Project summary

The interdisciplinary approach of SmallData is one of our unique features and key to success in small data challenges. We hypothesize that a comprehensive framework for addressing small data challenges can only be achieved by integrating and fusing approaches from different disciplines. For joint overarching topics, we will formalize the integration and fusion of techniques from different disciplines. In particular, we will advance definitions of small data, develop a taxonomy of similarity comprising uncertainty, extend similarity as a pre-condition for transfer, transfer few-shot learning across disciplines, and unify work on attention mechanisms. The joint conceptual developments also will steer the structure of a SmallData Compendium, for making results more broadly useful.