Periodic Reporting for period 1 - AI4EOSC (Artificial Intelligence for the European Open Science Cloud)
Période du rapport: 2022-09-01 au 2023-08-31
The AI4EOSC project delivers an enhanced set of advanced services for the development of Artificial Intelligence (AI), Machine Learning (ML) and Deep Learning (DL) models and applications in the European Open Science Cloud (EOSC). These services are bundled together into a comprehensive platform providing advanced features such as distributed, federated and split learning; novel provenance metadata for AI/ML/DL models; event-driven data processing services or provisioning of AI/ML/DL services based on serverless computing. The project builds on top of the DEEP-Hybrid-DataCloud outcomes and the EOSC compute platform and services in order to provide this specialized compute platform. Moreover, AI4EOSC offers customization components in order to provide tailor made deployments of the platform, adapting to the evolving user needs.
The main outcomes of the AI4EOSC project will be a measurable increase of the number of advanced, high level, customizable services available through the EOSC portal, serving as a catalyst for researchers, facilitating the collaboration, easing access to high-end pan-European resources and reducing the time to results; paired with concrete contributions to the EOSC exploitation perspective, creating a new channel to support the build-up of the EOSC Artificial Intelligence and Machine Learning community of practice.
The main outcomes of the AI4EOSC project will be a measurable increase of the number of advanced, high level, customizable services available through the EOSC portal, serving as a catalyst for researchers, facilitating the collaboration, easing access to high-end pan-European resources and reducing the time to results; paired with concrete contributions to the EOSC exploitation perspective, creating a new channel to support the build-up of the EOSC Artificial Intelligence and Machine Learning community of practice.
During the first year of the project the AI4EOSC consortium has focused on improving the DEEP-HDC platform and services, superseding the DEEP production platform. The project is delivering a platform (the AI4EOSC platform) and a software stack (the AI4OS software) that can be used to build customizable AI platforms following a cloud service model.
• The AI4EOSC platform provides a set of integrated and specific services (i.e. marketplace, API and training dashboard, development environments, serverless inference service) to build AI applications in the EOSC context. These assets are executed in a distributed and federated manner on top of Cloud resources that are already part of the EOSC ecosystem, being also part of the EGI Federated cloud. As a consequence, EOSC users can get easy access to the state-of-the-art pan-European e-Infrastructures to build artificial intelligence, machine learning and deep learning based applications and models, without dealing directly with the infrastructure.
• The AI4OS software stack is the software collection used to build the aforementioned platform. This software can be used to build customised systems compatible with the AI4EOSC platform. This software stack has been designed in a modular way, so that components can be reused (e.g. consuming models from the AI4EOSC marketplace instead of having its own marketplace) to provide different customization paths depending on the needs of the operators of the platform. The AI4OS based platforms can be easily integrated with the mainstream AAI systems in the EOSC ecosystem based on OpenID Connect and provides mechanisms to be easily integrated with accounting systems and other EOSC-core related systems.
The AI4EOSC consortium is offering its services and the AI4OS software stack to industrial and innovation stakeholders through the EOSC Digital Innovation Hub (EOSC DIH), and it is actively engaging with other projects and initiatives under the INFRAEOSC topics and beyond.
• The AI4EOSC platform provides a set of integrated and specific services (i.e. marketplace, API and training dashboard, development environments, serverless inference service) to build AI applications in the EOSC context. These assets are executed in a distributed and federated manner on top of Cloud resources that are already part of the EOSC ecosystem, being also part of the EGI Federated cloud. As a consequence, EOSC users can get easy access to the state-of-the-art pan-European e-Infrastructures to build artificial intelligence, machine learning and deep learning based applications and models, without dealing directly with the infrastructure.
• The AI4OS software stack is the software collection used to build the aforementioned platform. This software can be used to build customised systems compatible with the AI4EOSC platform. This software stack has been designed in a modular way, so that components can be reused (e.g. consuming models from the AI4EOSC marketplace instead of having its own marketplace) to provide different customization paths depending on the needs of the operators of the platform. The AI4OS based platforms can be easily integrated with the mainstream AAI systems in the EOSC ecosystem based on OpenID Connect and provides mechanisms to be easily integrated with accounting systems and other EOSC-core related systems.
The AI4EOSC consortium is offering its services and the AI4OS software stack to industrial and innovation stakeholders through the EOSC Digital Innovation Hub (EOSC DIH), and it is actively engaging with other projects and initiatives under the INFRAEOSC topics and beyond.
The AI4EOSC project is pushing the envelope in terms of distributed platforms for AI with a special focus on scientific users and the easy exploitation of pan-European e-Infrastructures. On the technological side, we are exploiting distributed compute and storage by implementing a scientific service mesh to bring together resources coming from EU e-Infrastructures (and more specially Cloud resources) transparently for the users, making possible for resource providers to share resources between different platform installations. The AI4EOSC technological framework allows to implement, at the platform level, sidecar tasks that are executed transparently for the users, making it possible to deliver more complex functionality (like access to specialized storage systems or the implementation of smart caching mechanisms). Moreover, at a higher level, we are delivering advances solutions for distributed and federated learning, composite AI, model monitoring and drift detection (MLOps) among others integrated into the AI4EOSC platform and the AI4OS software stack.
The AI4EOSC project is delivering a public catalogue of services following the layered cloud service model (IaaS, PaaS, FaaS, SaaS) exploiting in a structural way resources coming from these different layers in the cloud-based EOSC resources, providing rich services to build and deploy customizable machine learning, deep learning and artificial intelligence applications following a platform and serverless approach with horizontal scalability over the EOSC continuum. In this context, the project is also working towards the creation an EOSC AI exchange and community of practice around AI/ML/DL application development and deployment at scale, including but not limited to a noticeable knowledge body in the form of best practices, documentation and publications.
The AI4EOSC project is delivering a public catalogue of services following the layered cloud service model (IaaS, PaaS, FaaS, SaaS) exploiting in a structural way resources coming from these different layers in the cloud-based EOSC resources, providing rich services to build and deploy customizable machine learning, deep learning and artificial intelligence applications following a platform and serverless approach with horizontal scalability over the EOSC continuum. In this context, the project is also working towards the creation an EOSC AI exchange and community of practice around AI/ML/DL application development and deployment at scale, including but not limited to a noticeable knowledge body in the form of best practices, documentation and publications.