Skip to main content

Experience-based Computation: Learning to Optimise

Periodic Reporting for period 1 - ECOLE (Experience-based Computation: Learning to Optimise)

Reporting period: 2018-04-01 to 2020-03-31

Europe is facing challenges for growth of technological companies as the increasing complexity of products, development and production processes requires advanced integration skills and innovative minds. There is an acute shortage of human experts with sufficient skills in tackling current industrial challenges in a holistic manner, and meanwhile there is a need for solving ever more complex problems to tackle such challenges.
The aim of the ECOLE innovative training network is to contribute towards shortening product cycle, reducing resource consumption during complete engineering processes and creating more balanced and innovative products. This will involve not only the development of novel approaches to solving complex engineering optimisation problems, but also the equipment of early career researchers with the skills needed to advance European technological companies in the years to come, enhancing their employability and strengthening European innovation capacity.

ECOLE’s aim will be achieved through the following core objectives:
(1) To develop a “Learning to Optimise” framework. ECOLE will take a bold step forward in solving complex engineering optimisation problems. Instead of just developing technologies to solve a given optimisation problem, it will propose novel techniques to optimise automatically across problems. Through knowledge, skill, and practice derived from problem solving processes in time, the experience of optimising one product or process will be learned and transferred automatically to better solve other complex optimisation problems.
(2) To develop early career researchers’ transferrable skills and industrial experiences. Among others, this will include technical knowledge that enables them to understand the interplay between learning and optimisation, practical skills in integrating various disciplines into joint solutions to complex industrial problems, competences related to intellectual property, management and entrepreneurship, and communication skills.
Eight early career researchers have been recruited. They completed several training activities including training on the core algorithms and theory required to conduct research on Learning to Optimise, reading skills, presentation skills and networking. Training in public engagement and outreach plays an important role for disseminating ECOLE’s vision. The early career researchers have been organizing and participating in outreach events. They also participated in a workshop under the guidance of industry personnel, which focused on project management and IP related topics relevant in industrial settings. Each of them has already completed at least one secondment with industrial partners, putting their acquired skills into practice to develop learning and optimisation approaches for real-world applications.
The Learning to Optimise framework for solving complex engineering optimisation problems requires (a) advancing learning, (b) advancing optimisation, and (c) developing novel approaches to integrate learning with optimisation. The following progress has been achieved so far:
a) Learning:
• Learning from imbalanced data: Real world problems frequently suffer from data imbalances, where data of a given type is available in much more limited quantity than data of other types. Machine learning algorithms struggle to learn well from such data. We investigated the adequacy of different class imbalance techniques to deal with this problem and applied them to classify mesh quality in aerodynamics simulation of a benchmark vehicle body, which allows to realize a fast and resource efficient design process. Algorithms to automatically tune parameters of class imbalance techniques have also been proposed, leading to a more reliable and effective learning process.
• Text representation: Many real-world problems involve making predictions based on text data. For that, existing machine learning algorithms use text representations that are not human-interpretable, leading to lack of human trust in the results. We proposed a novel human-interpretable representation for text and applied it to interpretable user profiling. We validated the method on user generated product review data and showed promising explanation about users’ preference for feature optimization of product design. We wish to later on connect this representation utilizing customers feedback on vehicles into the vehicle development process.
• Prediction approaches: We developed a novel prediction approach with demonstrated improvements in medical data analysis for patient status prediction. We integrated patient meta data, such as disease and demographic information, to model correlations of patient time series, by which prediction depends on not only historic patient information but also information of related patients. Different from existing work, prediction is enhanced by information propagation over the interconnected data network, which can potentially help with automobile design suggestions in the future. Other work focused on the prediction of potential user design modifications for an improved man-machine design process.
b) Optimisation:
• Supporting automobile design: We collected data from various steps of vehicle development process to create a challenging test framework for new optimisation and learning methods. We improved a novel geometric representation based on cutting-edge deep learning technology for practical application on 3D (car) shapes. Such representation has shown to enable efficient optimisation of the design of a benchmark vehicle body.
• Optimisation under uncertainty: Real-world optimisation frequently has to deal with problems whose features are not entirely and accurately known beforehand. We developed a novel optimisation algorithm able to improve optimisation of such challenging problems.

c) Integration:
• Learning to optimise for dynamic problems: Real-world optimisation problems frequently suffer changes over time. We proposed a new algorithm to learn from the optimisation results obtained before a change to more efficiently optimise after the change.
Progress beyond state-of-the-art so far is explained in the previous section and contributes towards different aspects of the Learning to Optimise framework. By the end of the project, the Learning to Optimise framework will be fully developed, leading to demonstrated improvements on efficiency and effectiveness of solving complex real-world optimisation problems, and in particular automobile design optimisation. Potential impact includes shortening product cycle, reducing resource consumption during the automobile engineering process, and an integrative and seamless man-machine design approach towards the creation of automobile innovations. The framework is expected to be transferrable to other domains, contributing towards European technologic industry in general. Impact so far has been mainly in terms of enhancing the career perspectives of the early career researchers, who are expected to continue contributing towards addressing challenges faced by industry after completing their ECOLE training. The ECOLE training model is also expected to contribute towards structuring early-stage research training that involves contributions of the non-academic sector at the European level.