5 papers
Task Vector Descent: Learning from Non-IID Batches
Anton Baumann, Jonas Hübotter, Zeynep Akata +1
A central challenge in continual learning is to acquire new knowledge without forgetting what the model has already learned. This challenge appears in language model training when…
On Repulsive and Attractive Teachers: Separating Correctness from Behavior in Self-Distillation
Anton Baumann, Akmal Ashirmatov, Leo Schmidt-Traub +4
On-policy self-distillation provides dense, token-level supervision by conditioning a model on privileged information and distilling the resulting teacher distribution back into th…
Reinforcement Learning via Self-Distillation
Jonas Hübotter, Frederike Lübeck, Lejs Behric +8
Large language models are increasingly post-trained with reinforcement learning in verifiable domains such as code and math. Yet, current methods for reinforcement learning with ve…
Post-hoc Probabilistic Vision-Language Models
Anton Baumann, Rui Li, Marcus Klasson +5
Vision-language models (VLMs), such as CLIP and SigLIP, have found remarkable success in classification, retrieval, and generative tasks. For this, VLMs deterministically map image…
Probabilistic MIMO U-Net: Efficient and Accurate Uncertainty Estimation for Pixel-wise Regression
Anton Baumann, Thomas Roßberg, Michael Schmitt
Uncertainty estimation in machine learning is paramount for enhancing the reliability and interpretability of predictive models, especially in high-stakes real-world scenarios. Des…