Here we introduce MLentory, an FDO registry for Machine Learning models and their corresponding workflows, from creation to deployment. MLentory relies on FAIR Digital Objects (FDOs) to improve Findability, Accessibility, Interoperability, and Reusability while also improving reproducibility and transparency. MLentory aggregates, harmonizes and FAIRifies data from various ML model and model-related repositories and platforms. Here we present the initial architecture for data extraction, transformation, and loading.
Due to the proliferation of Machine Learning (ML) models, it is necessary a systematic approach to report them. To this end, the ML model cards were proposed in 2019 [1], complemented by Dataset cards [2] by providing additional information on the training datasets. A parallel and complementary effort are platforms facilitating the storing, sharing and reporting of ML models and other artifacts needed for training and deployment, e.g., HuggingFace, neptune.ai, SpaceML, Kipoi, BioImagine Model Zoo, etc. Interoperability across platforms and connection to other ML experiments and related artifacts are still challenging as the corresponding metadata is stored in different ways by different platforms. Harmonization across different efforts is a gap being addressed by different communities, e.g., Research Data Alliance (RDA) FAIR4ML Interest Group, ELIXIR ML Focus Group (ELIXIR MLFG), and National Research Data Infrastructure for Data Science (NFDI4DS) in Germany.
This work has been partially supported by NFDI4DataScience, a consortium funded by the German Research Foundation (DFG), project number 460234259.