The AI4EOSC (AI for the European Open Science Cloud) project is aiming to deliver an enhanced set of services for the development of Artificial Intelligence (AI), Machine Learning (ML), and Deep Learning (DL) models and applications. This is to be reached by increasing the service offered in the EU landscape by expanding the European Open Science Cloud (EOSC) ecosystem. The services will make use of advanced features such as distributed, federated, and split learning; provenance metadata; event-driven data processing services or provisioning of services based on serverless computing.
The project will focus on tools to provide AI, and ML DL services by integrating into it real-life use cases to co-design the project proposal and drive our integration activities. AI4EOSC bases its activities on the technological framework delivered by the DEEP-Hybrid-DataCloud H2020 project. The DEEP platform (provided through the EOSC portal 2) is a production-ready system that is being effectively used by researchers in the EU to train and develop machine learning and deep learning models. A special emphasis is made on ensuring that all the research outputs and sub-products (data, models, metadata, publications, etc.) adhere to the FAIR data and research principles.
The AI4EOSC consortium has been assembled to ensure a skills-balanced and complementary set of partners with a strong research, development, technological, and innovation background. The consortium gathers several of the most active institutions in the EOSC in terms of development, implementation, deployment, and operation of distributed pan-European e-infrastructures experienced and highly innovative SMEs with a huge potential in the AI field, and a wide experience in technological endeavors. All partners involved in the project activities have wide experience in software development. As such, several academic partners have developed key components used in the production of EU e-Infrastructures. The consortium is comprised of 10 partners including academic - the project coordinator CSIC, PSNC, LIP, KIT, UPV, IISAS, INFN, industrial - Predictia, MicroStep-MIS, and WODR.
The MicroStep-MIS' role in the project is to provide an agrometeorological use case, especially a new innovative nowcasting model for thunderstorms on the EOSC platform. It is expected that due to climate change, there will be higher numbers and more intense thunderstorms, which can cause farmers damaged by a variety of means. Prompt actions before predicted thunderstorms may help minimize the impact, therefore more accurate and easily available nowcasting model of thunderstorms is our plan. Thunderstorm nowcasting state-of-the-art models are usually based on extrapolation. Employment of AI techniques will be a significant step forward because this technology can grasp complex non-linear features of natural processes. The usage of large-scale radar data and precise local point measurement provides both preciseness (inherited from point stations) and spatial coverage. The new model will be trained by AI techniques on the EOSC platform using all available inputs as predictors, measured data from stations, outputs of the NWP model, and outputs of commonly used extrapolation methods on radar data.
