Filip Kučera

PhD Candidate at JMU Würzburg. Member of the Media Bias Group. Currently at NII Tokyo!

FilipK_NTK.jpg

2-1-2 Hitotsubashi

Chiyoda-ku

Tokyo

My research focuses on evaluating and developing methods of mechanistic interpretability to detect, quantify and mitigate biases within Vision-Language Models.

I am in my first year of PhD, supervised by Radu Timofte at the JMU Würzburg, working closely with the Media Bias Group in-person at NII Tokyo, invited here by Prof. Echizen-sensei.

I’ve extensive teaching experience - taught Computer Architectures course at the CTU Prague university, taught robotics course at a Grammar School, years of high-school robotic workshops teaching for the National Library of Technology in Prague, years of teaching preparatory programming Bootcamp for newly admitted students at the FEE CTU. I have also worked as a research fellow in various labs since my first year of studies and after my first year I’ve delivered a conference talk about my first project.

research Interests

  • Mechanistic Interpretability
  • Vision-language models
  • Uncovering models inside models

current Projects

  • Systematic Literature Review
  • Automated LLM (DSPy powered) definition extraction from scientific papers

selected publications

  1. arXiv
    SciDef: Automating Definition Extraction from Academic Literature with Large Language Models
    Filip Kučera, Christoph Mandl, Isao Echizen, and 2 more authors
    2026
  2. RiE
    Single Session Walking Robot Workshop for High School Students
    Martin Zoula and Filip Kučera
    In International Conference on Robotics in Education, 2023