Anton Baumann in the mountains

Anton Baumann

PhD student at ETH Zurich
antbaumann [at] ethz [dot] ch


About Me

I'm a PhD student at ETH Zurich, supervised by Andreas Krause and Zeynep Akata (TUM). I'm interested in probabilistic machine learning, continual learning, and sample-efficient reinforcement learning for LLMs.

Always feel free to reach out :)

News

Blog Posts

Selected Publications

  1. Task-vector descent across three tasks: scaled task vectors update model parameters along average-loss contours.
    Task Vector Descent: Learning from Non-IID Batches
    Anton Baumann, Jonas Hübotter, Zeynep Akata, Andreas Krause
    NeurIPS 2026 FLLMPT workshop
    OralNeurIPS 2026 Workshop FLLMPT
  2. Visual abstract of contrastive self-distillation: attractive and repulsive teacher signals separate correct and incorrect answers from student behavior.
    On Repulsive and Attractive Teachers: Separating Correctness from Behavior in Self-Distillation
    Anton Baumann, Akmal Ashirmatov, Leo Schmidt-Traub, Frederike Lübeck, Jonas Hübotter, Thomas Kleine Buening, Andreas Krause
    NeurIPS 2026 FLLMPT workshop
    Preprint Blog Post
    PosterNeurIPS 2026 Workshop FLLMPT
  3. Comparison of GRPO and SDPO: GRPO assigns a shared signal to generated tokens, while SDPO provides detailed feedback across tokens and vocabulary alternatives.
    Reinforcement Learning via Self-Distillation
    Jonas Hübotter, Frederike Lübeck*, Lejs Behric*, Anton Baumann*, Marco Bagatella, Daniel Marta, Ido Hakimi, Idan Shenfeld, Thomas Kleine Buening, Carlos Guestrin, Andreas Krause
    ICML 2026 *equal contribution
    Preprint Project Page Code
    Best PaperICLR 2026 Workshop on Test-Time Updates
    OralICLR 2026 Workshop on Scaling Post-training for LLMs
  4. BayesVLM overview: a Laplace approximation turns frozen image and text encoder outputs into probabilistic embeddings and cosine similarities.
    Post-hoc Probabilistic Vision-Language Models
    Anton Baumann, Rui Li, Marcus Klasson, Santeri Mentu, Shyamgopal Karthik, Zeynep Akata, Arno Solin, Martin Trapp
    ICLR 2026
    Preprint Project Page Code
    SpotlightNeurIPS 2024 Workshop on Bayesian Decision-making and Uncertainty
  5. Probabilistic MIMO U-Net architecture: three permuted input batches pass through separate encoders, a shared core, and separate decoders with skip connections.
    Probabilistic MIMO U-Net
    Anton Baumann, Thomas Roßberg, Michael Schmitt
    ICCV 2023 UnCV workshop
    Preprint Code
    OralICCV 2023 Workshop on Uncertainty in Computer Vision

Powered by the Minimal Light theme.