Semih Cayci — Personal Webpage

Department of Mathematics
RWTH Aachen University

Pontdriesch 14–16
52062 Aachen, Germany

I am an Assistant Professor for Mathematics of Machine Learning at RWTH Aachen University.

My research concerns the mathematical foundations of machine learning, with a focus on deep learning theory, reinforcement learning, and optimization.

Before joining RWTH in 2022, I was an NSF TRIPODS Postdoctoral Research Fellow at the University of Illinois at Urbana-Champaign. I received my PhD in Electrical and Computer Engineering from The Ohio State University in 2020.

CV (PDF) Email RWTH profile Google Scholar

Research themes

Geometry-aware optimization for deep learning · Reinforcement learning: MDPs and POMDPs · Geometry and optimization

selected publications

Geometric Convergence of Gauss–Newton for Neural Networks: Riemannian Geometry and Adaptive Damping

Semih Cayci

Schematic comparing Gauss–Newton and gradient flow in parameter space and on the prediction manifold

In Proceedings of the 43rd International Conference on Machine Learning, 2026

HTML Bib

@inproceedings{cayci2026geometric,
  author = {Cayci, Semih},
  title = {{Geometric Convergence of Gauss--Newton for Neural Networks: Riemannian Geometry and Adaptive Damping}},
  booktitle = {Proceedings of the 43rd International Conference on Machine Learning},
  series = {Proceedings of Machine Learning Research},
  volume = {306},
  pages = {12189--12214},
  year = {2026},
  publisher = {PMLR},
  url = {https://proceedings.mlr.press/v306/cayci26a.html},
}

Non-Asymptotic Generalization and Optimization Bounds for Stochastic Gauss-Newton in Deep Neural Networks

Semih Cayci

Schematic of stochastic Gauss–Newton trajectories in parameter space for training sets differing by one sample, with a larger-damping comparison inset

In Proceedings of the 29th International Conference on Artificial Intelligence and Statistics, 2026

HTML Bib

@inproceedings{cayci2026stochasticgaussnewton,
  author = {Cayci, Semih},
  title = {{Non-Asymptotic Generalization and Optimization Bounds for Stochastic Gauss-Newton in Deep Neural Networks}},
  booktitle = {Proceedings of the 29th International Conference on Artificial Intelligence and Statistics},
  series = {Proceedings of Machine Learning Research},
  volume = {300},
  pages = {1549--1557},
  year = {2026},
  publisher = {PMLR},
  url = {https://proceedings.mlr.press/v300/cayci26a.html},
}

Recurrent Natural Policy Gradient for POMDPs

Semih Cayci and Atilla Eryilmaz

Black-and-white schematic of recurrent policies for a POMDP, showing hidden states, noisy observations, rewards, recurrent memory, and action inputs across two time steps

Transactions on Machine Learning Research, 2025

HTML PDF Code Bib

@article{cayci2025recurrent,
  author = {Cayci, Semih and Eryilmaz, Atilla},
  title = {{Recurrent Natural Policy Gradient for POMDPs}},
  journal = {Transactions on Machine Learning Research},
  year = {2025},
  url = {https://openreview.net/forum?id=6G01e0vgIf},
}

Optimal Rates of Convergence for Entropy Regularization in Discounted Markov Decision Processes

Johannes Müller and Semih Cayci

Information and Inference: A Journal of the IMA, 2026

DOI Bib

@article{muller2026optimal,
  author = {M{\"u}ller, Johannes and Cayci, Semih},
  title = {{Optimal Rates of Convergence for Entropy Regularization in Discounted Markov Decision Processes}},
  journal = {Information and Inference: A Journal of the IMA},
  volume = {15},
  number = {1},
  pages = {iaaf034},
  year = {2026},
  doi = {10.1093/imaiai/iaaf034},
  url = {https://doi.org/10.1093/imaiai/iaaf034},
}

Convergence of Entropy-Regularized Natural Policy Gradient with Linear Function Approximation

Semih Cayci, Niao He, and R. Srikant

Policy simplex comparing policy gradient and natural policy gradient trajectories toward an entropy-regularized optimum

SIAM Journal on Optimization, 2024

DOI Bib

@article{cayci2024convergence,
  author = {Cayci, Semih and He, Niao and Srikant, R.},
  title = {{Convergence of Entropy-Regularized Natural Policy Gradient with Linear Function Approximation}},
  journal = {SIAM Journal on Optimization},
  volume = {34},
  number = {3},
  pages = {2729--2755},
  year = {2024},
  doi = {10.1137/22M1540156},
  url = {https://doi.org/10.1137/22M1540156},
}