Pflichtseminar Deep Learning (Bachelor) — WS 2026/27
Pflichtseminar Deep Learning, Winter Semester 2026/27
Seminar for Bachelor students in Computer Science, offered as group 10 of the module Pflichtseminar (5 ECTS).
Schedule: Wednesday, 10:30–12:00
Location: Room 2533.00.41
First meeting: Wednesday, October 14th, 2026
Participants: up to 15
Format
The aim of the seminar is to maximize understanding. Each participant works on one research paper and must know the paper and its implementation in detail:
- Talk: 30 minutes, followed by 15 minutes of discussion. Rough structure: 5 minutes introduction, 15 minutes method, 10 minutes experiments and reimplementation.
- Reimplementation: as far as possible, reimplement a feasible part of the paper, e.g. a very small version of one central experiment. What to reimplement is decided together with me as supervisor. GPU access and some Claude tokens are available on request.
- Review: hand in your presentation one week in advance for review.
- Office hours: in the coming weeks, the seminar slots (Wednesday, 10:30–12:00) are used as office hours for discussing the reimplementation and other questions. Please write to me in advance.
Topics are assigned in the first meeting.
Schedule
Two talks per session, starting after the office-hour weeks. Send me your preferred date. No sessions during the Christmas break (23 Dec – 3 Jan).
| Date | Slot | Topic | Presenter |
|---|---|---|---|
| Wed 11 Nov 2026 | 1 | free | |
| Wed 11 Nov 2026 | 2 | free | |
| Wed 18 Nov 2026 | 1 | free | |
| Wed 18 Nov 2026 | 2 | free | |
| Wed 25 Nov 2026 | 1 | 17. Emotion concepts in an LLM | Tommy Marcel Horrig |
| Wed 25 Nov 2026 | 2 | 2. Implicit bias of gradient descent | Hadi Al-Helbawi |
| Wed 2 Dec 2026 | 1 | 12. Harness design for coding agents | Jordy Van den Akker |
| Wed 2 Dec 2026 | 2 | 20. LLMs vs. professional poker players | Nikita Schmunk |
| Wed 9 Dec 2026 | 1 | 6. Edge of stability | Yusa Özdemirkan |
| Wed 9 Dec 2026 | 2 | 5. Wide neural networks as linear models | David Darius Rajski |
| Wed 16 Dec 2026 | 1 | 11. Why do multi-agent LLM systems fail? | Alexander Jan Sadowski |
| Wed 16 Dec 2026 | 2 | 18. LLM Chess | Micha Schopen |
| Wed 6 Jan 2027 | 1 | 9. Multiagent debate | Eren Mersin |
| Wed 6 Jan 2027 | 2 | 10. MetaGPT | Dilara Aldemir |
| Wed 13 Jan 2027 | 1 | 16. LLMs do not simulate human psychology | Fynn Hoffmann |
| Wed 13 Jan 2027 | 2 | 13. Where do AI coding agents fail? | Jad Alsioufi |
| Wed 20 Jan 2027 | 1 | 19. Game Arena | Firas Tounsi |
| Wed 20 Jan 2027 | 2 | 22. HardcoreLogic | Mouhib Kaabachi |
Topics
Training dynamics
- Saxe, McClelland & Ganguli 2013: Exact solutions to the nonlinear dynamics of learning in deep linear neural networks
- Soudry et al. 2017: The Implicit Bias of Gradient Descent on Separable Data
- Draxler, Veschgini, Salmhofer & Hamprecht 2018: Essentially No Barriers in Neural Network Energy Landscape (ICML 2018)
- Belkin et al. 2018: Reconciling modern machine learning practice and the bias-variance trade-off
- Lee et al. 2019: Wide Neural Networks of Any Depth Evolve as Linear Models Under Gradient Descent
- Cohen et al. 2021: Gradient Descent on Neural Networks Typically Occurs at the Edge of Stability
LLMs and mathematical tools
- Requena et al. 2026: A Minimal Agent for Automated Theorem Proving (ICML 2026)
- Xia, Gomes, Selman & Szeider 2026: Agentic Neurosymbolic Collaboration for Mathematical Discovery: A Case Study in Combinatorial Design
Multi-agent systems
- Du et al. 2023: Improving Factuality and Reasoning in Language Models through Multiagent Debate (ICML 2024)
- Hong et al. 2023: MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework (ICLR 2024)
- Cemri et al. 2025: Why Do Multi-Agent LLM Systems Fail?
Working with coding agents
- Fan et al. 2026: An Empirical Study of Harness Design for Coding Agents
- Ehsani et al. 2026: Where Do AI Coding Agents Fail? An Empirical Study of Failed Agentic Pull Requests in GitHub (MSR 2026)
Automated research
- Zhang et al. 2025: MLRC-Bench: Can Language Agents Solve Machine Learning Research Challenges? (NeurIPS 2025)
- Huang et al. 2026: Reward Hacking Challenges Oversight of Autonomous Research Agents
LLMs as test subjects
- Schröder et al. 2025: Large Language Models Do Not Simulate Human Psychology
- Sofroniew et al. 2026: Emotion Concepts and their Function in a Large Language Model (Anthropic; arXiv)
LLMs in games
- Kolasani et al. 2025: LLM CHESS: Benchmarking Reasoning and Instruction-Following in LLMs through Chess
- Doerschuk-Tiberi et al. 2026: Game Arena: Strategic LLM Evaluation in Competitive Environments
- Lin et al. 2026: How Far Are LLMs from Professional Poker Players? Revisiting Game-Theoretic Reasoning with Agentic Tool Use (ICLR 2026)
- Duffy et al. 2025: Democratizing Diplomacy: A Harness for Evaluating Any Large Language Model on Full-Press Diplomacy
- Liang et al. 2025: HardcoreLogic: Challenging Large Reasoning Models with Long-tail Logic Puzzle Games