Ege Erdogan
PhD Student at the University of Amsterdam
Hi! I am a PhD student at the University of Amsterdam with Ana Lucic as part of the SIAS, CLC, and AI4Science groups.
My main research interest is mechanistic interpretability with a particular focus towards AI4Science. I am interested in two interrelated questions: how can interpretability of ML models facilitate scientific discovery, and how can scientific knowledge inform the design of interpretability methods?
I previously did my MSc at the Technical University of Munich and wrote my thesis on geometric generative models of neural network weights, advised by David Rügamer and Bastian Rieck.
Feel free to reach out at ege at erdogan.dev.
news
| Sep 2025 | 🎉 Paper accepted at the NeurIPS Mech Interp and UniReps workshops! Group Equivariance Meets Mechanistic Interpretability: Equivariant Sparse Autoencoders |
|---|---|
| May 2025 | 🎉 Paper accepted at FAccT 2025! Gender Bias in Explainability: Investigating Performance Disparity in Post-hoc Methods |
| May 2025 | 👨⚕️ Started my PhD at the University of Amsterdam! |
| Mar 2025 | 🎉 Paper accepted at the ICLR MLMP Workshop! FreeFlow: Latent Flow Matching for Free Energy Difference Estimation |
| Jan 2025 | 🎓 Completed my MSc at TUM! Read my thesis here |
selected publications
- Mech Interp ’25Group Equivariance Meets Mechanistic Interpretability: Equivariant Sparse AutoencodersNeurIPS Mechanistic Interpretability and UniReps Workshops 2025
- FAccT ’25Gender Bias in Explainability: Investigating Performance Disparity in Post-hoc MethodsACM Conference on Fairness, Accountability, and Transparency 2025
- MLMP ’25FreeFlow: Latent Flow Matching for Free Energy Difference EstimationICLR Workshop on Machine Learning Multiscale Processes 2025
- GLFrontiers ’23Poisoning × Evasion: Symbiotic Adversarial Robustness for Graph Neural NetworksNew Frontiers in Graph Learning Workshop (at NeurIPS ’23) 2023
