Jan Sobotka

CS Master’s Student & AI/ML Research Assistant

Swiss Federal Institute of Technology in Lausanne (EPFL)

About me

My name is Jan, I am a master’s student in computer science at EPFL and a research assistant at the Autonomous Systems Group at the University of Texas at Austin, where I work on Large Language Models (LLMs) in strategy games.

At a high level, I am interested in (1) understanding how our mind and cognition work and (2) building machines that can perceive, think, and learn. This intersection of artificial intelligence and cognitive computational neuroscience excites me the most. Currently, my primary research interests include the following areas:

(Mechanistic) interpretability
Foundation models
Meta-learning

I am always happy to discuss these topics, so if you have related thoughts or questions, please do not hesitate to contact me.

Interests

(Mechanistic) interpretability
Foundation models
Meta-learning

Education

Master's degree in Computer Science, 2024 - 2026
Swiss Federal Institute of Technology in Lausanne (EPFL)
Bachelor's degree in Informatics, Specialization in Artificial Intelligence, 2021 - 2024
Czech Technical University in Prague

Recent Publications & Preprints

Jan Sobotka, Luca Baroni, Ján Antolík (2025). MEIcoder: Decoding Visual Stimuli from Neural Activity by Leveraging Most Exciting Inputs. Conference on Neural Information Processing Systems (NeurIPS 2025).

PDF OpenReview

Myeongho Jeon, Jan Sobotka, Suhwan Choi, Maria Brbic (2025). Weak-to-Strong Generalization under Distribution Shifts. Conference on Neural Information Processing Systems (NeurIPS 2025).

Poster OpenReview

Jan Sobotka, Auke Ijspeert, Guillaume Bellegarda (2025). Reverse-Engineering Memory in DreamerV3: From Sparse Representations to Functional Circuits. Conference on Neural Information Processing Systems (NeurIPS 2025, Spotlight at Mechanistic Interpretability Workshop).

PDF OpenReview

Mustafa O. Karabag, Jan Sobotka, Ufuk Topcu (2025). Do LLMs Strategically Reveal, Conceal, and Infer Information? A Theoretical and Empirical Analysis in The Chameleon Game. ArXiv.

PDF DOI

Jan Sobotka, Petr Šimánek, Pavel Kordík (2024). Enhancing Fractional Gradient Descent with Learned Optimizers. ArXiv.

PDF DOI

Jan Sobotka, Petr Šimánek, Daniel Vašata (2024). Investigation into the Training Dynamics of Learned Optimizers. 16th International Conference on Agents and Artificial Intelligence (ICAART 2024).

PDF DOI

Jan Sobotka, Petr Šimánek (2024). Investigation into Training Dynamics of Learned Optimizers (Student Abstract). The 38th Annual AAAI Conference on Artificial Intelligence (AAAI-24).

PDF DOI