AI4EOSC : customizable AI platform in the EOSC context

The AI/ML/DL landscape is evolving rapidly, thanks to the advances in computing hardware technologies and data availability. Machine learning and deep learning methods are now used in more and more scientific domains and industries. AI4EOSC project [1] aims to significantly pursue the goal of "Enabling an operational, open and FAIR EOSC ecosystem" (INFRAEOSC) by creating an open platform tailored to the needs of researchers and engineers working with AI/ML/DL techniques. The research effort targets to provide advanced features to cover all stages of the AI/ML/DL life cycle on the platform: data preparation and model development, experiment tracking, use of various modern enhanced techniques such as e.g. federated learning, split learning, composite AI, support for domain metadata and model provenance, the share of developed models with scientific communities adhering to the FAIR data and research principles, serving models in production in the scalable environment, monitoring of models for a possible concept drift and data drift, leverage MLOps approach.

First, there were conducted extensive scientific reviews of privacy-preserving and federated frameworks and tools, and of the machine learning life cycle management landscape. ... mehrThese reviews identified current approaches, their advantages and applicability for real-life applications. Another important direction of research is on domain metadata and model provenance, where semantics in AI/ML/DL pipelines, vocabulary and ontology for the data and model description become important. Since in real life, AI/ML/DL cycle does not end with the deployment of an AI model in production, one other research focus is on monitoring models and data in production. From the computer science point of view, the platform we are developing not only combines different solutions for the AI/ML/DL life cycle but also prioritises support of distributed computing and usage of heterogeneous and federated resources through a service mesh. Three real-life applications in research topics of agrometeorology, plant protection, and automated thermography, are used to verify ongoing activities.

We are going to demonstrate our complex studies and discuss current approaches.

References:
[1] https://ai4eosc.eu