Oğuz Kaan Yüksel
I am a fourth-year Ph.D candidate @ Theory of Machine Learning, EPFL advised by Nicolas Flammarion. My research focuses on understanding language models from first principles with a focus on mathematical modeling. Here are some keywords that spark my curiosity:
- learning-without-mixing
- non-i.i.d. learning theory of autoregressive processes
- in-context learning
- length generalization
- incremental learning
- saddle to saddle trajectories
- implicit bias
- Markovian structure
- mixing
- stability
- mixture processes
Short bio. Previously, M.Sc. in Data Science (minor in Mathematics) @ EPFL and B.Sc. in Computer Engineering & Mathematics @ Boğaziçi University. Worked as ML Engineer/Co-founder in NLP and CV applications, infrastructure and model life cycle maintenance. Full-stack web developer. Hobbyist game developer.
For more information, check my CV.
News
- Jul 12, 2025 : Presented a talk on Incremental Learning of Sparse Attention Patterns in Transformers at PriGM, EurIPS 2025.
- Jul 12, 2025 : Presented two posters Incremental Learning of Sparse Attention Patterns in Transformers and Generalization Bounds for Autoregressive Processes and In-Context Learning at PriGM, EurIPS 2025.
No matching items