3. Applied Research in Joint Research Projects
Collaboration activities with Universities were performed as joint projects dedicated to specific topics, as follows:
- Analyses of thermal images of thermal processes in large electrical circuits and electrical appliances, jointly with the Electrical Faculty of Technical University - Sofia;
- Joint study of high-speed processes concerning the briquetting of metal shavings and powders necessary for the optimization of briquetting of waste products with the purpose of secondary melting. Performed with the Faculty of Mechanical Engineering of Technical University - Sofia;
- Examination of temperature regimes in large buildings to deploy powerful breakers, fuses and distributors, done jointly with the Plovdiv Branch of Technical University - Sofia;
- Study of the high-speed processes of interaction and deformation of different physical bodies by shocks in hit railings and guardrails, performed for the Institute of Mechanics – Bulgarian Academy of Sciences to support the standardisation of protective fences, guardrails, handrails etc.;
- Examination of the temperature image of the cooling process of welding metals for different kinds of welding for medium nano particles in the range of 2000°C– 500°C, performed jointly with the Institute of Metal Sciences – Bulgarian Academy of Sciences;
- Development of models for a distribution of particles with a size of 0.01 - 2000 μm in samples of bulk materials - various types of powders of dried and ground food, performed together with experts from the Bulgarian Agricultural Academy which performs a research project related to dried and freeze-dried foods that are used in extreme sports (mountaineering) and when traveling in aerospace;
- Educational analytics of learning data performed together with the team of the most popular Bulgarian K-12 platform with educational videos and exercises. The results helped them to identify users who are likely to renew their subscription and attract other subscribers;
- Applying language technologies for automatic processing of Bulgarian clinical narratives for automatic construction of the National Diabetic Register, together with the University Specialized Hospital for Active Treatment of Endocrinology "Acad. Ivan Penchev", Medical University - Sofia;
- Non-destructive testing of rolling bearings using the AComIn acoustic camera, performed together with the Faculty of Mechanical Engineering of Technical University - Sofia;
- Microstructure analysis and numerical homogenization of epoxybased nanocomposites based on industrial CT scanning data. Done jointly with the Institute of Mechanics – Bulgarian Academy of Sciences;
- Participation in the project "Study of composition, structure, properties and production of yellow pavement in Sofia" together with the "Bulgarian crystallographic society" and experts from the Institute of Mineralogy and Crystallography - BAS, Institute of Physical Chemistry - BAS, Institute of Catalysis - BAS, Institute of Information and Communication Technologies - BAS, Mining and Geology University "St. Ivan Rilski "and the University of Architecture, Civil Engineering and Geodesy – Sofia. The project was funded by the Sofia Municipality (the yellow pavement is a prominent symbol of Sofia, which was declared national heritage in the beginning of 2014 by the Sofia municipal council);
- Applications of CT Microstructure analysis in Paleoanthropology developed jointly with the Institute of Experimental Morphology, Pathology and Anthropology with Museum – Bulgarian Academy of Sciences;
- Advanced CT study of structural and material properties of fiber reinforced concrete, performed with the Institute of Mechanics – Bulgarian Academy of Sciences;
- Investigations and microstructural analysis of geological materials by scanning with AComIn 3D industrial tomography, performed with the Institute of Mechanics – Bulgarian Academy of Sciences for Ruhr-University of Bochum (Germany);
- Preparation and completing of 3D models of historical figures related to the Battle of Pavia (1525), 3D printing of the figures and printing of 3D tactile matrices that display medieval tapestries to visually impaired people. Performed jointly with the University of Pavia for the Exhibition "The Battle of Pavia" held in Visconti Castle, Pavia, as an accompanying event of EXPO 2015, Milan, Italy.
These applied activities show that the practical results achieved in AComIn are based on the high-tech equipment purchased within the project.
Potential Impact:
I. Potential Impact
Here we present the potential project impact in three dimentions: (i) plans for exploitation of the AComIn research foreground, (ii) socio-economic impact and (iii) wider societal exploitations.
1. Plans for using the AComIn foreground
In Wigner Monte Carlo Algorithms for Quantum Transport in Nanoelectronics: The research carried out in AComIn was focussed on ground-breaking applications that support building up and structuring of a new interdisciplinary scientific community in emerging themes dealing with radically new future technologies. For instance our results involve studying of special nano-structures viewed as potential candidates for future quantum computers since we have considered technologically produced structures. In this way we have studied experimentally nano-structures which are considered as a candidate for a quantum computer. Our research in this fiels directly supports innovations in semiconductor devices. That could lead to the simulation of 3D devices like FinFETs, nanowires, multi-gate FETs, i.e. devices that are the most likely to be the candidates to substitute the MOSFET technology that is suffering from the miniaturization effects. Those devices are already taken into account by industries such as Intel and AMD, as publicly advertised. IICT-BAS plans joint research with Technical University Wien which is particularly focused on magnetic field, entangled electronics, and parallelisation algorithms. A systematic derivation of simulation models accounting for the magnetic field in the Wigner picture is still missing. A major challenge is the incorporation of the electromagnetic potentials and the choice of the gauge. A computationally feasible theoretical approach is pursued, which allows an efficient inclusion of the vector potential in the signed particle method. Entanglement characterises the state evolution in electron lenses, mesoscopic (Aharonov-Bohm) rings, and Quantum Point Contacts (QPC's). Electron transport in such structures comprises phenomena with yet not explored physical and application aspects. For instance, there is no relevant theory which can explain the experimentally observed behaviour in QPC's. All these issues are included in our research agenda for the near future.
In Robust Finite Element Methods and Algorithms for Advanced Computer Simulations: The future plans for research will target top achievements in robust FEM discretization and solution methods. The obtained theoretical results for strongly heterogeneous elliptic problems of high contrast and high frequency will be further developed to the case of coupled problems including models of fluid-structure interactions, flows in deformable porous media, thin plates and shells, etc. The expected new results will play a key role in such high tech applications as new constructive materials, clean technologies, nano membranes, energy efficient combustion in inert porous media. The related problems are inherently nonlinear, three dimensional in space and time dependent. This means that HPC will be without alternative in many of the real life applications. The big challenge here concerns the development of scalable parallel algorithms for the new generation heterogeneous systems including GPU and MIC accelerators.
In Advanced Computing in Dynamical Analysis of Elastic Structures: The research plans for next period will be focused on development of efficient methods, algorithms and codes for dynamical analysis of more complicated structural elements. The first step will address the related problems in the theory of plates. The standard model is described by PDE of fourth order which beans that the FEM approximations should provide higher smoothness of the test functions. Even for linear problems this leads to serious increase of the size of discrete problem from one side and robustness problems from another. One alternative approach (which will be preferred) is based on development of efficient methods and algorithms in terms of Mindlin–Reissner plate theory. Mixed FEM, nonconforming and reduced integration FEM will be studied. The obtained results will be generalized to the case of shells. In both cases, plates and shells the methods should be robust with respect of the thickness which is assumed by default to be small. The parallel implementation of the algorithms (e.g. the shooting algorithm) will be required in many real life applications in machine and civil engineering.
In Advanced Methods, Algorithms and Innovations Based on 3D Digitisation and Prototyping: This is a completely advanced topic. Some of the reported results are really pioneering. There are (at least) two kinds of challenges in the future research and development activities. The first one is related to the extremely fast development of the equipment for 3D digitalization and prototyping. The sub-micro computed tomography is already available. This means in particular new qualitative and quantitative requirements to the algorithms for segmentation of CT images. Even the formulation of the problem of mass conservative segmentation for multicomponent porous media is a separate challenge. The next plans in the related field of numerical upscaling will be targeted to homogenization of nonlinear properties. The results in the field of numerical upscaling of anisotropic materials have a strong potential for generalizations in the general case of unstructured Big Data analytics. Considering the future plans in prototyping we should notice the new opportunities provided (or expected to appear) in the on-linecontrol of 3D printing with respect to, e.g. changing the material or/and controlling the density etc.
In Language and Semantic Technologies:
Integrating Language and Vision in general requires better understanding of the annotation processes and the psychology of personal behavior in social media where users often assign emotional tags to their photos. Therefore we believe that considerable success in automatic tag disambiguation can be achieved in specific niches e.g. collections of professional images that document mostly material objects denoted by nouns. We plan to extend the present results in the area of disambiguation by integration of further external resources that provide reference sets of senses: like Wikipedia and DBPedia. Then there will be a voting scenario where different disambiguation platforms propose the correct tag sense. Further we plan to make experiments with embedding fine-tuned tags in the image recognition phase when the decision of automatic annotation is made.
Regarding the News Media Analysis and Creation of Language Resources, the results achieved in AComIn will be integrated in the development of a Language Technology and Resources Centre for Linked Open Data. This includes: (i) Further extension of the BTB WordNet through reliable (semi)automatic processes; (ii) Alignment of BTB WordNet with English and other languages wordnets as well as world knowledge databases such as DBpedia, Freebase and other datasets from Linked Open Data cloud; (iii) Further improvement of the knowledge graph performance through the addition of new relations; (iv) Developing of ensemble and supervised methods for solving several language technology tasks; (v) Crosslingual event analysis and information extraction for big news data in Bulgarian and English (also possibly Dutch); as well as (vi) Deployment of the semantic tagging into an MT system between Bulgarian and English.
As for the Language Technologies Applied for Generation of a Diabetes Register, the technologies used at present will be at first extended for automatic extraction of values in the section “Patient Status” in the Outpatient records submitted to the National Health Insurance Fund. Then the plan is to extend the extractors to further chronic diseases in order to generate other registers: like Diseases of the circulatory system, Malignant neoplasms, Mental and behavioural disorders, etc. In addition to the Outpatient records, the extractors should process selected zones in other types of clinical narratives: e.g. Discharge letters and Reports about Clinical examinations issued by Independent Medical Diagnostic Laboratories. A major challenge is the automatic processing of temporal information in clinical narratives: recognition of absolute and relative temporal markers, and ordering them in timelines.
In Speech Processing: A disadvantage of all beam search methods is that they find a suboptimal, but not the best, path in the search space. This leads to propagation of the recognition errors. To the best of our knowledge a method that finds the best path is not known so far. Such a method could significantly improve not only the state of the art of speech recognition systems, but also any other system based on beam search like statistical machine translation systems, optical character recognition system etc. Despite of the fact that the open problem for the best path is considered very hard, we plan further research in this direction that could lead to better quality in comparison with beam search methods. In these ambitious tasks we rely on our previous experience and successes in efficient computations.
In Signal and Image Processing:
Initial plans concern theoretical research with the acoustic camera. This device is a high-quality piece of equipment but belongs to the entry-level due to comparatively low number of microphones and small aperture's size. The number of microphones could not be easily increased (the hardware price depends mainly of number of channels/microphones). The main ideas about acoustic camera improvement are directed to aperture modification and implementation of different algorithms for resolution enhancement, enlargement of applicable frequency band, smart feature extraction and so on. The already purchased microphone connectors and cables provide the necessary hardware for experimenting with new configurations of microphones on newly designed apertures.
We foresee to develop further the proposed smart approach for features extraction from the multi-dimensional data received by acoustic camera towards:
(i) Refinement of initial features extraction procedures aimed at:
- Investigation of proper frequency ranges for specific diagnostic purposes using focalized spectra;
- Increasing of the resolution of the produced by our algorithm “acoustic picture” aimed at fine detection of sound sources.
(ii) Investigation of different kind of clustering/classification procedures aimed at:
- Increased accuracy of diagnostic;
- Improved visualization of multi-dimensional data.
There are also considerations how to use the research results achieved in Biometric Authentication esp. Ear Biometrics. The plan is to disseminate on-line the 3D Ears Database created in IICT-BAS (for about 100 subjects). This would support the experiments carried out in the ear biometric research community which lacks high accurate 3D ear mesh data. The IICT-BAS team will continue developing and supporting this valuable resource, and will exploit it further in the development of methods and prototypes for Biometric Authentication.
In Optimisation and Intelligent Control:
In New Technologies, Mechatronic Systems and Robotics: in AComIn an advanced nanotechnology was developed for laying of a nickel coating with included nanoparticles on the working surfaces of extruding shafts used in industrial manufacturing of sheet materials. In our future work, the technology will be used for development of new types of innovative metal coatings with increased hardness and wear resistance that can be used for renovation of different mechanical devices in industry and production processes. A robotised system for inspections of coatings with high mechanical wear-resistance and surface-smoothness will be designed and put into exploitation. The research on mechatronic systems and robotics will continue by studying and design of autonomous mobile robots intended for multisensory environment perception and exploration, transportation of equipment and objects and working in harmful environment. New methods and algorithms for group control of mobile robots will also be developed and examined.
In Intelligent Methods and Systems: In our studies we have suggested original methods for diagnostics and predictive maintenance of industrial assets using intelligent techniques for analysis of the facilities condition. As a next step these methods will be used in the development of complex intelligent systems for advanced technological and operational management of various industrial and manufacturing processes.
In the area of radar data processing and target detection, an advantageous method was proposed using a Track-Before-Detect processing technique, which allows for the previously collected data to be used in the target detection process and parameter estimation. The method will be further developed in order to increase the accuracy of the estimated moving target parameters in a dynamic radio location environment. Examination and development of new robust and reliable algorithms for simultaneous trajectory and target detection is also a perspective field of research. The development of new algorithms that can be used to retrieve information about targets, applying a mathematical transformation on the received signals yielding estimates of the parameters of moving targets with extremely high precision in a dynamically changing radar environment is a new and very promising direction in modern information and communication technologies.
One of the most recent directions of the scientific research conducted by the intelligent control team within the project frame is the newly proposed Inter Criteria Decision Making aapproach. Further development and practical applications of this approach will support a decision maker in evaluating data or measurements of multiple objects against multiple criteria, to more profoundly understand the nature of the criteria involved and their correlation.
Information and Communication Systems. In the area of data transfer and optimisation in communication networks, we have proposed a family of models of the incoming traffic in a crossbar switch node together with algorithms for nonconflict scheduling and optimal data transfer as well as numerical procedures for estimating the maximal throughput of the communication node. Further exploitation of these results will include substantial improvements of the models in order to cope with complex traffic flows with different intensity and stochastic parameters, development of new scheduling algorithms, simulation studies including large scale computations in order to obtain precise performance evaluations of the communication network.
Important results obtained in course of the AComIn project also include a sensor system for meteorological data capturing using hierarchical cluster-tree topology and a software platform for smart sensor network data integration based on Service Oriented Architecture. The research work in this direction will continue by developing new software tools and mechanisms for cooperative data mining, self-organization, networking, and energy optimization in order to build higher level service structures.
Knowledge and experience acquired in the ICT sector within the project AComIn will be extensively used in the design of methods, algorithms and software products for local and web-based optimization and decision support systems, development of intelligent procedures and devices leading to innovative applications in industry, manufacturing, production planning and corporate business management.
In Advanced Control of Transport Systems. Our work in the development of multilevel optimisation for transportation system has the potential to improve the control policies and the exploitation of traffic networks in intensive urban areas. The control policies always target reducing the congestions, which deteriorate traffic conditions. Applying multilevel optimization the control space is increased and additional goals can be achieved as minimization of waiting time and maximization of outgoing flows from the urban network.
Potential improvement and future prospects from the multilevel optimization can be expected by development of fast algorithm for solving such class of control problems. Provisionally this could be the derivation of appropriate coordination strategies, which reduce the computational workload for real time purposes. Future developments will be focused on: (i) derivation of analytical approximation of inexplicit relations, concerning the solution of upper and lower level optimization problems; (ii) development of coordination strategies, which reduce the information exchange between the upper and lower level problems.
Having powerful methods to solve in real time multilevel problem, the control policies for transportation system can benefit from the increased control space, which can contain both the duration of the traffic lights, time cycle and the offset between the traffic nodes. Additionally variable speed limits and information signs can be incorporated in the traffic control. The multilevel approach can support additionally the integration of the on-line traffic estimations and optimization of the control states. Thus the multilevel approach can support both control policies for free way traffic and ramp metering.
A mathematical model for improvement of the rail-way passenger transportation in Bulgaria will be developed as an academic response of IICT-BAS towards the request of the Ministry of Transport of Bulgaria for the definition of mathematical model for the future development and exploitation of the Bulgarian state railway passenger transport. This type of transport has to become a backbone for the transportation structure of Bulgaria and has to provide rearrangements and modifications to the current bus transportation systems on national, regional and community scales.
2. Socio-economic impact
AComIn has indirect impact of employments and income. The project clearly helped several Bulgarian companies to improve their competitiveness by
- Increasing the quality of their production and services via new research methods, provided by IICT-driven research, integrated within their production cycles,
- Improving the energy efficiency of their production processes,
- Demonstrating solutions for material analysis and cheaper non-destructive testing that can be employed in their everyday work.
AComIn has direct impact on public services. It provided algorithms and software components for analysis of clinical narratives in Bulgarian language that were integrated in the environment for automatic generation of the (anonymous) national diabetic register (which was done by the University Specialised Hospital in Endocrinology, Medical University Sofia). These language technologies were applied over millions of outpatient records submitted to the Bulgarian Health Insurance Fund in 2012-2014. This allowed for automatic extraction of numeric values concerning diabetes compensation and enabled the statistical assessment of originally unstructured data. The experiment is unique and opens the door to new strategies for better control of chronic diseases like Diabetes Mellitus and Hypertonia.
AComIn also contributes to quality of life since the project elaborated several research proposals how to tackle:
- health problems by analysis of microstructures with various applications and
- ecological problems by simulation of BOD removal in horizontal subsurface flow constructed wetlands via high-performance computations.
3. Wider societal implications
AComIn is coordinated by a female researcher and was performed by many female researchers and experts (something relatively rare in ICT). Among the incoming researchers, who were employed in IICT with long term contracts, about 30% were female: 5 out of 16. As for the permanent staff, the team consisted of 27 women and 39 men.
The research results of the project were published in Open Access Journals whenever possible and uploaded in Open Access Repositories like the Anthology of the Association for Computational Linguistics. Two linguistic resources have been developed within the project, one of them is completely free (core Bulgarian WordNet) and the other one is free for academic research.
The project published a lot of promotional materials in Bulgarian language (and some project leaflets were only in Bulgarian as they were oriented to the visitors of national events).
Last but not least AComIn provided the results of a scientific study (analytics of learning data logs) to UCHA.SE the most popular Bulgarian on-line portal for K-12 users. Such a large-scale research study is made for the first time in the country. In this way the project supported the emerging eLearning activities in Bulgaria.
II. Main Dissemination Activities of AComIn Project
Seven high-quality scientific Conferences were organised within AComIn as dissemination forums for the project research results:
- The International Conference "Advanced Computing for Innovation" ACOMIN 2015, 10-11 November 2015, Sofia, Bulgaria, with Proceedings "Innovative Approaches and Solutions in Advanced Intelligent Systems" published as Volume 648 of the series Studies in Computational Intelligence, Springer 2015, ISSN:1860-949X
- The 10th International Conference "Recent Advances in Natural Language Processing" RANLP-2015, 5-11 September 2015, Hissar, Bulgaria. The Proceedings is uploaded in the SCOPUS indexed Anthology of the Association of Computational Linguistics (a Digital Archive of Research Papers in Computational Linguistics) at http://aclanthology.info/events/ranlp-2015
- The 10th International Conference "Large-Scale Scientific Computations" LSSC'15, 8-12 June 2015, Sozopol, Bulgaria, with Proceedings published as Volume 9374 in the series Lecture Notes in Computer Science, Springer 2015, ISSN: ISSN 0302-9743
- The 16th International Conference "Artificial Intelligence: Methodology, Systems, and Applications" AIMSA'2014, 10-13 September 2014, Varna, Bulgaria, with Proceedings published as Volume 8722 in the series Lecture Notes in Artificial Intelligence, Springer 2014, ISBN: 978-3-318-10553-6, 92-104
- The 8th International Conference "Numerical Methods and Applications" (NMA'14), 20-24 August 2014, Borovets, Bulgaria, with Proceedings published as Volume 8962 in the series Lecture Notes in Computer Science, Springer 2015, ISSN: ISSN 0302-9743
- The 9th International Conference "Recent Advances in Natural Language Processing" RANLP-2013, 7-13 September 2013, Hissar, Bulgaria. The Proceedings is uploaded in the SCOPUS indexed Anthology of the Association of Computational Linguistics (a Digital Archive of Research Papers in Computational Linguistics) at http://aclanthology.info/events/ranlp-2013
- The 9th International Conference "Large-Scale Scientific Computations" LSSC'13, 3-7 June 2013, Sozopol, Bulgaria, with Proceedings published as Volume 8353 in the series Lecture Notes in Computer Science, Springer 2014, ISSN 0302-9743
Nine Workshops with international participation were organised within AComIn as dissemination forums for the applied research results of the project:
- The International Workshop "Advanced Industrial Control Applications"AICA-15, 8 October 2015, Sofia, Bulgaria
- The International Workshop on Information Fusion, 25 September 2015, Sofia, Bulgaria, with Selected papers published in the Scopus indexed Journal "Cybernetics and Information Technologies", 15(7), 2015, ISSN 1314-4081
- The International Workshop on Big Data in Education and Digital Collections, 29 June 2015, Sofia, Bulgaria, with Selected papers published in the Scopus indexed Journal "Cybernetics and Information Technologies", 15(7), 2015, ISSN 1314-4081
- The International Workshop "Control in Transportation Systems", 10-11 September 2014, Sofia, Bulgaria, with Selected workshop papers published in a Special issue of the Journal "Cybernetics and Information Technologies" 15(5), 2015, ISSN 1314-4081
- The International Workshop on Biometrics (BIOMET'2014), 23-24 June 2014, Sofia, Bulgaria, with Proceedings published as Volume 8897 in the series Lecture Notes in Computer Science, Springer 2014, Print ISBN: 978-3-319-13385-0
- The International Workshop "Advanced Control and Optimisation: Step Ahead" ACOSA, 8-10 May 2014, Bankya, Bulgaria, with Proceedings published by Prof. Marin Drinov Academic Publishing House, 2014, ISSN 1314-4634
- The Workshop "ICT for New Materials and Nanotechnologies" (New Nano), 8-10 October 2013, a co-event of the International Conference "Robotics, Automation and Mechatronic" 2013, Bankya, Bulgaria, with Proceedings published by Prof. Marin Drinov Academic Publishing House, 2013, ISSN 1314-4634
- The International Workshop "Autonomic Computing and Automatic Control in Computer Systems" (ACACCS), 3-4 October 2013, a co-event of the International Conference "Аutomatics and Informatics 2013", 3-7 October 2013, Sofia, Bulgaria, with Selected papers published in the Scopus indexed Journal "Cybernetics and Information Technologies", 13(4), 2013, ISSN 1314-4081
- The Workshop "Information and Communication Technologies for Human Health and Quality of Life" (ICT-HuHeQuL), 15-17 May 2013, Stara Zagora - Mineral Baths, Bulgaria. The extended versions of 6 talks presented at the Workshop were published in two referenced Bulgarian scientific journals: "Bulgarian Journal of Agricultural Science" Vol. 20 (2014) ISSN 1310-0351 and "Cybernetics and Information Technology" 13(4), 2013, ISSN 1314-4081
Twenty Technology Transfer Events enabled know-how transfer to User Communities in AComIn:
- International Workshop on Industrial and Applied Mathematics, 21 December 2015
- International Workshop on Mathematics in Industry, 14 September 2015
- Know How Transfer on Speech Technologies, June-September 2015
- Know-How Transfer on Acoustic Imaging, April-September 2015
- International Workshop on Advanced Material Characterisation, Modelling, and Numerical Simulations, 28 June - 1 July 2015
- International Workshop on Advanced Techniques in NonDestructive Testing, 18-19 June 2015
- International Workshop on Advanced Computing for Innovation – Industrial Applications, 14-15 May 2015
- Workshop on 3D Digitisation and Virtual Reality, 18 April 2015
- Workshop on Microstructure Material Analysis, 17 April 2015
- International Workshop on Biomedical Simulations, 4 December 2014
- Workshop on Robotics and Innovations, 18-19 September 2014
- International Workshop on 3D Visualization of Cultural Heritage, 10 September 2014
- International Workshop "Walk on Equations and Sequential Monte Carlo to Solve Linear Systems", 21 August 2014
- International Seminar "Challenges connected with the use of resources of large digital libraries in education and civil science", 9 July 2014
- International Seminar "Introduction of game elements in training systems and education", 3 July 2014
- Workshop on 3D-Scanning and digitization, 17 March 2014
- Workshop on Thermography and its Applications, 11-13 February 2014
- International Seminar on Industrial Mathematics, 19 December 2013
- Technology Transfer Seminar "3D Technologies in the Textile Industry and Fashion", 2-5 September 2013
- Technology Transfer Seminar on Computational Vision Applied to Medical Diagnostics, 24-26 July 2013
Three Doors Open Days with demonstrations of AComIn applied results (using the SmartLab equipment) were organised:
- AComIn Doors Open Days 15-16 January 2016, Sofia, Bulgaria
- AComIn Doors Open Days 17-18 April 2015, Sofia, Bulgaria
- AComIn Doors Open Days 28-29 March 2014, Sofia, Bulgaria
Further presentations to leading figures of the State Authorities and the European Commission as well as the society at large are listed below:
- AComIn was presented to the President of the Republic of Bulgaria – Rosen Plevneliev who visited the Institute of Information and Communication Technologies on 10 February 2014
- AComIn was presented to Mr. Wolfgang Burtscher, Deputy Director General of DG Research and Innovation, European Commission, who visited the Institute of Information and Communication Technologies on 17 February 2014
- AComIn was presented at the Meeting of the Science, Technologies and Innovation Expert Council to the Mayor of Sofia Municipality held in the Institute of Information and Communication Technologies on 14 July 2015
- AComIn poster was shown in Sofia downtown as part of the exhibition "Bulgarian Academy of Sciences - science for the society and the country", dedicated to the 145th Anniversary of the Bulgarian Academy of Sciences, in June-July 2014
- AComIn results were demonstrated at the Promotional presentation of the Exhibition "the Battle of Pavia", Expo 2015 Milan, Italy, on 25 May 2015
- AComIn results were shown in the Exhibition "The Battle of Pavia 1525" in Visconti Castle, Pavia - an associated event of Expo 2015 in Milan, Italy. The Exhibition was opened in June-November 2015
- Public presentation of AComIn Evaluation Report was organised in Sofia, Bulgaria on 24 March 2016
III. Exploitation of Results
The exploitation of results already started in various applied activities, as listed in the Section "Description of main S & T results/foregrounds". The Table with "Exploitable Foreground" in this report lists six patent applications that have been submitted recently to the Bulgarian Patent Office (BPO), the World Intellectual Property Organisation (WIPO) and the European Patent Office (EPO). One research result is now integrated in a real-life application for commercial exploitation and will be soon in use.
To ensure further and more active exploitation of results the project team plans to keep the sustainability of liaisons with AComIn User Communities. We managed to build a Network of Users from specific areas who want to increase their knowledge about modern ICT applications and to take part in pilot developments. Such a network existed before AComIn too, related to e.g. the Bulgarian GRID community, but AComIn and the installation of SmartLab accelerated significantly the process of User Communities formation. The attracted participants in the Technology Transfer Seminars in AComIn year 3 (more than 400 people) will further disseminate the news about SmartLab potential in particular and IICT-BAS capacity in general. These figures show that the AComIn idea to focus on User Communities (among other objectives) was very successful too. It is interesting to note that recently IICT-BAS has established itself as a non-formal transfer center, making successful connections between business and academia. News about the Doors Open Days in IICT-BAS are broadcasted on the national TV channels and the national radio (thanks to the PR-Department of the Bulgarian Academy of Sciences). Many pupils from high-schools come to the Days, as well as students.
Following the recommendations of the AComIn Reviewers, the Network of Users will be extended to foreign countries. Social media will be used to disseminate information about the institute capacity and to keep contacts with potential clients.
List of Websites:
Project website:
http://www.iict.bas.bg/acomin
Project Newsletters in English:
http://iict.bas.bg/acomin/e-newsletters.html
Project Newsletters in Bulgarian:
http://iict.bas.bg/acomin/bg/e-newsletters.html
Scientific events organised within AComIn (list and short reports):
http://iict.bas.bg/acomin/events.html
Technology Transfer events organised within AComIn (list and short reports):
http://iict.bas.bg/acomin/user_communities.html
AComIn Movies, media reactions and other signs of AComIn recognition (list and short reports):
http://iict.bas.bg/acomin/appreciation.html
Project Deliverables with public access:
http://iict.bas.bg/acomin/deliverables.html
Lists of all project publications including ones without DOI:
http://iict.bas.bg/acomin/publications_talks.html
http://iict.bas.bg/acomin/monographs.html
Contact details:
Coordinator: Prof. Galia Angelova, Dr.Sc.
Coordinator's e-mail: galia@lml.bas.bg
Institute of Information and Communication Technologies - Bulgarian Academy of Sciences
Acad. G. Bonchev St., block 2
Sofia - 1113
Bulgaria
tel. +3592 979 6611
e-mail: iict@bas.bg
The general objective of AComIn is to strengthen the research and innovation capacity of the Institute of Information and Communication Technologies (IICT), Bulgarian Academy of Sciences (BAS) by increasing the knowledge and skills of IICT researchers in advanced areas, by purchasing modern research infrastructure as well as by development of innovation capacity. AComIn should help the institute to successfully accomplish its strategic mission: by 2016, i.e. 5 years after its creation, IICT has to become a leading RTD Centre in Eastern Europe, providing facilities and working conditions comparable to the average standards of the EU Centres of Excellence in ICT. The institute will support the sustainable regional and national growth and employment by providing RTD results to advanced industrial organisations and users from the public sector.
In terms of human capacity, IICT managed to improve its human resources. AComIn project attracted 16 incoming post-doctoral researchers, who worked in IICT with long term contracts, and eight visiting professors who were recruited in IICT with short term employments. Four of the post-doctoral researchers remained in the institute after AComIn end. As for the staff of the institute with permanent contracts, comparing the figures before and after AComIn (2011 vs. 2016), overall the number of permanent staff remains the same but the average age is now between 40 and 50 compared to 50-60 in 2011. Many female researchers obtained scientific degrees and habilitations; new young researchers and doctoral students were attracted. Moreover, IICT offers continued research education to obtain a PhD with independent preparation or paid PhD study to young experts coming from the industry. This illustrates the capacity building and multiplicator role of IICT.
In terms of publication activity, all research teams in AComIn produced internationally acknowledged results. Several research directions (not envisioned in the original proposal) emerged as result of new ideas brought by the new postdocs. In the category “papers with IF and SJR with DOI” AComIn scientists published 161 papers. The project actually implied for 3 years more than doubled indicators as follows: Total IF, Average IF, and the SJR SCImago Journal & Country Rank. Another effect is the strong reduction of the number of publications in domestic Volumes.
The equipment acquired for the SmartLab is the most visible and understandable asset of IICT. It clearly complements the existing high performance computing core and extends significantly the research tasks that can be tackled using the new devices. The activities triggered by the SmartLab acquisition highlight that this is a perfect mechanism to attract additional interest and to generate industrial projects and collaborations. SmartLab reinforces the existing research with complementary activities. This infrastructure opens a clear perspective for cooperation with the industry and makes IICT one of the key actors in advanced computing in South East Europe.
The innovation activities of IICT are considerably extended in 2012-2015. There are six projects in AComIn-related topics, completed within the Competitiveness Operational Programme, coordinated by Bulgarian SMEs which collaborate with IICT as a research partner providing innovation. Based on AComIn results, IICT performed eight projects for contracted research with seven companies and one client from the public sector, in the areas of material analysis, audio processing and 3D modelling and simulation. Fifteen collaboration project for joint research and development of pilot prototypes were accomplished with academic organisations, users from the public sector and companies. Eleven patent applications of six inventions were submitted to the Bulgarian Patent Office, the World Intellectual Property Organisation and the European Patent Office. Certificates for one Registered design and one Utility model were issued by the Bulgarian Patent Office.
The dissemination activities within AComIn (16 Scientific Forums in the project research areas, twenty Technology Transfer events, yearly Doors Open Days, numerous promotional materials including three movies and various media reactions) made the project visible not only in the research community but in the society at large.
In this way the AComIn project had a definite, direct and clearly visible positive impact through significant strengthening of human potential across all units involved in it. This impact concerns not only personnel directly involved in the project, but all other researchers of the IICT BAS. The indirect effect comes, among others, from strengthening the international reputation of the IICT BAS as a prime research institution, and making it a key player at the regional level.
Project Context and Objectives:
The Institute of Information and Communication Technologies (IICT), Bulgarian Academy of Sciences (BAS) (http://www.iict.bas.bg) is an independent legal entity established on 1 July 2010. It arose after the Academy has been reorganised, as recommended in its International Evaluation performed by the European Science Foundation/ALLEA (November 2009). As a part of the reorganisation, involving extensive personal attestation and structural optimisation, the IICT was founded with the aim to integrate the research teams from three former Institutes of the Academy: the Institute of Parallel Processing (IPP), Institute for Information Technologies (IIT) and Institute for Computer and Communication Systems (ICCS).
The IICT is a self-governing permanent research unit of BAS established to perform the following major tasks: (i) to carry out basic and applied scientific research; (ii) to train students, on a contractual basis with the higher schools, granting them a possibility to participate in the scientific research process; (iii) to provide postgraduate tutorship to the PhD students and professional researchers. The IICT governing bodies are: (i) the Assembly of Research Scientists, (ii) the Scientific Council and (iii) the Director. The legal basis of the IICT-BAS is legally grounded by the (i) Law of Bulgarian Academy of Sciences and (ii) Statutes of the Bulgarian Academy of Sciences.
IICT-BAS has permanent staff of about 100 scientists and carries out research in different areas. The Institute is leader of two national infrastructures: the supercomputing one and the CLARIN-BG infrastructure for development of language resources and language technologies for Bulgarian. The AComIn project is generally focused on Advanced Computing, Language and Semantic Technologies, Image and Signal Processing as well as Optimisation and Intelligent Control. Nine of the 13 IICT departments worked in the four AComIn topics as follows:
- In Advanced Computing: Department of Parallel Algorithms, Department of Scientific Computations and Department of Grid Technologies and Applications;
- In Language and Semantic Technologies: Department of Linguistic Modelling and Knowledge Processing;
- In Image and Signal Processing: Department of Mathematical Methods for Sensor Data Processing;
- In Optimization and Intelligent Control: Department of Embedded Intelligent Technologies, Department of Intelligent Systems, Department of Hierarchical Systems and Department of Modelling and Optimization.
The general objective of AComIn is to strengthen the IICT research and innovation capacity by increasing the knowledge and skills of IICT researchers in emerging areas as well as by purchasing modern research infrastructure. AComIn should help the institute to successfully accomplish its strategic mission: by 2016, i.e. 5 years after its creation, IICT has to become a leading RTD Centre in Eastern Europe, providing facilities and working conditions comparable to the average standards of the EU Centres of Excellence in ICT. The institute will support the sustainable regional and national growth and employment by providing RTD results to advanced industrial organisations; it will be a focal point of high-quality research training in South-East Europe.
More specifically, the project objectives are as follows:
- Strengthening the IICT Human Potential by recruitment of incoming experienced researchers, foreigners and Bulgarian nationals who have left the country, including long-term employments of 7 post-docs;
- Purchase of modern equipment for setting up a Smart Periphery Lab that enables to fully exploit the potential of the existing IICT core Grid infrastructure and the Bulgarian Supercomputer. Organisation and Training of User Communities with company representatives who need deeper expertise and innovative RTD approach in their areas of activities;
- Networking with the partnering organisations and knowledge exchange with established ICT experts as well as participation of IICT researchers in prestigious international scientific events and exhibitions;
- Building an IICT Innovation Strategy and development of innovation potential and IPR management skills. Patent applications to the European Patent Office and the Bulgarian Patent Office are foreseen as well;
- Organisation of various dissemination activities to spread out the project results and the IICT excellence: scientific Conferences and Workshops, Technology Transfer events oriented to User Communities and innovation-absorbing Bulgarian companies, Information Days and Doors Open Days, a non-scientific Stakeholder Meeting and a Round Table, as well as publishing books and monographs, leaflets and Newsletters, posters and other promotional materials, making three movies for dissemination via TV channels, YouTube, Facebook etc.
- Evaluation of IICT by international EC-selected reviewers, scheduled for months 35-42. Four independent reviewers assess the IICT-BAS overall research quality and capacity in the post-evaluation facility in order to assign a "quality label" to the institute.
AComIn is focused on research topics which are fundamental for ICT and enable radical progress and development of novel applications. The state-of-the-art frontiers of these technologies tackle computational tasks with high complexity which underpin the challenges of Horizon 2020. IICT has proven record of excellence in these synergistically connected areas and it is expected that the project will strengthen further the horizontal collaboration within the institute. Advanced computing becomes a broadly-used technology relatively recently because it requires substantial infrastructural developments; thus the available IICT capacity in advanced computing is a chance for the other IICT departments to quickly switch to most modern computational paradigms and to generate state-of-the-art results and corresponding innovation. More precisely, the scientific focus of the project is defined by the following research fields:
- Area 1: Advanced computing and Finite Elements applications including multiscale and multiphysics simulations of strongly heterogeneous media with strongly nonlinear and/or anisotropic behaviour as well as high-performance computing in engineering and environmental problems.
- Area 2: Monte Carlo methods, algorithms and distributed computing including sensitivity analysis of large mathematical models as well as advanced applications of Monte Carlo simulations in computational physics and environmental sciences.
- Area 3: Multimodal enrichment of voice communication by investigation a new generation of intelligent voice-activated technologies that focus not only on the precise recognition of spoken words but on interpretation of meaning and context to deliver more accurate speech recognition results.
- Area 4: Large-scale approach to multilingual terminology available in public sites by integration of separate resources, taking into account the ambiguity and the complex links between terms, and producing dynamic reference collections that will be publicly available at the AComIn site.
- Area 5: 3D modelling and recognition in biometrics applied to 3D face recognition based on large DB of face images, lips dynamics processing to help speech segmentation and recognition, iris dynamics and colors to enrich current eye tracking techniques and/or iris diagnostics, etc.
- Area 6: Digital preservation of cultural heritage for research and education by integration of facilities provided by Smart Lab devices and advanced computing solutions.
- Area 7: High spatial resolution based on near-field focalisation by using knowledge (delivered by the Smart Lab acoustic holography) about sources of noisy signals and any sound field descriptor such as sound pressure, sound intensity, or particle velocity as a function of position.
- Area 8: Hierarchical optimization in real time applications related to transportation systems, in particular for solving problems of optimal control, real time decision making, and on-line resource allocation in communication and transportation networks.
- Area 9: Energy efficient production technologies by using the Smart Lab devices and related IT methods for the precise study of dynamical behaviour of complex industrial systems and processes, further development of energy efficient technologies and improving the production quality.
- Area 10: Maintenance of industrial facilities operating in aggressive environment by creation of IT models for diagnostics and predictive maintenance of facilities operating under high risk of incidents. This includes modelling the dynamic behaviour of complex industrial systems and processes using the Smart Lab devices.
As for the selected equipment, which upgrades the computational facilities of the institute, the AComIn objective is to purchase devices that ensure the "data autonomy" of IICT, i.e. devices that enable flow of real-world data for the IICT research tasks. The core computational infrastructure of IICT integrates GRID clusters and the Bulgarian supercomputer purchased with considerable national funding within various programmes. However, insufficient advanced ′periphery′ (in a broad sense) to the HPC core was available. IICT needed modern devices to acquire data about objects and processes that might be of interest for the Bulgarian high-tech industry. Therefore AComIn plans to strengthen the IICT research infrastructure by purchasing SmartLab: a set of complementary high-tech devices that enable 3D input (by measuring shapes, microstructure, sound, temperature), 3D output (vizualisation, 3D printing), study of system dynamics, a speech processing studio, as well as an integrating server with software for traffic simulation and modelling. These devices enable highly innovative scientific activities related to material sciences, energy, health, industrial control and optimisation etc.
Project Results:
I. Scientific Foreground
AComIn foreground is defined in terms of new knowledge, generated in modern IT areas as result of the project research activities and technological achievements. The lists of most representative publications and exploitable knowledge items, presented in this Final report, support the claims for innovation and originality.
1. Wigner Monte Carlo Algorithms for Quantum Transport in Nanoelectronics
Efficient Monte Carlo Methods (MCM) are of great importance for the analysis of large-scale computer models. Because of their robustness, MCM is the only viable method for a wide range of high-dimensional problems ranging from Atomic Physics to Finance. MCM is a powerful tool for sensitivity analysis of large and very large mathematical models.
Our research in the frame of AComIn is focused on ground-breaking applications like studying of special nano-structures considered as potential candidates for future quantum computers, where we have considered technologically produced structures. The miniaturisation of devices did bring the developers in the realm of quantum transport. While technologically important quantum effects were only seen as perturbations in the past, today they are so relevant that specialists are exploiting them as principal effects to make new innovative devices work. The MCM has been extremely successful in the description of such devices and effects. The outcomes could be of great impact in the semiconductor community giving the creation of the first fully quantum simulator able to take into account also scattering effects. That could lead to the simulation of 3D devices like FinFETs, nanowires, multi-gate FETs, i.e. devices that are the most likely to be the candidates to substitute the MOSFET technology that is suffering from the miniaturisation effects. Those devices are already considered by industries such as Intel and AMD, as publicly advertised. The code that implements our MC algorithm is parallelised using the MPI library. This allows us to maintain the portability among different operating systems and/or architectures including available computational infrastructure at IICT-BAS.
The applications we are sealing produce advancements of three species: Development of physical insights in the field of Silicon based quantum computing devices; Development of a completely new numerical technique for the time dependent simulation of chemical systems; Development of mathematical advancements and insights in the theories of quantum transport. The following research activities were performed: (i) The research has been focused on the development of a time-dependent MC algorithm. Our MCM represents the first successful attempt in the world to simulate the Wigner equation time-dependently and multi-dimensionally. Groups at the University of Antwerp (Belgium) and in TU-Wien (Austria) are now using this new method on a daily basis. (ii) The two-dimensional Wigner MCM (ballistic regime) has been applied to the study of ordered and disordered arrays of dopants in order to explain the experimental results of various groups (in particular the Shinada’s group from ASMeW, Japan). The results are able to explain the experimentally observed enhanced transport characteristics obtained by ordered arrays. (iii) A sensitivity analysis of the coherence length in the Wigner MCM has been carried with success using the concepts of L1, L2 norms and the cosine similarity. The aim of this work is to show the efficiency of sensitivity studies of the design of semiconductor devices. (iv) A modification to Density Functional Theory to include the one-body Wigner MCM has been developed and validated in the cases of Hydrogen molecules in different geometrical configurations, Boron and Lithium. This is a very important achievement since the use of single body Wigner MCM allows for the first time the use of the Wigner formalism in the study of complex chemical systems. Let us mention that the many-body quantum problem was considered (before appearing the Wigner MCM) as unsolvable because of the non-polynomial growth of the computational complexity when other methods are used.
2. Robust Finite Element Methods and Algorithms for Advanced Computer Simulations
The modern Finite Element Methods (FEM) provide a computational technology for solving important classes of multiscale and multiphysics problems in science and engineering. The obtained results concern such challenging topics as novel FEM discretization techniques and robust scalability for problems with extremely heterogeneous and anisotropic coefficients.
The FEM discretization is based on conforming or nonconforming approximations. In both cases, the FEM basis functions should have a local support, ensuring sparsity of the matrices of FEM linear systems. The obtained new discretization/approximation results include: (i) interpolation, cubature rules, and least squares fitting of harmonic functions based on Radon projections; (ii) mixed finite element methods based on weighted Hdiv bilinear forms and equivalence between mixed and nonconforming FEM discretization. The robustness of mixed FEM discretization is studied in terms of uniform boundlessness of the constant in the inf-sub condition and the related LBB (Ladyzenskaja-Babuška-Brezzi) inequality.
Many of the nowadays real-life applications involve large-scale parallel processing of FEM linear systems with millions or even billions of degrees of freedom (unknowns). Such huge computational problems are beyond the scope of commercial software tools. The obtained new results are in the field of scalable preconditioned conjugate gradients (PCG) iterative solution methods. They include new or significantly improved multilevel, multigrid and domain decomposition methods, namely: robust algebraic multilevel iteration (AMLI) and semi-coarsening AMLI methods for strongly anisotropic problems; auxiliary space multigrid method based on additive Schur complement approximation; spectral analysis of geometric multigrid methods and full multigrid schemes for isogeometric analysis; finite element tearing and interconnect (FETI) solvers for non-standard finite element equations based on boundary integral operators.
The obtained results allow for advanced computer simulations and studies of complex processes and phenomena in science and high technologies. Among others, the robust solvers are tested on world-wide recognized SPE (society for petroleum engineering) benchmarks. The applications staying beyond the above theoretical achievements include industrial flows in porous media, as well as advanced civil engineering, biomedical engineering, environmental engineering, etc. simulations.
Some algorithms based on the numerical investigation for the seismic analysis of tall reinforced concrete (RC) Civil Engineering structures, which have been degradated due to extreme environmental actions and are strengthened by cable elements are developed. The effects of multiple earthquakes on such RC building frames are computed. Damage indices are estimated in order to compare the seismic response of the structures before and after the retrofit by cable element strengthening, and so to elect the optimum strengthening version.
A new computational model is developed for the mathematically rigorous analysis of Civil Engineering structures, which have been environmentally damaged and subsequently strengthened by cable-elements. The problem is treated as an inequality one, where the governing conditions are equalities as well as inequalities. The cable behavior is considered as nonconvex and non-monotone and is described by generalized subdifferential relations including loosening, elastoplastic-fracturing and other effects. Using piece-wise linearization for the cable behavior, a linear complementarity problem, with a reduced number of unknowns, is solved by optimization algorithms.
3. Advanced Computing in Dynamical Analysis of Elastic Structures
Understanding in details the dynamical behavior of elastic structures provides valuable information for the engineers, which can be used in the design, maintenance and health monitoring. The parametric study in frequency domain by the concepts of nonlinear normal modes (NNM) and nonlinear frequency-response function (NFRF) presents such knowledge that describes completely the dynamical characteristics of the structure. The necessity of obtaining reliable and accurate results requires consideration of nonlinear physical models and development of proper mesh discretization methods, e.g. finite element methods. As a consequence, the whole process of computing the NNM and NFRF becomes computationally expensive and cumbersome. Its effective parallel implementation on high performance computers is essential and unavoidable for the future development of new technologies for design, optimization and health monitoring of engineering structures.
New parallel algorithms for computing numerically the NFRF and NNM of dynamical systems are developed. The numerical methods are applied to large-scale dynamical systems that arise from space discretization of real-life elastic structures. New results, which show the optimal scalability of the proposed parallel implementation, are obtained.
The proposed numerical approach for dynamical analysis presents an iterative process for computing periodic responses of nonlinear systems by varying a parameter (such as frequency of vibration), determination of stability of the periodic solution, establishment of bifurcation points and following the secondary branches of solutions that arise from the bifurcation points. The computation of the NNM or NFRF is achieved by combination of finite element method, shooting and continuation methods. The application of the finite element method to elastic structures, considering geometrical type of nonlinearity, results into nonlinear system of second order ordinary differential equations. The shooting method computes the initial conditions that lead to steady-state periodic response of the dynamical system while the continuation method defines a prediction for the next point from the bifurcation diagram and enables to pass turning points (cyclic-fold bifurcation points). Additionally, the shooting method requires a time integration scheme, achieved by Newmark's method and solution of nonlinear algebraic system, achieved by Newton-Raphson's method.
New parallel realization of the shooting method is developed. The parallel process involves simultaneously efficient algorithms and basic matrix operations for sparse and dense matrices. Effective implementation on high performance computers considering equal distribution of computations and memory among available processors is obtained. New results with optimal parallel speedup and efficiency of the shooting method applied to nonlinear dynamical systems are achieved.
The proposed new technology gives the opportunity to academia and industry to compute efficiently the dynamical properties of real-life applications on high performance computers.
4. Advanced Methods, Algorithms and Innovations Based on 3D Digitisation and Prototyping
A key topic in this area is the voxel data processing based on computed tomography (CT) images of microstructures. The addressed advanced applications vary from fabless technologies for design of new composite and/or porous materials, trough quality control of technological processes and nondestructive defectoscopy, to high-tech innovations in biomedical engineering.
Accurate segmentation of 3D CT data of porous media is crucial for the numerical simulations and the computation of the material/object’s macro characteristics at the next stage. Due to the highly irregular structure of the segmentation phases and the presence of noise in the image, the classical methods are not reliable and the results of different standard algorithms may differ drastically (even on up to 50% of the data). A completely new direction of image segmentation is investigated, where some physical properties of the scanned specimen are incorporated in the segmentation process as constraints. In particular, the volume of the solid phase can be determined from the material’s density and weight measurements and can be a priori prescribed (mass conservation), while the solid phase itself, should be connected whenever the specimen is a single material piece.
Two different families of 2-phase image segmentation methods are proposed, analyzed and implemented. The conducted numerical experiments are promising in terms of both improved accuracy and decreased computational time. The computational efficiency is further optimized. MPI parallel implementations of the new algorithms on the high performance cluster at IICT-BAS are developed. It is important to notice, that experimental measurements of the macro characteristics of porous materials could be rather expensive and not always possible. In such cases, virtual material design is a modern research direction that significantly speeds up the analysis and the derivation of new porous materials with a priori given physical properties.
New methods, algorithms and software (including parallel) implementations for numerical upscaling are developed. The computed anisotropic tensors of material properties are beyond the nowadays abilities of the lab equipment for measurements and testing. The reported applications include: (i) numerical homogenization (upscaling) of heterogeneous anisotropic linear elastic materials; the studied specimens vary from innovative porous metals (lighted aluminums) to trabecular bone tissues, where the impact of osteoporosis is inspected; (ii) calibration of biomechanical characteristics of hepatic tissues and related parameters for radio-frequency tumor ablation simulation.
A completely new innovative methodology of 3D digitalization and prototyping is developed. Here, we are able to report the obtained extremely promising results concerning a lightweight 3D printed polymer prototype antenna, metallization and experimental test at 14-18GHz.
5. Language and Semantic Technologies
Automatic Processing of Image Annotations in Large-Scale Image Databases. The main challenge in the emerging area “Language and Vision” is how to produce (automatically) high quality annotations that describe the image semantics, with least effort and minimal costs. Automatic assignment of keywords to images is a rather difficult task and current solutions based on machine learning are unsatisfactory.
We proposed an approach for post-editing of keywords which are assigned to images. The suggested post-editing tackles "noisy" tags: mistakes, inflexions, doublets, normalization of keywords-names, abbreviations, and various linguistic and punctuation phenomena that occur at token level. Further we suggested employing linguistic resources that define English lexical semantics (WordNet) for consolidation of tags by removing “unnecessary” annotations.
The emotional classification of images depends on the individual opinion of each person but we proposed and investigated an idea how to compute image sentiment scores using external resources. We presented an approach for analysis of sentiments and emotions in image tagging using SentiWordNet as an external linguistic resource of emotional words. Calculating a "sentiment score" for each image, the system classifies images into three classes (positive, negative, and neutral). Our approach works with 63.53% precision, 58.7% recall and 61.02% F-measure which is coherent with the results reported for sentiment analysis in general.
Automatic tag sense disambiguation is another hot issue in the “Language and Vision” field. In general automatic tag disambiguation is a fundamental issue for modern management of digital resources because content objects with identical tags can be linked together allowing users to search for similar or related content in blogs, videos, image collections, learning objects or categories of web-pages. Tags ambiguity leads to inaccuracy and misunderstandings. We proposed an integrated method for tag disambiguation of images and showed its effectiveness for arbitrary tags in English with nominal and verbal senses using WordNet as an external resource defining tag senses. The main advantage of the method is the combination of known approaches which ensures some non-zero result for all annotation keywords belonging to WordNet. The evaluation is done on a corpus of about 5,600 tags with some 10,000 senses and showed that the correctly disambiguated tags are 95% for professional images and 87% for social images (that are often annotated manually).
Application of Educational Data Mining for Analysis of the eLearning Portal UCHA.SE. Educational Data Mining allows the discovery of new knowledge based on learners’ usage data in order to help validate and/or evaluate educational systems, to potentially improve some aspects of the quality of education and to lay the groundwork for a more effective learning process. Its potential explains the significant interest it raised in the educational community and the developers of online learning environments. A pilot research projects was carried out in the frame of AComIn. We applied educational data mining to the most popular Bulgarian educational site, UCHA.SE which offers instructional videos and practice exercises in Bulgarian language. The site can be used to support formal and informal education, that is, in schools and for self-learning. Currently it offers more than 4,300 video lessons in 17 subjects, including the basic K-12 subjects, as well as Introductory level English, German, French, and Spanish, and Introduction to Programming. As of September 2015, the site has more than 400,000 registered users - students (including college students), teachers, and parents. The lessons are seen more than 13 mln times.
The goal of the project was to improve the quality of educational services and the subscription rate for the site by leveraging educational data mining. In order to achieve these goals we extracted a “learner model” by applying BITool (business intelligence software supporting a multidimensional data model) to UCHA.SE relational database in which system logs and students’ performance data are stored. As a result the constructed “learner model” contained 30 attributes that reflect four type of information:
- general information about the user (e.g. age, gender, etc.),
- information related to user activities (e.g. “Percent of the total number of all taken exercises”, “Average length watched videos per day”, “Consecutive days the user has accessed the site”, etc),
- information related to user knowledge and skills (e.g. “Percentage of all attempts of these exercises, for which the total score exceeds the threshold of 75 points”, “Percent of the watched videos, which the user also commented”, etc, and
- information about the use of some gamification elements proposed by the site (e.g. “leadership level”, “charisma”, “influence”, etc.).
The goal for increasing the site subscription rate was reformulated as a classification task for predicting whether the user will renew his current subscription in the period of 3 month after its expiration. The classification task was solved by means of a machine learning approach. The training data reflects the user model and was constructed as of a concrete moment in time (end of February 2015). Since the learning data was highly unbalanced (only 18% of all data represented “positive” examples, i.e. the users who renewed their subscription in due time). The learning dataset was improved by combination of subsampling and undersampling methods. The predictive model was created as a combination of rules learned by two machine learning algorithms - JRIP and CN2, as JRIP rules generalized “positive” examples, and CN2 rules – the “negative” ones.
The predictive quality of this easily understandable to the end-users rule-based model, has classification accuracy 86.1% and F-measure 68%, respectively, which is close to the predictive quality of the model produced by one of best classification algorithms - SVM (87% and 68%). However, the SVM model is not understandable by the end-user.
News Media Analysis and Creation of Language Resources. The contemporary Natural Language Processing (NLP) applications, such as the analysis and extraction of relevant information from huge streams of data as well as context-aware Machine Translation, require high quality knowledge-rich processing. The Web provides an extensive volume of relevant knowledge, but most of it is in a textual or semi-structured format. Also, in the last decades the language and culturally specific information constantly increase. This calls for the development of language technology and resources (LTR) for national and local languages. These LTR need to reflect the language and culture specific features of the raw data. In order to support the interaction with rest of the world they are aligned by rule with LTRs for other languages.
Together with partners from Vrije Universiteit, Amsterdam (Computational Lexicology and Terminology Lab (CLTL)) we have been adapting the IICT-BAS in-house NLP processor for Bulgarian to analyse the events and stories in big news media data.
The NLP processor performs the following analyses: tokenization, POS and morphosyntactic tagging, lemmatisation, syntactic parsing, named entity identification and semantic analysis. Our efforts have been invested mainly in the last step as the most complex component of the processing chain. The semantic tagging includes the following submodules: (i) assigning the correct sense from the BTB WordNet on verbs, nouns, adjectives and adverbs; (ii) assigning the correct semantic role.
The first submodule uses the BTB WordNet, which was developed and is being extended within ACOMIN. It has been made freely available at the following link: http://compling.hss.ntu.edu.sg/omw/
The second submodule was developed through the WordNet verb semantic classes that were mapped onto the Bulgarian valency frames. The initial general semantic role labels have been distributed over BulTreeBank (BTB), and now the specialization process is in progress. Additionally, the first automatic transfer of Predicate Matrix (http://adimen.si.ehu.es/web/PredicateMatrix) was performed on Bulgarian data from the Setimes parallel corpora.
For the purposes of semantic annotation and disambiguation, the knowledge-graph tool UKB, developed by the IXA group, Basque country, was used. The experiments have been done with Bulgarian (BulTreeBank) and English data (SemCor). The results show improvement when syntactic relations are added for both languages.
Language Technologies Applied for Generation of a Diabetes Register. This pilot project for large-scale application of language technologies was performed jointly with the University Specialised Hospital for Active Treatment of Endocrinology ″Acad. Ivan Penchev″ (USHATE), Medical University – Sofia. USHATE is authorised by the Bulgarian Ministry of Health to host an anonymous Register of diabetic patients in Bulgaria. This Register contains 28 indicators of diabetic patients including age, sex, codes of diagnoses of diabetes and its complications, diabetes duration, risk factors, data about compensation, laboratory results, hospitalisations and prescribed medication. The AComIn experienced researchers helped for the automatic construction of the Register using language technologies and business analytics tools.
The register is automatically generated from a Repository of more than 112 million pseudonymised reimbursement requests (Outpatient records, ORs) submitted to the National Health Insurance Fund (NHIF) in 2012-2014 for more than 5 million citizens, including 436,000 diabetic patients. ORs are semi-structured in XML format; in each file some tags contain free-text fields with important explanations about the patient: “Anamnesis”, “Status”, “Clinical examinations” and “Therapy”. In AComIn, pilot prototypes for large scale processing of Bulgarian clinical texts were developed that extract automatically most essential data.
We developed a drug extractor using regular expressions to describe linguistic patterns. There are more than 80 different patterns for matching text units to ATC drug names/codes and NHIF drug codes. This extractor identifies the medication name, dosage, frequency and route of admission. Currently, it handles 2,239 drug names included in the NHIF nomenclatures. For treatment effect assessment we provide a comparative study of patients with insulin therapy, incretine-based therapy and oral medications therapy.
For extraction of clinical examination we designed a Numeric value extractor that processes lab and test results. Using collocation extractor we generate patterns for clinical exam data. After analysis of the positive and negative examples for such data we define the clinical exam data language L by a context-free grammar. For parsing clinical examination data we propose a hybrid method inspired by dependency grammars, constituents and Government and binding theory. The obtained result is a list of attribute-value pairs. For monitoring the compensation of Diabetes Type 2 and risk factors we extract values for body mass index (BMI), weight (W), blood pressure (BP), glycated haemoglobin (HbA1c), blood glucose (GLU), etc.
To identify complex relations between Diabetes Mellitus Type 2, its complications and other chronic diseases we defined a Data mining module for finding (i) frequent patterns and (ii) frequent temporal sequences of chronic diseases. For task (i) we apply modification of the classical frequent itemsets mining algorithms dEclat and association rules generation. We proposed a cascade method that reduces the search space for task (ii). In task (ii) several experiments were done – with no limitations for the distance between events (only the order matters), and with different window limitations between events – 1 to 3 months, etc.
Speech Processing. Modern state-of-the art speech recognition systems crucially use a beam search method in huge graph that combines acoustic model probabilities and language model probabilities. A disadvantage of all beam search methods is that they find a suboptimal, but not the best, path in the search space. This leads to propagation of the recognition errors. We have introduced a novel method for real-time lattice rescoring in speech recognition. It improves the performance of speech recognition systems by providing the option to partially compile the word lattice into a deterministic finite-state automaton, making it suitable for the rescoring step in the speech recognition process. In contrast to the widely used n-best approach our method permits the consideration of significantly larger number of alternatives within the same time-constraint thus providing better recognition results. A description of the new method is presented together with empirical evaluation of its performance in comparison with the n-best method. The achieved WER reduction is up to 3.77% at a p-value below 3%. An important advantage of our method is its applicability for real-time speech recognition. A pilot prototype of speech-to-text system for Bulgarian is demonstrated.
AComIn supported also the development of a novel effective method for searching similar audio segments in large audio collections. This method is based on original ideas for similarity search and ensures high reliability in noisy environments and exclusively high speed of searching for similar audio records. A pilot prototype for reliable automatic annotation of subtitled audio signals at phonemic level is demonstrated as well.
Annotated corpus of Bulgarian speech. Using the Speech Lab (purchased in AComIn) a multi-signal corpus of Bulgarian speech was developed. The corpus incorporates sounds, laringograph signals and physiological signals for some of the readers. The sentences in the corpus are segmented automatically by original software with manual correction of the accuracy. The resource is available at http://lml.bas.bg/BulPhonC.
6. Signal and Image Processing
Neuro-fuzzy Approach. In the considered period a theoretical investigation of effect of Intrinsic Plasticity (IP) improvement of Echo State Networks (ESN) reservoirs were summarised and commented. It was observed that IP training “captures” input data structure into the reservoir steady state in a way that could be useful for clustering purposes. In search of explanation of these results the Memory Capacity of Reservoir Equilibrium State (MCRES) was introduced. Achieved due to IP tuning MCRES of ESN reservoirs with different sizes was investigated using three benchmark artificial data sets. Two applications of neuro-fuzzy approach were realised. The first one visualises 3D acoustic waves propagation based on multi-dimensional data clustering. The second one implements adaptive critic designs for optimisation of fuzzy rule base parameters, designed to adapt the measurement noise covariance matrix. The proposed approach assures fast tuning of Kalman filter and improvement of the quality of its estimates in changing working conditions of the MEMS in real time application.
Enhancement of Acoustic Noise Source Localization and Identification. Results in several directions are achieved:
- Generation of the acoustic images in range - cross range - elevation coordinates. The approach enables to determine automatically the range to source (Acoustic camera software require measuring it in advance). The approach enables to localize the source with frequency bandwidth about 10 kHz, which is processed simultaneously;
- Enhancement of angular resolution of the acoustic images up to 2.72 (2.56) times in respect to "delay and sum" beamforming (built-in standard beamforming). The resolution enhancement was obtained for center frequencies from 1 kHz to 18 kHz, at bandwidth 10% and 23% of the center frequency. It was obtained based-on modified Capon`s approach, at range span comparable to dimensions of microphone array of the Acoustic Camera;
- A software tool for bearing noise estimation is prepared. It was applied for development of new tools for non-destructive testing of the bearings;
- Improvement of resolution of SONAH method in high frequency range, based on virtual microphones. The small number of microphones in Acoustic Camera limits SONAH method resolution. Virtual microphones are defined by using autoregressive relation for signal interpolation. In this way measurement surface density and/or area are numerically enlarged and resolution improvement of SONAH has been achieved;
- Fast direction of arrival estimation. A single frequency can be calculated on the basis of at least three data samples. Having a set of local frequencies one can calculate a set of direction angles and draw several rays from corresponding points. Estimation of the shift parameter of the distribution of these points gives the source coordinates;
- Multichannel frequency estimation. The task considers estimation of the power spectral density matrix, which consists of elements related to auto-spectrums and cross-spectrums for each pair of channels. By using the maximum likelihood approach the autoregressive parameter estimator has been synthesized. It has been shown that in the single channel case it is equivalent to the known modified covariance method obtained by a least squares approach. The simulation results demonstrate that the new estimator has similar precision in the middle frequency range to a simple one averaged by multiple channels, but the proposed method becomes much better at low frequencies. Additional attention has been paid to influence of the impulsive noise. The proposed solution assumes removing of noise impulses by the separation of the input sample and interpolation of the detected corrupted values. It has been shown that the proposed approach has advantage over the conventional filtering techniques. Statistical model of the signal sample, description of the detection criterion, flowchart of the filtering procedure and comparison results are published.
Research results achieved in Biometric Authentication. Ear Biometrics is a young topic that is intensively investigated recently. Even if ear has a uniform distribution of color, human external ear characteristics are considered enough unique to each individual and permanent during the lifetime of an adult, almost like fingerprint and/or iris data, but ear data can be gathered without any discomfort for people. A new method representing most essential data of human ear has been proposed, namely the matching potential for ear discrimination was tested via Extended Gaussian Image (EGI) representation and experimentally proved using 3D ear data gathered via the 3D Scanner of the AComIn SmartLab. Another approach to Biometrics Authentication by Ear has been proposed based on an original CBIR (Content Based Image Retrieval) method of IICT-BAS for rapid and reliable recognition of 3D objects using a database of precedents. Each object of interest for recognition is presented in the DB through a sufficient number of 2D projections (images), each from a different view point. Input images for recognition are obtained from conventional 2D cameras, and the most appropriate ear data are retrieved from the DB trough the CBIR method to access that is fast enough and noise resistant. We have already applied successfully this appearance-based approach two times: (i) for recognition of palm signs from a sign language alphabet and (ii) for human face recognition. The effective application for human ears became possible thanks to the 3D scanning technology available in the SmartLab of IICT-BAS (in the frames of AComIn project).
Speech Detection. A new robust feature for contour-based speech endpoint detection is proposed. It is called Group Delay Mean Delta (GDMD) feature and combines the properties of the known Modified Group Delay Spectrum (MGDS) and the original Mean Delta (MD) approach. The effectiveness of proposed feature is experimentally evaluated in the fixed-text Dynamic Time Warping (DTW) - based speaker verification task with short phrases of telephone speech. In comparison with three well know features - Modified Teager Energy (MTE), Energy-Entropy (EE) and MD feature - the experiments have shown that the GDMD feature demonstrates the best performance in endpoint detection tests in terms of verification rate.
Signature verification. A combined off-line signature verification approach is proposed using Neural networks, where global and grid signature features are combined to generate specific feature sets, one for each person represented in the verification system.
Video Stabilisation. A method and robust algorithm for fast 2D video stabilization for handheld devices in real-time is proposed. The BSC (Boundary Signal Computation) chip of TI (Texas Instruments) is essentially used (or emulated herein) for searching of correlations between the 1D integral projections, horizontal and vertical ones, by a SAD (Sums of Absolute Differences) approach. The proposed method is based on an accurate vector model allowing interpretations of increasing complexity for the transformations among frames. Experiments, conducted on testing video clips, are very promising for the future R&D of the method. This method has been extended for the problem of video stabilisation ‘in a point’, related to providing a static background for correct measurements with high-speed industrial cameras, like the one available in the SmartLab of AComIn. This new approach is called “3x3OF9x9” and combines (fuses) the basic method we use with the most lightweight version of the known Optical Flow approach, applying simple Otsu segmentation for eliminating the influence of moving objects in the video. The obtained results show better stabilisation (low that 1% instability for frame size of 640x480) in comparison with commercial software packets, like Warp Stabilizer of Adobe After Effects CS6.
7. Optimisation and Intelligent Control
A New Type of Chemical Nickel Coatings Including Nano Elements. The research on this topic includes a comprehensive study of abrasive wear and starting friction of nickel chemical coatings containing nanosized particles of silicon carbide (SiC) and aluminum oxide (Al2O3) of various sizes. It was carried out with the purpose of replacing old and non-ecological industrial technologies for laying a chrome based coating on different working surfaces. As a result, an advanced technology is developed and tested for renovation of extruding shafts used in industrial manufacturing of sheet materials, i.e. PVC, Plexiglas, other plastics, etc. The commonly used renovation technologies in this area are known to be exclusively harmful due to the waste cyanide products. Our innovative approach overcomes this drawback by applying advanced nanotechnology for nonelectric chemical laying of a nickel coating with included nanoparticles. Enhanced tribological characteristics are obtained due to the embedded micro- and nanosized particles. The technology is environmentally harmless and the new coating features an increased hardness and wear resistance. A special mechanical manipulator is developed as a part of a robotized systems for inspections of coatings with high mechanical wear-resistance and surface-smoothness.
Intelligent Methods for Technical Diagnostics. Technical diagnostics is a classical interdisciplinary area where results from narrow scientific sciences and specific trends are applied: probability and statistics, pattern recognition, decision making, mathematical logic. Modern industry requires efficient fault discovering and isolation solutions in process equipment service which is a real-world problem of typically ill-defined systems, hard to model, with large-scale solution spaces. Design of precise models is impractical, too expensive, or often non-existent. To cope with this problem, methods and tools are developed for intelligent diagnostics, monitoring and decision making for real detection of potential abnormalities in technological processes without applying expensive diagnostic apparatus. Especially the usage of AI-based methodologies enables us to deal with imprecise, uncertain data and incomplete domain knowledge typically encountered in practical applications. Several types of diagnostic approaches with different structures are applied to a mill fan device in the Bulgarian Maritsa East 2 Thermal Power Plant being the largest thermal power plant on the Balkan Peninsula. The possibility to predict eventual damage or wearing out without switching off the device turns out to be of great importance for providing faultless and reliable work of the plant.
Traffic Optimization in Communication Networks. In communication networks, a crossbar switch node routes traffic from the input to output where a message packet is transmitted from the source to the destination. The randomly incoming traffic must be controlled and scheduled to eliminate conflict at the crossbar switch. The goal of the traffic-scheduling for the crossbar switches is to maximize the throughput of packet through a switch and to minimize packet blocking probability and packet waiting time. In this area of research, new models of the incoming traffic in crossbar switch nodes are proposed including several families of input traffic patterns. As compared to the modeling methods and techniques available in the current literature, an important advantage of our models is that the generation of the traffic patterns does not depend on the type of hardware and software tools. Simulation studies and evaluations for the throughput of a switch node by the proposed family of patterns are performed using the grid-structure of the IICT-BAS. A numerical procedure for computation of the upper bound of the throughput is suggested which enables us to estimate the maximal throughput of the switch for different traffic scheduling algorithms.
Target Detection and Parameter Estimation. The research on this topic includes a thorough study of the latest trends in the design of highly efficient and fully automated systems for processing radar data in terms of a priori uncertainty about the targets and disturbances. Parameter estimation and high-speed small targets detection in randomly arriving impulse interference environment is a challenging task in radar systems design. Conventional radar detection architectures like spectral discrimination and non-coherent integration have been considered and used but with limited success. Within this framework, an improved target detection architecture is developed making use of highly efficient Hough velocity estimation technique. The proposed algorithm is based on a Track-Before-Detect processing method, which allows for the previously collected data to be used in the target detection process and parameter estimation. The presented technique has a lot of advantages compared to traditional ones. The obtained results can be successfully applied for radar target parameter estimation as well as in the existing communication network receivers making use of pulse signals.
Mechatronic Systems for Industrial Automation and Robotics. The investigation on this topic is aimed at developing new mechatronic systems incorporating mechanical constructions, electronic modules and intelligent software for automation of technological processes and operations. Main subjects of study are robots, dedicated for multisensory environment perception and exploration, measurements and samples taking, discovering and putting a mark on the objects as well as environment interactions like transportation, carrying in and out of equipment and objects. An important result obtained in this area is an advanced modular system developed for the mechanical construction of the mobile robots. It includes special programmable logical controllers for robot control and electronic modules for the wireless communication. New methods, means and algorithms for adaptive environment behaviour and group control of mobile robots are also examined. An innovative application of the international standard BDS ISO 10825 in mechanical transmissions is also proposed.
Intelligent Transportation Systems. The work on Autonomic transportation aims at meeting the challenge of engineering autonomic behavior in Intelligent Transportation Systems. The research integrates approaches from the disciplines of traffic engineering and autonomic computing. Autonomic computing is inspired by a biological example of body’s autonomic nervous system and it allows for a more efficient management of heterogeneous distributed computing systems. This type of systems are endowed with a number of properties that are generally referred as “self” properties, including self-configuration, self-healing, self-optimization, self-protection or in general term self-management. In this domain, an advanced approach is developed providing a new formal description of the traffic control policy. More complicated formalization is applied for the control of urban traffic systems using bi-level optimization. Such an approach gives potential for increase of the control space of the control problem. The new formalization of the control policy allows for the traffic flows to be controlled in urban networks not only by the green light duration of traffic lights as it is the common manner, but simultaneously by changing the duration of the traffic lights cycle obtained as solution of optimal control problem. This innovation benefits minimization of the waiting time of the vehicles in the urban network, maximization of the traffic outflows, decrease of traffic congestions, minimizing time for travel.
Numerically Effective Kalman Estimator Algorithm for Urban Transportation Network. The research on this topic addresses a problem of vehicles amount estimation in queues, which appear in front of controlled junctions in a congested Urban Transportation Network. The innovative result of the research is an approach, which leads to an effective numerical algorithm of building steady-state Kalman estimator for the store-and-forward urban transportation network model. The evaluation of the numerical algorithm shows its efficiency and applicability for on-line computations in the urban intelligent transportation system. The research presents an original approach, which uses a special structure of the optimization problem data and expands method of resolvent for constructing a stabilizing solution of discrete algebraic Riccati equation.
According to the AComIn Reviewers (their Review is publicly available at http://iict.bas.bg/acomin/docs/deliverables/D6_2.pdf) "all research teams produced internationally acknowledged results. Several research directions (not envisioned in the original proposal) emerged as result of new ideas brought by the new postdocs". It is also a positive effect that the new research topics are in line with merging activities at an international level, so they may grow and extend the institute research potential.
II. Technological Results
In terms of technological and innovation achievements, the results of AComIn show the high potential of the project team to transfer know-how to Users. Here we briefly summarize the figures illustrating the scale of applied research performed in AComIn.
1. Completed projects in the Competitiveness Operational Programme, coordinated by Bulgarian SMEs which collaborate with IICT as a research partner providing innovation. There are six projects tackling AComIn topics that were performed in 2013-2015:
- Project BG161PO003-1.1.06-0001-C0001 "Innovative technology for efficiency evaluation of ERP systems in small and medium enterprises", with beneficiary ERP Bulgaria Ltd. with R&D activities related to the development of software for efficient management of business processes;
- Project BG161PO003-1.1.06-0004-C0001 "Innovative technology solutions for radiofrequency thermoablation" with beneficiary AMET Ltd. Sofia, Bulgaria, with R&D activities related to the development, modern manufacturing and distribution of electronic medical equipment and modules, mechanical parts and units for incorporation;
- Project BG161PO003-1.1.06-0023-C0001 “Analysis and identification of inexplicit relations in large scale numerical data: applications in economics and technological analysis” with beneficiary ADISS Lab Ltd, with R&D activities related to the development of integrated information systems and business analytics tools;
- Project BG161PO003-1.1.06-0023-C0001 "Distributed Information System for Group Control, Distance Diagnostic and Service of Specialised Industrial Robots" with beneficiary SPESIMA GMBH, Sofia, Bulgaria - part of the FRECH Holding, Germany, with R&D activities related to the development of automation systems;
- Project BG161PO003-1.1.06-0023-C0001 "Experimental Development of Software Libraries for Improvement of Image Quality and Stabilisation using Inertial Sensors" with beneficiary MM Solutions AD, Sofia; with R&D activities in the area of complete imaging solutions for mobile camera devices;
- Project BG161PO003-1.1.06-0023-C0001 "Industrial Research for New Technologies in Image Stabilization and Image Quality Enhancement by Implementation of Inertial Sensors" with beneficiary MM Solutions AD, Sofia; with R&D activities in the area of complete imaging solutions for mobile camera devices.
2. Contracted Research
There are eight projects performed for external clients based on results of AComIn research activities, as follows:
- Research and development of models for the distribution of particles with a size of 0.01 - 2000 μm in samples of bulk materials - various types of metal powders and/or nanoparticles by using the laser nano particle sizer Analysette 22 Nano Tech + for measuring the size and distribution of particles. Performed for RETEL JSK, a company dealing with the delivery of complete equipment for the mining industry regarding processes of milling and enrichment of ore materials;
- Research for the development of 3D models with the mobile 3D scanner VIUScan, parameterization of 3D models, processing of the received data, followed byre search, analysis, design and visualization of 3D models of industrial, municipal and archaeological sites with the aim to find innovative solutions and testing new models in the field of 3D printing by using the professional 3D Printer Pro Jet 460+. Performed for ECO BIO ENGINEERING, a company engaged in large-scale modelling and engineering of industrial, archaeological and urban installations;
- Development of models for the distribution of particle with a size of 0.01 - 2000 μm in samples of solutions of bulk materials - various solutions, emulsions and mixtures and/or the contained nanoparticles after filtration and percolation using the laser nano particle sizer Analysette 22 Nano Tech + for measuring the size and distribution of particles. Performed for Diamond Properties, a company performing studies of particle distribution in powders as well as their mixtures and solutions for industrial users;
- Analysis and visualization of the size and distribution of particles in the concentrates of fruit juices and emulsions using the laser nano particle sizer Analysette 22 Nano Tech +, performed for Interaroma - a producer of different clear fruit juices and nectar;
- Construction of 3D models of different types of badges and medals and 3D printing in order to find the proper size. This application is done for B&W Consultant - a company engaged in design and production of promotional materials;
- Research and development of models concerning the distribution of particles with a size of 0.01 - 2000 μm in soil samples - various types of clays using the Laser Nano particle sizer Analysette 22 Nano Tech + for measuring the size and distribution of particles. Performed for the National Institute of Geophysics, Geodesy and Geography – Bulgarian Academy of Sciences;
- CT scanning and microstructure analysis of large series of bone tissue samples, done for an International Research Team;
- Development of a real-life application of an effective method for searching similar audio segments in large audio collections, performed for the Bulgarian TV-related company H-Tech.
3. Applied Research in Joint Research Projects
Collaboration activities with Universities were performed as joint projects dedicated to specific topics, as follows:
- Analyses of thermal images of thermal processes in large electrical circuits and electrical appliances, jointly with the Electrical Faculty of Technical University - Sofia;
- Joint study of high-speed processes concerning the briquetting of metal shavings and powders necessary for the optimization of briquetting of waste products with the purpose of secondary melting. Performed with the Faculty of Mechanical Engineering of Technical University - Sofia;
- Examination of temperature regimes in large buildings to deploy powerful breakers, fuses and distributors, done jointly with the Plovdiv Branch of Technical University - Sofia;
- Study of the high-speed processes of interaction and deformation of different physical bodies by shocks in hit railings and guardrails, performed for the Institute of Mechanics – Bulgarian Academy of Sciences to support the standardisation of protective fences, guardrails, handrails etc.;
- Examination of the temperature image of the cooling process of welding metals for different kinds of welding for medium nano particles in the range of 2000°C– 500°C, performed jointly with the Institute of Metal Sciences – Bulgarian Academy of Sciences;
- Development of models for a distribution of particles with a size of 0.01 - 2000 μm in samples of bulk materials - various types of powders of dried and ground food, performed together with experts from the Bulgarian Agricultural Academy which performs a research project related to dried and freeze-dried foods that are used in extreme sports (mountaineering) and when traveling in aerospace;
- Educational analytics of learning data performed together with the team of the most popular Bulgarian K-12 platform with educational videos and exercises. The results helped them to identify users who are likely to renew their subscription and attract other subscribers;
- Applying language technologies for automatic processing of Bulgarian clinical narratives for automatic construction of the National Diabetic Register, together with the University Specialized Hospital for Active Treatment of Endocrinology "Acad. Ivan Penchev", Medical University - Sofia;
- Non-destructive testing of rolling bearings using the AComIn acoustic camera, performed together with the Faculty of Mechanical Engineering of Technical University - Sofia;
- Microstructure analysis and numerical homogenization of epoxybased nanocomposites based on industrial CT scanning data. Done jointly with the Institute of Mechanics – Bulgarian Academy of Sciences;
- Participation in the project "Study of composition, structure, properties and production of yellow pavement in Sofia" together with the "Bulgarian crystallographic society" and experts from the Institute of Mineralogy and Crystallography - BAS, Institute of Physical Chemistry - BAS, Institute of Catalysis - BAS, Institute of Information and Communication Technologies - BAS, Mining and Geology University "St. Ivan Rilski "and the University of Architecture, Civil Engineering and Geodesy – Sofia. The project was funded by the Sofia Municipality (the yellow pavement is a prominent symbol of Sofia, which was declared national heritage in the beginning of 2014 by the Sofia municipal council);
- Applications of CT Microstructure analysis in Paleoanthropology developed jointly with the Institute of Experimental Morphology, Pathology and Anthropology with Museum – Bulgarian Academy of Sciences;
- Advanced CT study of structural and material properties of fiber reinforced concrete, performed with the Institute of Mechanics – Bulgarian Academy of Sciences;
- Investigations and microstructural analysis of geological materials by scanning with AComIn 3D industrial tomography, performed with the Institute of Mechanics – Bulgarian Academy of Sciences for Ruhr-University of Bochum (Germany);
- Preparation and completing of 3D models of historical figures related to the Battle of Pavia (1525), 3D printing of the figures and printing of 3D tactile matrices that display medieval tapestries to visually impaired people. Performed jointly with the University of Pavia for the Exhibition "The Battle of Pavia" held in Visconti Castle, Pavia, as an accompanying event of EXPO 2015, Milan, Italy.
These applied activities show that the practical results achieved in AComIn are based on the high-tech equipment purchased within the project.
Potential Impact:
I. Potential Impact
Here we present the potential project impact in three dimentions: (i) plans for exploitation of the AComIn research foreground, (ii) socio-economic impact and (iii) wider societal exploitations.
1. Plans for using the AComIn foreground
In Wigner Monte Carlo Algorithms for Quantum Transport in Nanoelectronics: The research carried out in AComIn was focussed on ground-breaking applications that support building up and structuring of a new interdisciplinary scientific community in emerging themes dealing with radically new future technologies. For instance our results involve studying of special nano-structures viewed as potential candidates for future quantum computers since we have considered technologically produced structures. In this way we have studied experimentally nano-structures which are considered as a candidate for a quantum computer. Our research in this fiels directly supports innovations in semiconductor devices. That could lead to the simulation of 3D devices like FinFETs, nanowires, multi-gate FETs, i.e. devices that are the most likely to be the candidates to substitute the MOSFET technology that is suffering from the miniaturization effects. Those devices are already taken into account by industries such as Intel and AMD, as publicly advertised. IICT-BAS plans joint research with Technical University Wien which is particularly focused on magnetic field, entangled electronics, and parallelisation algorithms. A systematic derivation of simulation models accounting for the magnetic field in the Wigner picture is still missing. A major challenge is the incorporation of the electromagnetic potentials and the choice of the gauge. A computationally feasible theoretical approach is pursued, which allows an efficient inclusion of the vector potential in the signed particle method. Entanglement characterises the state evolution in electron lenses, mesoscopic (Aharonov-Bohm) rings, and Quantum Point Contacts (QPC's). Electron transport in such structures comprises phenomena with yet not explored physical and application aspects. For instance, there is no relevant theory which can explain the experimentally observed behaviour in QPC's. All these issues are included in our research agenda for the near future.
In Robust Finite Element Methods and Algorithms for Advanced Computer Simulations: The future plans for research will target top achievements in robust FEM discretization and solution methods. The obtained theoretical results for strongly heterogeneous elliptic problems of high contrast and high frequency will be further developed to the case of coupled problems including models of fluid-structure interactions, flows in deformable porous media, thin plates and shells, etc. The expected new results will play a key role in such high tech applications as new constructive materials, clean technologies, nano membranes, energy efficient combustion in inert porous media. The related problems are inherently nonlinear, three dimensional in space and time dependent. This means that HPC will be without alternative in many of the real life applications. The big challenge here concerns the development of scalable parallel algorithms for the new generation heterogeneous systems including GPU and MIC accelerators.
In Advanced Computing in Dynamical Analysis of Elastic Structures: The research plans for next period will be focused on development of efficient methods, algorithms and codes for dynamical analysis of more complicated structural elements. The first step will address the related problems in the theory of plates. The standard model is described by PDE of fourth order which beans that the FEM approximations should provide higher smoothness of the test functions. Even for linear problems this leads to serious increase of the size of discrete problem from one side and robustness problems from another. One alternative approach (which will be preferred) is based on development of efficient methods and algorithms in terms of Mindlin–Reissner plate theory. Mixed FEM, nonconforming and reduced integration FEM will be studied. The obtained results will be generalized to the case of shells. In both cases, plates and shells the methods should be robust with respect of the thickness which is assumed by default to be small. The parallel implementation of the algorithms (e.g. the shooting algorithm) will be required in many real life applications in machine and civil engineering.
In Advanced Methods, Algorithms and Innovations Based on 3D Digitisation and Prototyping: This is a completely advanced topic. Some of the reported results are really pioneering. There are (at least) two kinds of challenges in the future research and development activities. The first one is related to the extremely fast development of the equipment for 3D digitalization and prototyping. The sub-micro computed tomography is already available. This means in particular new qualitative and quantitative requirements to the algorithms for segmentation of CT images. Even the formulation of the problem of mass conservative segmentation for multicomponent porous media is a separate challenge. The next plans in the related field of numerical upscaling will be targeted to homogenization of nonlinear properties. The results in the field of numerical upscaling of anisotropic materials have a strong potential for generalizations in the general case of unstructured Big Data analytics. Considering the future plans in prototyping we should notice the new opportunities provided (or expected to appear) in the on-linecontrol of 3D printing with respect to, e.g. changing the material or/and controlling the density etc.
In Language and Semantic Technologies:
Integrating Language and Vision in general requires better understanding of the annotation processes and the psychology of personal behavior in social media where users often assign emotional tags to their photos. Therefore we believe that considerable success in automatic tag disambiguation can be achieved in specific niches e.g. collections of professional images that document mostly material objects denoted by nouns. We plan to extend the present results in the area of disambiguation by integration of further external resources that provide reference sets of senses: like Wikipedia and DBPedia. Then there will be a voting scenario where different disambiguation platforms propose the correct tag sense. Further we plan to make experiments with embedding fine-tuned tags in the image recognition phase when the decision of automatic annotation is made.
Regarding the News Media Analysis and Creation of Language Resources, the results achieved in AComIn will be integrated in the development of a Language Technology and Resources Centre for Linked Open Data. This includes: (i) Further extension of the BTB WordNet through reliable (semi)automatic processes; (ii) Alignment of BTB WordNet with English and other languages wordnets as well as world knowledge databases such as DBpedia, Freebase and other datasets from Linked Open Data cloud; (iii) Further improvement of the knowledge graph performance through the addition of new relations; (iv) Developing of ensemble and supervised methods for solving several language technology tasks; (v) Crosslingual event analysis and information extraction for big news data in Bulgarian and English (also possibly Dutch); as well as (vi) Deployment of the semantic tagging into an MT system between Bulgarian and English.
As for the Language Technologies Applied for Generation of a Diabetes Register, the technologies used at present will be at first extended for automatic extraction of values in the section “Patient Status” in the Outpatient records submitted to the National Health Insurance Fund. Then the plan is to extend the extractors to further chronic diseases in order to generate other registers: like Diseases of the circulatory system, Malignant neoplasms, Mental and behavioural disorders, etc. In addition to the Outpatient records, the extractors should process selected zones in other types of clinical narratives: e.g. Discharge letters and Reports about Clinical examinations issued by Independent Medical Diagnostic Laboratories. A major challenge is the automatic processing of temporal information in clinical narratives: recognition of absolute and relative temporal markers, and ordering them in timelines.
In Speech Processing: A disadvantage of all beam search methods is that they find a suboptimal, but not the best, path in the search space. This leads to propagation of the recognition errors. To the best of our knowledge a method that finds the best path is not known so far. Such a method could significantly improve not only the state of the art of speech recognition systems, but also any other system based on beam search like statistical machine translation systems, optical character recognition system etc. Despite of the fact that the open problem for the best path is considered very hard, we plan further research in this direction that could lead to better quality in comparison with beam search methods. In these ambitious tasks we rely on our previous experience and successes in efficient computations.
In Signal and Image Processing:
Initial plans concern theoretical research with the acoustic camera. This device is a high-quality piece of equipment but belongs to the entry-level due to comparatively low number of microphones and small aperture's size. The number of microphones could not be easily increased (the hardware price depends mainly of number of channels/microphones). The main ideas about acoustic camera improvement are directed to aperture modification and implementation of different algorithms for resolution enhancement, enlargement of applicable frequency band, smart feature extraction and so on. The already purchased microphone connectors and cables provide the necessary hardware for experimenting with new configurations of microphones on newly designed apertures.
We foresee to develop further the proposed smart approach for features extraction from the multi-dimensional data received by acoustic camera towards:
(i) Refinement of initial features extraction procedures aimed at:
- Investigation of proper frequency ranges for specific diagnostic purposes using focalized spectra;
- Increasing of the resolution of the produced by our algorithm “acoustic picture” aimed at fine detection of sound sources.
(ii) Investigation of different kind of clustering/classification procedures aimed at:
- Increased accuracy of diagnostic;
- Improved visualization of multi-dimensional data.
There are also considerations how to use the research results achieved in Biometric Authentication esp. Ear Biometrics. The plan is to disseminate on-line the 3D Ears Database created in IICT-BAS (for about 100 subjects). This would support the experiments carried out in the ear biometric research community which lacks high accurate 3D ear mesh data. The IICT-BAS team will continue developing and supporting this valuable resource, and will exploit it further in the development of methods and prototypes for Biometric Authentication.
In Optimisation and Intelligent Control:
In New Technologies, Mechatronic Systems and Robotics: in AComIn an advanced nanotechnology was developed for laying of a nickel coating with included nanoparticles on the working surfaces of extruding shafts used in industrial manufacturing of sheet materials. In our future work, the technology will be used for development of new types of innovative metal coatings with increased hardness and wear resistance that can be used for renovation of different mechanical devices in industry and production processes. A robotised system for inspections of coatings with high mechanical wear-resistance and surface-smoothness will be designed and put into exploitation. The research on mechatronic systems and robotics will continue by studying and design of autonomous mobile robots intended for multisensory environment perception and exploration, transportation of equipment and objects and working in harmful environment. New methods and algorithms for group control of mobile robots will also be developed and examined.
In Intelligent Methods and Systems: In our studies we have suggested original methods for diagnostics and predictive maintenance of industrial assets using intelligent techniques for analysis of the facilities condition. As a next step these methods will be used in the development of complex intelligent systems for advanced technological and operational management of various industrial and manufacturing processes.
In the area of radar data processing and target detection, an advantageous method was proposed using a Track-Before-Detect processing technique, which allows for the previously collected data to be used in the target detection process and parameter estimation. The method will be further developed in order to increase the accuracy of the estimated moving target parameters in a dynamic radio location environment. Examination and development of new robust and reliable algorithms for simultaneous trajectory and target detection is also a perspective field of research. The development of new algorithms that can be used to retrieve information about targets, applying a mathematical transformation on the received signals yielding estimates of the parameters of moving targets with extremely high precision in a dynamically changing radar environment is a new and very promising direction in modern information and communication technologies.
One of the most recent directions of the scientific research conducted by the intelligent control team within the project frame is the newly proposed Inter Criteria Decision Making aapproach. Further development and practical applications of this approach will support a decision maker in evaluating data or measurements of multiple objects against multiple criteria, to more profoundly understand the nature of the criteria involved and their correlation.
Information and Communication Systems. In the area of data transfer and optimisation in communication networks, we have proposed a family of models of the incoming traffic in a crossbar switch node together with algorithms for nonconflict scheduling and optimal data transfer as well as numerical procedures for estimating the maximal throughput of the communication node. Further exploitation of these results will include substantial improvements of the models in order to cope with complex traffic flows with different intensity and stochastic parameters, development of new scheduling algorithms, simulation studies including large scale computations in order to obtain precise performance evaluations of the communication network.
Important results obtained in course of the AComIn project also include a sensor system for meteorological data capturing using hierarchical cluster-tree topology and a software platform for smart sensor network data integration based on Service Oriented Architecture. The research work in this direction will continue by developing new software tools and mechanisms for cooperative data mining, self-organization, networking, and energy optimization in order to build higher level service structures.
Knowledge and experience acquired in the ICT sector within the project AComIn will be extensively used in the design of methods, algorithms and software products for local and web-based optimization and decision support systems, development of intelligent procedures and devices leading to innovative applications in industry, manufacturing, production planning and corporate business management.
In Advanced Control of Transport Systems. Our work in the development of multilevel optimisation for transportation system has the potential to improve the control policies and the exploitation of traffic networks in intensive urban areas. The control policies always target reducing the congestions, which deteriorate traffic conditions. Applying multilevel optimization the control space is increased and additional goals can be achieved as minimization of waiting time and maximization of outgoing flows from the urban network.
Potential improvement and future prospects from the multilevel optimization can be expected by development of fast algorithm for solving such class of control problems. Provisionally this could be the derivation of appropriate coordination strategies, which reduce the computational workload for real time purposes. Future developments will be focused on: (i) derivation of analytical approximation of inexplicit relations, concerning the solution of upper and lower level optimization problems; (ii) development of coordination strategies, which reduce the information exchange between the upper and lower level problems.
Having powerful methods to solve in real time multilevel problem, the control policies for transportation system can benefit from the increased control space, which can contain both the duration of the traffic lights, time cycle and the offset between the traffic nodes. Additionally variable speed limits and information signs can be incorporated in the traffic control. The multilevel approach can support additionally the integration of the on-line traffic estimations and optimization of the control states. Thus the multilevel approach can support both control policies for free way traffic and ramp metering.
A mathematical model for improvement of the rail-way passenger transportation in Bulgaria will be developed as an academic response of IICT-BAS towards the request of the Ministry of Transport of Bulgaria for the definition of mathematical model for the future development and exploitation of the Bulgarian state railway passenger transport. This type of transport has to become a backbone for the transportation structure of Bulgaria and has to provide rearrangements and modifications to the current bus transportation systems on national, regional and community scales.
2. Socio-economic impact
AComIn has indirect impact of employments and income. The project clearly helped several Bulgarian companies to improve their competitiveness by
- Increasing the quality of their production and services via new research methods, provided by IICT-driven research, integrated within their production cycles,
- Improving the energy efficiency of their production processes,
- Demonstrating solutions for material analysis and cheaper non-destructive testing that can be employed in their everyday work.
AComIn has direct impact on public services. It provided algorithms and software components for analysis of clinical narratives in Bulgarian language that were integrated in the environment for automatic generation of the (anonymous) national diabetic register (which was done by the University Specialised Hospital in Endocrinology, Medical University Sofia). These language technologies were applied over millions of outpatient records submitted to the Bulgarian Health Insurance Fund in 2012-2014. This allowed for automatic extraction of numeric values concerning diabetes compensation and enabled the statistical assessment of originally unstructured data. The experiment is unique and opens the door to new strategies for better control of chronic diseases like Diabetes Mellitus and Hypertonia.
AComIn also contributes to quality of life since the project elaborated several research proposals how to tackle:
- health problems by analysis of microstructures with various applications and
- ecological problems by simulation of BOD removal in horizontal subsurface flow constructed wetlands via high-performance computations.
3. Wider societal implications
AComIn is coordinated by a female researcher and was performed by many female researchers and experts (something relatively rare in ICT). Among the incoming researchers, who were employed in IICT with long term contracts, about 30% were female: 5 out of 16. As for the permanent staff, the team consisted of 27 women and 39 men.
The research results of the project were published in Open Access Journals whenever possible and uploaded in Open Access Repositories like the Anthology of the Association for Computational Linguistics. Two linguistic resources have been developed within the project, one of them is completely free (core Bulgarian WordNet) and the other one is free for academic research.
The project published a lot of promotional materials in Bulgarian language (and some project leaflets were only in Bulgarian as they were oriented to the visitors of national events).
Last but not least AComIn provided the results of a scientific study (analytics of learning data logs) to UCHA.SE the most popular Bulgarian on-line portal for K-12 users. Such a large-scale research study is made for the first time in the country. In this way the project supported the emerging eLearning activities in Bulgaria.
II. Main Dissemination Activities of AComIn Project
Seven high-quality scientific Conferences were organised within AComIn as dissemination forums for the project research results:
- The International Conference "Advanced Computing for Innovation" ACOMIN 2015, 10-11 November 2015, Sofia, Bulgaria, with Proceedings "Innovative Approaches and Solutions in Advanced Intelligent Systems" published as Volume 648 of the series Studies in Computational Intelligence, Springer 2015, ISSN:1860-949X
- The 10th International Conference "Recent Advances in Natural Language Processing" RANLP-2015, 5-11 September 2015, Hissar, Bulgaria. The Proceedings is uploaded in the SCOPUS indexed Anthology of the Association of Computational Linguistics (a Digital Archive of Research Papers in Computational Linguistics) at http://aclanthology.info/events/ranlp-2015
- The 10th International Conference "Large-Scale Scientific Computations" LSSC'15, 8-12 June 2015, Sozopol, Bulgaria, with Proceedings published as Volume 9374 in the series Lecture Notes in Computer Science, Springer 2015, ISSN: ISSN 0302-9743
- The 16th International Conference "Artificial Intelligence: Methodology, Systems, and Applications" AIMSA'2014, 10-13 September 2014, Varna, Bulgaria, with Proceedings published as Volume 8722 in the series Lecture Notes in Artificial Intelligence, Springer 2014, ISBN: 978-3-318-10553-6, 92-104
- The 8th International Conference "Numerical Methods and Applications" (NMA'14), 20-24 August 2014, Borovets, Bulgaria, with Proceedings published as Volume 8962 in the series Lecture Notes in Computer Science, Springer 2015, ISSN: ISSN 0302-9743
- The 9th International Conference "Recent Advances in Natural Language Processing" RANLP-2013, 7-13 September 2013, Hissar, Bulgaria. The Proceedings is uploaded in the SCOPUS indexed Anthology of the Association of Computational Linguistics (a Digital Archive of Research Papers in Computational Linguistics) at http://aclanthology.info/events/ranlp-2013
- The 9th International Conference "Large-Scale Scientific Computations" LSSC'13, 3-7 June 2013, Sozopol, Bulgaria, with Proceedings published as Volume 8353 in the series Lecture Notes in Computer Science, Springer 2014, ISSN 0302-9743
Nine Workshops with international participation were organised within AComIn as dissemination forums for the applied research results of the project:
- The International Workshop "Advanced Industrial Control Applications"AICA-15, 8 October 2015, Sofia, Bulgaria
- The International Workshop on Information Fusion, 25 September 2015, Sofia, Bulgaria, with Selected papers published in the Scopus indexed Journal "Cybernetics and Information Technologies", 15(7), 2015, ISSN 1314-4081
- The International Workshop on Big Data in Education and Digital Collections, 29 June 2015, Sofia, Bulgaria, with Selected papers published in the Scopus indexed Journal "Cybernetics and Information Technologies", 15(7), 2015, ISSN 1314-4081
- The International Workshop "Control in Transportation Systems", 10-11 September 2014, Sofia, Bulgaria, with Selected workshop papers published in a Special issue of the Journal "Cybernetics and Information Technologies" 15(5), 2015, ISSN 1314-4081
- The International Workshop on Biometrics (BIOMET'2014), 23-24 June 2014, Sofia, Bulgaria, with Proceedings published as Volume 8897 in the series Lecture Notes in Computer Science, Springer 2014, Print ISBN: 978-3-319-13385-0
- The International Workshop "Advanced Control and Optimisation: Step Ahead" ACOSA, 8-10 May 2014, Bankya, Bulgaria, with Proceedings published by Prof. Marin Drinov Academic Publishing House, 2014, ISSN 1314-4634
- The Workshop "ICT for New Materials and Nanotechnologies" (New Nano), 8-10 October 2013, a co-event of the International Conference "Robotics, Automation and Mechatronic" 2013, Bankya, Bulgaria, with Proceedings published by Prof. Marin Drinov Academic Publishing House, 2013, ISSN 1314-4634
- The International Workshop "Autonomic Computing and Automatic Control in Computer Systems" (ACACCS), 3-4 October 2013, a co-event of the International Conference "Аutomatics and Informatics 2013", 3-7 October 2013, Sofia, Bulgaria, with Selected papers published in the Scopus indexed Journal "Cybernetics and Information Technologies", 13(4), 2013, ISSN 1314-4081
- The Workshop "Information and Communication Technologies for Human Health and Quality of Life" (ICT-HuHeQuL), 15-17 May 2013, Stara Zagora - Mineral Baths, Bulgaria. The extended versions of 6 talks presented at the Workshop were published in two referenced Bulgarian scientific journals: "Bulgarian Journal of Agricultural Science" Vol. 20 (2014) ISSN 1310-0351 and "Cybernetics and Information Technology" 13(4), 2013, ISSN 1314-4081
Twenty Technology Transfer Events enabled know-how transfer to User Communities in AComIn:
- International Workshop on Industrial and Applied Mathematics, 21 December 2015
- International Workshop on Mathematics in Industry, 14 September 2015
- Know How Transfer on Speech Technologies, June-September 2015
- Know-How Transfer on Acoustic Imaging, April-September 2015
- International Workshop on Advanced Material Characterisation, Modelling, and Numerical Simulations, 28 June - 1 July 2015
- International Workshop on Advanced Techniques in NonDestructive Testing, 18-19 June 2015
- International Workshop on Advanced Computing for Innovation – Industrial Applications, 14-15 May 2015
- Workshop on 3D Digitisation and Virtual Reality, 18 April 2015
- Workshop on Microstructure Material Analysis, 17 April 2015
- International Workshop on Biomedical Simulations, 4 December 2014
- Workshop on Robotics and Innovations, 18-19 September 2014
- International Workshop on 3D Visualization of Cultural Heritage, 10 September 2014
- International Workshop "Walk on Equations and Sequential Monte Carlo to Solve Linear Systems", 21 August 2014
- International Seminar "Challenges connected with the use of resources of large digital libraries in education and civil science", 9 July 2014
- International Seminar "Introduction of game elements in training systems and education", 3 July 2014
- Workshop on 3D-Scanning and digitization, 17 March 2014
- Workshop on Thermography and its Applications, 11-13 February 2014
- International Seminar on Industrial Mathematics, 19 December 2013
- Technology Transfer Seminar "3D Technologies in the Textile Industry and Fashion", 2-5 September 2013
- Technology Transfer Seminar on Computational Vision Applied to Medical Diagnostics, 24-26 July 2013
Three Doors Open Days with demonstrations of AComIn applied results (using the SmartLab equipment) were organised:
- AComIn Doors Open Days 15-16 January 2016, Sofia, Bulgaria
- AComIn Doors Open Days 17-18 April 2015, Sofia, Bulgaria
- AComIn Doors Open Days 28-29 March 2014, Sofia, Bulgaria
Further presentations to leading figures of the State Authorities and the European Commission as well as the society at large are listed below:
- AComIn was presented to the President of the Republic of Bulgaria – Rosen Plevneliev who visited the Institute of Information and Communication Technologies on 10 February 2014
- AComIn was presented to Mr. Wolfgang Burtscher, Deputy Director General of DG Research and Innovation, European Commission, who visited the Institute of Information and Communication Technologies on 17 February 2014
- AComIn was presented at the Meeting of the Science, Technologies and Innovation Expert Council to the Mayor of Sofia Municipality held in the Institute of Information and Communication Technologies on 14 July 2015
- AComIn poster was shown in Sofia downtown as part of the exhibition "Bulgarian Academy of Sciences - science for the society and the country", dedicated to the 145th Anniversary of the Bulgarian Academy of Sciences, in June-July 2014
- AComIn results were demonstrated at the Promotional presentation of the Exhibition "the Battle of Pavia", Expo 2015 Milan, Italy, on 25 May 2015
- AComIn results were shown in the Exhibition "The Battle of Pavia 1525" in Visconti Castle, Pavia - an associated event of Expo 2015 in Milan, Italy. The Exhibition was opened in June-November 2015
- Public presentation of AComIn Evaluation Report was organised in Sofia, Bulgaria on 24 March 2016
III. Exploitation of Results
The exploitation of results already started in various applied activities, as listed in the Section "Description of main S & T results/foregrounds". The Table with "Exploitable Foreground" in this report lists six patent applications that have been submitted recently to the Bulgarian Patent Office (BPO), the World Intellectual Property Organisation (WIPO) and the European Patent Office (EPO). One research result is now integrated in a real-life application for commercial exploitation and will be soon in use.
To ensure further and more active exploitation of results the project team plans to keep the sustainability of liaisons with AComIn User Communities. We managed to build a Network of Users from specific areas who want to increase their knowledge about modern ICT applications and to take part in pilot developments. Such a network existed before AComIn too, related to e.g. the Bulgarian GRID community, but AComIn and the installation of SmartLab accelerated significantly the process of User Communities formation. The attracted participants in the Technology Transfer Seminars in AComIn year 3 (more than 400 people) will further disseminate the news about SmartLab potential in particular and IICT-BAS capacity in general. These figures show that the AComIn idea to focus on User Communities (among other objectives) was very successful too. It is interesting to note that recently IICT-BAS has established itself as a non-formal transfer center, making successful connections between business and academia. News about the Doors Open Days in IICT-BAS are broadcasted on the national TV channels and the national radio (thanks to the PR-Department of the Bulgarian Academy of Sciences). Many pupils from high-schools come to the Days, as well as students.
Following the recommendations of the AComIn Reviewers, the Network of Users will be extended to foreign countries. Social media will be used to disseminate information about the institute capacity and to keep contacts with potential clients.
List of Websites:
Project website:
http://www.iict.bas.bg/acomin
Project Newsletters in English:
http://iict.bas.bg/acomin/e-newsletters.html
Project Newsletters in Bulgarian:
http://iict.bas.bg/acomin/bg/e-newsletters.html
Scientific events organised within AComIn (list and short reports):
http://iict.bas.bg/acomin/events.html
Technology Transfer events organised within AComIn (list and short reports):
http://iict.bas.bg/acomin/user_communities.html
AComIn Movies, media reactions and other signs of AComIn recognition (list and short reports):
http://iict.bas.bg/acomin/appreciation.html
Project Deliverables with public access:
http://iict.bas.bg/acomin/deliverables.html
Lists of all project publications including ones without DOI:
http://iict.bas.bg/acomin/publications_talks.html
http://iict.bas.bg/acomin/monographs.html
Contact details:
Coordinator: Prof. Galia Angelova, Dr.Sc.
Coordinator's e-mail: galia@lml.bas.bg
Institute of Information and Communication Technologies - Bulgarian Academy of Sciences
Acad. G. Bonchev St., block 2
Sofia - 1113
Bulgaria
tel. +3592 979 6611
e-mail: iict@bas.bg