Schwinn team

SECURE MACHINE LEARNING

Schwinn's Lab research focuses on secure machine learning: understanding, measuring, and improving the resilience of machine learning models to adversarial attacks, distribution shifts, and unsafe outputs. A core focus of his work is studying the vulnerabilities of large language models (LLMs), including the reliability of safety evaluations, adversarial training, and machine unlearning. He also works on data-efficient learning, foundation models for time-series forecasting, and small detail processing in VLMs.

Leo is a faculty member of the ELLIS Unit Munich and a visiting researcher at the Mila Quebec AI Institute, where he is hosted by Prof. Gauthier Gidel. He joined Helmholtz AI in May 2026, hosted at Helmholtz Munich.

External website: Visit Leo Schwinn's research website →