Felix Jahn

Felix Jahn

PhD Student

DFKI – Neuro-Mechanistic Modeling

CaMoRe – Causal Models and Representations

Modeling and Simulation Group, Saarland University

Saarland Informatics Campus

Campus D3.2 (DFKI), Room +1.13

66123 Saarbrücken

Germany

About Me

After finishing my Masters in Mathematics & Computer Science at Saarland University in summer 2023, I joined the Modeling and Simulation (MOSI) group of Verena Wolf at the Computer Science Department of Saarland University as a Ph.D. student. In the meanwhile, I moved within the NMM/MOSI cosmos to the closely connected research department for Neuro-Mechanistic Modeling (NMM) at the German Research Center for Artificial Intelligence (DFKI), where I am associated with the research group for Causal Models and Representations (CaMoRe) under the lead of Jonas Wahl, who is also my main Ph.D. advisor. I work on the MAC-MERLin project dealing with Multi-level Abstractions and Causal Modeling for Enhanced Reinforcement learning.

My research thereby mainly targets the field of causal reinforcement learning, that is, the incorporation of causal knowledge into the reinforcement learning pipeline with the goal to improve sample efficiency, generalizability, interpretability, and/or safety. I focus on foundational research by trying to develop new formalisms enabling a causal perspective to sequential decision-making. This is largely inspired by existing work on causal bandits. Further, I am also interested in using ideas from the formal methods community to reason about causality, for instance, for finding causal explanations for the behavior of a reinforcement learning agent.

Other Work / Research Activity

Besides my main research in the area of causal reinforcement learning, I have an interest in a broad range of further (scientific) topics resulting in a number of side projects. In particular, it is important to me to also consider the social, ethical, and political aspects of our work as computer scientists. Therefore, I am grateful to also participate in the interdisciplinary research group Responsible AI and Machine Ethics (RAIME) under the lead of Kevin Baum, in which I not only do my own research on artificial moral agency, but also have the chance to discuss and think about valuable and exciting philosophical topics around digitalization and AI. Based on my passion for teaching, I have started a project to develop an artificial teaching assistant providing individual support for students practicing lecture contents. Lastly, I am also part of the KIsu project, which aims to promote digital sovereignty in internet searches for people with special needs across several generations by developing a tool that supports the search.

Interests

My research mainly targets the field of causal reinforcement learning, that is, the incorporation of causal knowledge into the reinforcement learning pipeline with the goal to improve sample efficiency, generalizability, interpretability, and/or safety. I focus on foundational research by trying to develop new formalisms enabling a causal perspective to sequential decision-making.
  • Causal Reinforcement Learning
  • Formal Methods
  • Sequential Decision-Making
  • Causal Explanations

Education

Publications (4)

2024

2 papers

Counterfactual Explanations for MITL Violations

Bernd Finkbeiner, Felix Jahn and Julian Siber

2024 — 44th IARCS Annual Conference on Foundations of Software Technology and Theoretical Computer Science (FSTTCS 2024). IARCS Annual Conference on Foundations of Software Technology and Theoretical Computer Science (FSTTCS-2024), December 16-18

@inproceedings{finkbeinerCounterfactualExplanationsForMITL2024,
 author = {Bernd Finkbeiner and Felix Jahn and Julian Siber},
 booktitle = {44th IARCS Annual Conference on Foundations of Software Technology and Theoretical Computer Science (FSTTCS 2024). IARCS Annual Conference on Foundations of Software Technology and Theoretical Computer Science (FSTTCS-2024), December 16-18},
 doi = {10.4230/LIPIcs.FSTTCS.2024.22},
 editor = {Siddharth Barman and Slawomir Lasota},
 isbn = {978-3-95977-355-3},
 month = {12},
 pages = {22:1-22:25},
 publisher = {Schloss Dagstuhl -- Leibniz-Zentrum für Informatik, Dagstuhl, Germany},
 series = {Leibniz International Proceedings in Informatics, LIPIcs},
 title = {Counterfactual Explanations for MITL Violations},
 volume = {323},
 year = {2024}
}
DOI

Acting for the Right Reasons: Creating Reason-Sensitive Artificial Moral Agents

Kevin Baum, Lisa Dargasz, Felix Jahn, Timo P. Gros and Verena Wolf

2024 — ArXiv e-prints (arxiv)

@article{baumActingForTheRight2024,
 author = {Kevin Baum and Lisa Dargasz and Felix Jahn and Timo P. Gros and Verena Wolf},
 doi = {arXiv:2409.15014},
 journal = {ArXiv e-prints (arxiv)},
 month = {9},
 pages = {1-10},
 publisher = {arXiv},
 title = {Acting for the Right Reasons: Creating Reason-Sensitive Artificial Moral Agents},
 volume = {1},
 year = {2024}
}
DOI

2023

2 papers

A Computational Cantor-Bernstein and Myhill's Isomorphism Theorem in Constructive Type Theory

Yannick Forster, Felix Jahn and Gert Smolka

2023 — CPP ‘23: 12th ACM SIGPLAN International Conference on Certified Programs and Proofs. ACM SIGPLAN International Conference on Certified Programs and Proofs (CPP-2023), January 16-17, Boston,, MA, USA

@inproceedings{forsterAComputationalCantorBernsteinAnd2023,
 author = {Yannick Forster and Felix Jahn and Gert Smolka},
 booktitle = {CPP '23: 12th ACM SIGPLAN International Conference on Certified Programs and Proofs. ACM SIGPLAN International Conference on Certified Programs and Proofs (CPP-2023), January 16-17, Boston,, MA, USA},
 doi = {10.1145/3573105.3575690},
 isbn = {979-8-4007-0026-2},
 month = {1},
 pages = {159-166},
 publisher = {ACM},
 title = {A Computational Cantor-Bernstein and Myhill's Isomorphism Theorem in Constructive Type Theory},
 url = {https://inria.hal.science/hal-03891390},
 year = {2023}
}
DOI URL

Constructive and Synthetic Reducibility Degrees: Post's Problem for Many-one and Truth-table Reducibility in Coq

Yannick Forster and Felix Jahn

2023 — CSL 2023 - 31st EACSL Annual Conference on Computer Science Logic. EACSL Annual Conference on Computer Science Logic (CSL), Warsaw, Poland

@inproceedings{forsterConstructiveAndSyntheticReducibility2023,
 author = {Yannick Forster and Felix Jahn},
 booktitle = {CSL 2023 - 31st EACSL Annual Conference on Computer Science Logic. EACSL Annual Conference on Computer Science Logic (CSL), Warsaw, Poland},
 doi = {10.4230/LIPIcs.CSL.2023.16},
 pages = {1-21},
 publisher = {Schloss Dagstuhl Leibniz-Zentrum für Informatik, Dagstuhl Publishing},
 title = {Constructive and Synthetic Reducibility Degrees: Post's Problem for Many-one and Truth-table Reducibility in Coq},
 url = {https://inria.hal.science/hal-03901942},
 year = {2023}
}
DOI URL

Teaching (8)

  • Spatiotemporal Models and Inference

    Assistant, in Summer 2025
    Seminar at Saarland University

  • Statistics Lab

    Assistant, in Summer 2024
    Lecture at Saarland University

  • Special Topics in Deep Reinforcement Learning

    Assistant, in Summer 2024
    Seminar at Saarland University

  • Reinforcement Learning

    Assistant, in Winter 2023/24
    Lecture at Saarland University

  • Ethics for Nerds

    Tutor, in Summer 2021
    Lecture at Saarland University

  • Concurrent Programming

    Tutor, in Summer 2020
    Lecture at Saarland University

  • Introduction to Theoretical Computer Science

    Tutor, in Winter 2019/20
    Lecture at Saarland University

  • Programming 1

    Tutor, in Winter 2018/19
    Lecture at Saarland University

Supervised Students (5)

  • Julius Gabelmann

    Artificial Learning Assistance: A Modular Architecture for LLM-Based Educational Chatbots

    Master's Thesis, ongoing

  • Pascal Schumann

    Learning Continuous Dosage Strategies for Sepsis via Offline Reinforcement Learning

    Master's Thesis, ongoing

  • Muhammad Haris Owais Ahmed

    A Causal Reinforcement Learning Benchmark

    Master's Thesis, ongoing

  • Jannik Müller

    Using Causal Environment Knowledge for Agent Guidance

    Bachelor's Thesis, 2025

  • Lisa Dargasz

    Integrating Reason-Based Moral Decision-Making in the Reinforcement Learning Architecture

    Master's Thesis, 2025