Harald Binder: Unified Approaches to Small Data. Integrating Similarity, Transfer, and Uncertainty through Data- and Knowledge-Driven Methods

Abstract

Recent advances in AI have been largely driven by big data and data-driven approaches like deep learning. However, many applications, particularly in biomedicine, face challenges with these approaches due to limited data availability. Addressing these “small data” scenarios requires innovative methods that integrate data-driven techniques, such as pre-training and meta-learning, with knowledge-driven approaches, like regression models and differential equations, and also foundation models.

Our Collaborative Research Center, SmallData, focuses on three core areas: similarity, transfer, and uncertainty, aiming to create a unified interdisciplinary framework to address this problem. By combining contributions from computer science, mathematics, and statistics with input from biomedical experts, we develop novel methods tailored to applications in gene therapy, nephrology, psychiatry, radiology, and more. These efforts aim to advance small data solutions, foster interdisciplinary collaboration, and drive biomedical discoveries.

About

Harald Binder

Small Data Initiative

https://www.smalldata-initiative.de/

Freiburg AI

https://freiburg.ai/