Periodic Reporting for period 1 - AIatEDGE (A secure and reusable Artificial Intelligence platform for Edge computing in beyond 5G Networks)
Période du rapport: 2021-01-01 au 2022-06-30
The EC has pointed out how high-performance, intelligent, and secure networks are fundamental for the evolution of the multi-service Next Generation Internet (NGI), a position further confirmed by the 5G-PPP initiative for a European public-private partnership on SN&S.
In this scenario, telecom operators will have the opportunity to provide innovative solutions to combine the advanced capabilities of 5G with the fluid cloud-based application development processes that just emerged. AI-enabled applications are being foreseen as one of the pillars that will boost the fourth industrial revolution.
However, while great progress has been made with respect to the accuracy and performance of AI-enabled applications, their integration in potentially autonomous decision-making systems or even critical applications requires End-to-End quality assurance, ubiquitous availability and low-latency transport of data to the compute resource involved in connected and trustworthy AI.
5G is a paradigm shift: its high performance in terms of latency, bitrate, and reliability calls for a technological and business convergence between cloud computing and the telecom worlds. 5G features like slicing, edge computing, and better and more flexible radio connectivity can be used to support qualitatively different applications and to deliver a richer user experience, faster interactions, large-scale data processing, and machine-to-machine communications. Nevertheless, the challenges to be overcome are still notable. In particular, the increasing number of control and optimization dimensions of the end-to-end 5G infrastructure may result in an overly complex network that operators and vendors may find it difficult to operate, manage, and evolve. The introduction of AI and Machine Learning technologies in the cloud-network convergence process will be crucial to help achieve a higher level of automation, increasing network performance and decreasing the time-to-market of new features. Early attempts at applying AI/ML in the cellular domain can be found. Nevertheless, it cannot be expected that each and every subsystem of future access, edge, core, and cloud segments will employ distinct and separated AI tools and datasets. Such an approach would lead to AI-silos slowing down advances.
The goal of AI@EDGE is to leverage the concept of reusable, secure, and trustworthy AI for network automation to achieve an EU-wide impact on industry-relevant aspects of the multi-stakeholders environments. AI@EDGE has two lines of action: to design, prototype, and validate a network and service automation platform capable of supporting flexible and programmable pipelines for the creation, utilization, and adaptation of secure and privacy-aware AI/ML models; and to orchestrate AI-enabled end-to-end applications. The novel concept of Artificial Intelligence Functions (AIFs) is introduced.
In this scenario, telecom operators will have the opportunity to provide innovative solutions to combine the advanced capabilities of 5G with the fluid cloud-based application development processes that just emerged. AI-enabled applications are being foreseen as one of the pillars that will boost the fourth industrial revolution.
However, while great progress has been made with respect to the accuracy and performance of AI-enabled applications, their integration in potentially autonomous decision-making systems or even critical applications requires End-to-End quality assurance, ubiquitous availability and low-latency transport of data to the compute resource involved in connected and trustworthy AI.
5G is a paradigm shift: its high performance in terms of latency, bitrate, and reliability calls for a technological and business convergence between cloud computing and the telecom worlds. 5G features like slicing, edge computing, and better and more flexible radio connectivity can be used to support qualitatively different applications and to deliver a richer user experience, faster interactions, large-scale data processing, and machine-to-machine communications. Nevertheless, the challenges to be overcome are still notable. In particular, the increasing number of control and optimization dimensions of the end-to-end 5G infrastructure may result in an overly complex network that operators and vendors may find it difficult to operate, manage, and evolve. The introduction of AI and Machine Learning technologies in the cloud-network convergence process will be crucial to help achieve a higher level of automation, increasing network performance and decreasing the time-to-market of new features. Early attempts at applying AI/ML in the cellular domain can be found. Nevertheless, it cannot be expected that each and every subsystem of future access, edge, core, and cloud segments will employ distinct and separated AI tools and datasets. Such an approach would lead to AI-silos slowing down advances.
The goal of AI@EDGE is to leverage the concept of reusable, secure, and trustworthy AI for network automation to achieve an EU-wide impact on industry-relevant aspects of the multi-stakeholders environments. AI@EDGE has two lines of action: to design, prototype, and validate a network and service automation platform capable of supporting flexible and programmable pipelines for the creation, utilization, and adaptation of secure and privacy-aware AI/ML models; and to orchestrate AI-enabled end-to-end applications. The novel concept of Artificial Intelligence Functions (AIFs) is introduced.
AI@EDGE identified the network and systems requirements, developed the system architecture, interfaces, and APIs, and mapped the AI@EDGE technological roadmap into new business models. Subsequently, the overall design has been defined and, in parallel, the project KPIs, socio-economic impact assessment and techno-economic analysis.
Additionally, AI@EDGE designed and implemented the project Network and Service Automation Platform (NSAP) to automate network management using the concept of closed-loop control. The overall architecture was defined, by a multi-tier orchestrator, an intelligent orchestration component (IOC), a slice manager and a non-RT RIC. Several AI/ML-based methods for automation and learning for network management purposes have been developed. The prototype implementation of the NSAP and the automation methods have been started.
Additionally, the AI@EDGE CCP has been designed by taking as reference the ETSI MEC framework and extending it with support from several features including serverless functions, AIFs, and multi-MEC system coordination. In addition, the CCP integrates various technologies, namely 5G access and core, Multipath TCP, disaggregated RAN, and hardware acceleration. The integration of the CCP at the integration site has started. The MEC platform has been developed using the LightEdge platform extended with support for serverless functions and AIFs. The development and integration of the AI@EDGE MEC Orchestrator have also begun. A major result is the initial definition of the AIF Descriptors.
The finalization of the mapping of the AI@EDGE use cases over the AI@EDGE architecture has started, outlining a possible integration of the connect-compute platform components into the use cases testbeds.
Additionally, AI@EDGE designed and implemented the project Network and Service Automation Platform (NSAP) to automate network management using the concept of closed-loop control. The overall architecture was defined, by a multi-tier orchestrator, an intelligent orchestration component (IOC), a slice manager and a non-RT RIC. Several AI/ML-based methods for automation and learning for network management purposes have been developed. The prototype implementation of the NSAP and the automation methods have been started.
Additionally, the AI@EDGE CCP has been designed by taking as reference the ETSI MEC framework and extending it with support from several features including serverless functions, AIFs, and multi-MEC system coordination. In addition, the CCP integrates various technologies, namely 5G access and core, Multipath TCP, disaggregated RAN, and hardware acceleration. The integration of the CCP at the integration site has started. The MEC platform has been developed using the LightEdge platform extended with support for serverless functions and AIFs. The development and integration of the AI@EDGE MEC Orchestrator have also begun. A major result is the initial definition of the AIF Descriptors.
The finalization of the mapping of the AI@EDGE use cases over the AI@EDGE architecture has started, outlining a possible integration of the connect-compute platform components into the use cases testbeds.
The goal of AI@EDGE is to create new and realistic opportunities for generating competitive advantages for the European ICT sector. The vision of innovative and demanding applications and services is set to transform the telecom industry to benefit from the same level of agility of the IT world: time to market for new innovative services significantly improved and an overall reduction of TCO. In the commercial field, AI@EDGE will help to open the market to new actors and will enable worldwide operators to compete and benchmark with edge service availability.
Up to now, the commercial 5G offerings focused on connectivity and capacity, but the ubiquitous connectivity revolution will be followed by the connectivity-computation-storage revolution. The AI@EDGE project accelerates this revolution to maximize the value creation of the mobile networks and give worldwide access to this surge in value creation.
Additionally, AI@EDGE is investigating solutions to automatically adjust and/or attempt to move the execution of services to another location or to a higher architectural layer of the system in case of failures. As a result of this, should any disaster knock out local connectivity, computation, or storage resources, the network will experience just a graceful degradation and will not halt to a stop.
The project has shown significant innovations in several domains. Concerning the overall system architecture and interfaces, significant advances have been made especially in the field of multi-MEC systems coordination. AI@EDGE proposed several architectural concepts that go beyond the state-of-the-art, i.e. the work on a novel interface between multiple MEOs and the definition of explicit and implicit network slices. From the components’ standpoint, the Intelligence Orchestrator Components have been introduced within the NSAP and the MTO. A model for reusable data pipelines has been designed, efficiently providing input data to data-driven network automation methods. Excellent advances have been made in the development of nine machine learning-based methods and algorithms for automation of network management using closed-loop control. Furthermore, methods for data augmentation to increase the robustness of learning and methods for scheduling and placement are developed. The Connect-Compute platform (CCP) extends the ETSI MEC/NFV architectures with applications and models capable of providing the AI@EDGE platform with the context and metadata necessary to take automatically actionable decisions and to realize intelligent data and computation offload control and management of applications and services deployed over the decentralized and distributed AI@EDGE platform.
Up to now, the commercial 5G offerings focused on connectivity and capacity, but the ubiquitous connectivity revolution will be followed by the connectivity-computation-storage revolution. The AI@EDGE project accelerates this revolution to maximize the value creation of the mobile networks and give worldwide access to this surge in value creation.
Additionally, AI@EDGE is investigating solutions to automatically adjust and/or attempt to move the execution of services to another location or to a higher architectural layer of the system in case of failures. As a result of this, should any disaster knock out local connectivity, computation, or storage resources, the network will experience just a graceful degradation and will not halt to a stop.
The project has shown significant innovations in several domains. Concerning the overall system architecture and interfaces, significant advances have been made especially in the field of multi-MEC systems coordination. AI@EDGE proposed several architectural concepts that go beyond the state-of-the-art, i.e. the work on a novel interface between multiple MEOs and the definition of explicit and implicit network slices. From the components’ standpoint, the Intelligence Orchestrator Components have been introduced within the NSAP and the MTO. A model for reusable data pipelines has been designed, efficiently providing input data to data-driven network automation methods. Excellent advances have been made in the development of nine machine learning-based methods and algorithms for automation of network management using closed-loop control. Furthermore, methods for data augmentation to increase the robustness of learning and methods for scheduling and placement are developed. The Connect-Compute platform (CCP) extends the ETSI MEC/NFV architectures with applications and models capable of providing the AI@EDGE platform with the context and metadata necessary to take automatically actionable decisions and to realize intelligent data and computation offload control and management of applications and services deployed over the decentralized and distributed AI@EDGE platform.