Matthias Hein

Unlearning That Lasts: Utility-Preserving, Robust, and Almost Irreversible Forgetting in LLMs

Advancing Compositional Awareness in CLIP with Efficient Fine-Tuning

Mahalanobis++: Improving OOD Detection via Feature Normalization

Robustness in Both Domains: CLIP Needs a Robust Text Encoder

A Realistic Threat Model for Large Language Model Jailbreaks

An Interpretable N-gram Perplexity Threat Model for Large Language Model Jailbreaks