Theory of Machine Learning Group University of Tübingen

Research Seminar "Machine Learning Theory"

A group member presenting research on explainable machine learning during a retreat

When and where

Each tuesday 15:00 - 16:00, A 401 (= big glassroom on level 4), Maria-von-Lindenstrasse 1. ( From September on we will meet at 14:00 instead of 15:00).

What

This is the internal research seminar by the Theory of Machine Learning group. Most sessions take place in form of a reading group: everybody reads the assigned paper before the meeting. Then we jointly discuss the paper in the meeting. Sometimes we also have talks by guests or members of the group.

Who

TML group members, but other PhD students / researchers can join as well.

Upcoming meetings

  • 20.10.2026 (!!! 15:00, MPI N0.002) Talk by Alendra Olteanu: (Un)rigorous AI
  • 27.10.2026 Paper discussion (Gunnar): LLM-SR: Scientific Equation Discovery via Programming with Large Language Models, ICLR 2025. Parshin Shojaee, Kazem Meidani, Shashank Gupta, Amir Barati Farimani, Chandan Reddy pdf
  • 3.11.2026 Internal discussion regarding AI and publication strategiespdf
  • 10.11.2026 tba pdf
  • 17.11.2026 (? Ulrike on a hiring committee) pdf
  • 24.11. tba (? Ulrike on a hiring committee)pdf
  • 1.12. tba pdf
  • 8.12. (NeurIPS)
  • 15.12. tba pdf
  • 22.12. tba pdf

Past meetings

Listed here.

Suggested papers for future meetings

Feel free to make suggestions!
If you do, please (i) try to select short conference papers rather than 40-page-journal papers; (ii) please put your name when entering suggestions; it does not mean that you need to present it, but then we can judge where it comes from; (iii) Please provide a link, not just a title.
  • Verified SHAP: Provable Bounds for Exact Shapley Values of Neural Networks David Boetius, Shahaf Bassan, Guy Katz, Stefan Leue, Tobias Sutter. icml 2026
  • (Gunnar) How Does the Pretraining Distribution Shape In-Context Learning? A Fundamental Trade-Off. Azizian et al. ICML 2026 link
  • (Ulrike) Shahaf Bassan et al:Formal Mechanistic Interpretability: Automated Circuit Discovery with Provable Guarantees, iclr 2026 link
  • Preference for Explainable AI, 2026, Alex Chan
  • Formal mechanistic interpretability: automated circuit discovery with provable guarantees, Itamar Hadad, Guy Katz, Shahaf Bassan, ICLR 2026.
  • (Ulrike) The Computational Complexity of Circuit Discovery for Inner Interpretability, ICLR 2025, Federico Adolfi, Martina Vilas, Todd Wareham.
  • Training on the Test Task Confounds Evaluation and Emergence. ICML 2025
  • (Ulrike) Classical Statistical (In-Sample) Intuitions Don't Generalize Well: A Note on Bias-Variance Tradeoffs, Overfitting and Moving from Fixed to Random Designs Alicia Curth. arxiv
  • The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity. Jul 2025. pdf
  • Ulrike:What makes an Ensemble (Un) Interpretable? ICML 2025. https://arxiv.org/pdf/2506.08216v1
  • Everything, Everywhere, All at Once: Is Mechanistic Interpretability Identifiable?, ICLR 2025, link (Gunnar)
  • Circuit Tracing: Revealing Computational Graphs in Language Models, link (Gunnar)
  • How to Probe: Simple Yet Effective Techniques for Improving Post-Hoc Explanations, ICLR 2025, link (Gunnar)
  • We should read up on mechanistic interpretability. Suggestions? I vaguely remember somebody mentioning a good paper. Here is one recent iclr paper: link
  • (Sebastian) How Do Transformers Learn Variable Binding in Symbolic Programs? ICML 2025
  • Chain of Log-Concave Markov Chains, Saeed Saremi, Ji Won Park, Francis Bach, 2024 pdf (ulrike)
  • The Value of Prediction in Identifying the Worst-Off, Unai Fischer-Abaigar, Christoph Kern, Juan Perdomo, ICML 2025 (Outstanding Paper) pdf (Gunnar)
  • link