Ömer Faruk Akgül

I am a Ph.D. student in Computer Science at the University of Southern California, advised by Prof. Viktor Prasanna, and I also closely work with Prof. Willie Neiswanger. I received my B.Sc. in Computer Science from Bilkent University. Previously, I worked with Prof. Tudor Dumitras at the University of Maryland on machine learning for security.

My current research interests include:

Language model reasoning
Efficient and reliable machine learning
Post-training and inference-time methods for large language models
Structured reasoning over temporal and relational data

news

Sep 2026	Our paper LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoning was accepted to NeurIPS 2026.
Sep 2026	I was recognized as a NeurIPS 2026 Top Reviewer.
Aug 2026	Our paper SPARC-RAG: Adaptive Sequential-Parallel Scaling with Context Management for Retrieval-Augmented Generation was accepted to EMNLP 2026.
Jul 2026	Our paper Resa: Transparent Reasoning Models via SAEs was accepted to COLM 2026.
May 2026	I joined the AWS Agentic AI Foundational Research Team as an Applied Science Intern, working on AI agents.
May 2026	I was recognized as an ICML 2026 Gold Reviewer.
May 2026	We released Rethinking RL for LLM Reasoning: It’s Sparse Policy Selection, Not Capability Learning on arXiv.
Jan 2026	Our paper Tina: Tiny Reasoning Models via LoRA was accepted to ICLR 2026.
Jan 2026	Our paper RECIPE-TKG was accepted to EACL 2026.
Jun 2025	I joined Amazon as an Applied Science Intern, working on Text-to-SQL systems.
Aug 2024	Our paper Conformal Prediction for Federated Graph Neural Networks with Missing Neighbor Information was accepted to UAI 2025.
May 2023	Our paper An Efficient Distributed Graph Engine for Deep Learning on Graphs was accepted to the SC’23 Workshop.
Jun 2021	I joined UMD MC2 as a summer research intern.

publications

arXiv

Rethinking RL for LLM Reasoning: It’s Sparse Policy Selection, Not Capability Learning

Ömer Faruk Akgül, Rajgopal Kannan, Willie Neiswanger, and 1 more author

arXiv preprint arXiv:2605.06241, 2026

arXiv Bib Code

@article{akgul2026rethinking,
  title = {Rethinking RL for LLM Reasoning: It's Sparse Policy Selection, Not Capability Learning},
  author = {Akg{\"u}l, {\"O}mer Faruk and Kannan, Rajgopal and Neiswanger, Willie and Prasanna, Viktor},
  journal = {arXiv preprint arXiv:2605.06241},
  year = {2026},
  ranking = {01},
}

NeurIPS

LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoning

Ömer Faruk Akgül, Yusuf Hakan Kalaycı, Rajgopal Kannan, and 2 more authors

In Advances in Neural Information Processing Systems (NeurIPS), 2026

arXiv Bib Code

@inproceedings{akgul2026lynx,
  title = {LYNX: Learning Dynamic Exits for Confidence-Controlled Reasoning},
  author = {Akg{\"u}l, {\"O}mer Faruk and Kalayc{\i}, Yusuf Hakan and Kannan, Rajgopal and Neiswanger, Willie and Prasanna, Viktor},
  booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
  year = {2026},
  ranking = {02},
}

COLM

Resa: Transparent Reasoning Models via SAEs

Shangshang Wang, Julian Asilis, Ömer Faruk Akgül, and 4 more authors

In Conference on Language Modeling (COLM), 2026

arXiv Bib Code

@inproceedings{wang2026resa,
  title = {Resa: Transparent Reasoning Models via SAEs},
  author = {Wang, Shangshang and Asilis, Julian and Akg{\"u}l, {\"O}mer Faruk and Bilgin, Enes Burak and Liu, Ollie and Fu, Deqing and Neiswanger, Willie},
  booktitle = {Conference on Language Modeling (COLM)},
  year = {2026},
  ranking = {03},
}

ICLR

Tina: Tiny Reasoning Models via LoRA

Shangshang Wang, Julian Asilis, Ömer Faruk Akgül, and 3 more authors

In International Conference on Learning Representations (ICLR), 2026

arXiv Bib Code

@inproceedings{wang2026tina,
  title = {Tina: Tiny Reasoning Models via LoRA},
  author = {Wang, Shangshang and Asilis, Julian and Akg{\"u}l, {\"O}mer Faruk and Bilgin, Enes Burak and Liu, Ollie and Neiswanger, Willie},
  booktitle = {International Conference on Learning Representations (ICLR)},
  year = {2026},
  ranking = {04},
}

EACL

RECIPE-TKG: From Sparse History to Structured Reasoning for LLM-based Temporal Knowledge Graph Completion

Ömer Faruk Akgül, Feiyu Zhu, Yuxin Yang, and 2 more authors

In Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL), 2026

arXiv Bib Code

@inproceedings{akgul2026recipe,
  title = {RECIPE-TKG: From Sparse History to Structured Reasoning for LLM-based Temporal Knowledge Graph Completion},
  author = {Akg{\"u}l, {\"O}mer Faruk and Zhu, Feiyu and Yang, Yuxin and Kannan, Rajgopal and Prasanna, Viktor},
  booktitle = {Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL)},
  year = {2026},
  ranking = {05},
}

UAI

Conformal Prediction for Federated Graph Neural Networks with Missing Neighbor Information

Ömer Faruk Akgül, Rajgopal Kannan, and Viktor Prasanna

In Proceedings of the 41st Conference on Uncertainty in Artificial Intelligence (UAI), 2025

arXiv Bib HTML

@inproceedings{akgul2025conformal,
  title = {Conformal Prediction for Federated Graph Neural Networks with Missing Neighbor Information},
  author = {Akg{\"u}l, {\"O}mer Faruk and Kannan, Rajgopal and Prasanna, Viktor},
  booktitle = {Proceedings of the 41st Conference on Uncertainty in Artificial Intelligence (UAI)},
  year = {2025},
  ranking = {06},
}

EMNLP

SPARC-RAG: Adaptive Sequential-Parallel Scaling with Context Management for Retrieval-Augmented Generation

Yuxin Yang, Gangda Deng, Ömer Faruk Akgül, and 6 more authors

In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2026

arXiv Bib

@inproceedings{yang2026sparcrag,
  title = {SPARC-RAG: Adaptive Sequential-Parallel Scaling with Context Management for Retrieval-Augmented Generation},
  author = {Yang, Yuxin and Deng, Gangda and Akg{\"u}l, {\"O}mer Faruk and Chitsazan, Nima and Govilkar, Yash and Tigalappanavara, Akasha and Zhang, Shi-Xiong and Sahu, Sambit and Prasanna, Viktor},
  booktitle = {Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing (EMNLP)},
  year = {2026},
  ranking = {07},
}

Preprint

Schema-Free Text-to-SQL: Learning Enterprise Database Structure from Query Logs

Ömer Faruk Akgül, Nikhil Kapahi, and Prajit Muppidi

Preprint, 2026

Bib

@article{akgul2026schemafree,
  title = {Schema-Free Text-to-SQL: Learning Enterprise Database Structure from Query Logs},
  author = {Akg{\"u}l, {\"O}mer Faruk and Kapahi, Nikhil and Muppidi, Prajit},
  journal = {Preprint},
  year = {2026},
  ranking = {08},
}

An Efficient Distributed Graph Engine for Deep Learning on Graphs

Gangda Deng, Ömer Faruk Akgül, Hongkuan Zhou, and 4 more authors

In Proceedings of the SC’23 Workshops of The International Conference on High Performance Computing, Network, Storage, and Analysis, 2023

Bib HTML

@inproceedings{deng2023engine,
  title = {An Efficient Distributed Graph Engine for Deep Learning on Graphs},
  author = {Deng, Gangda and Akg{\"u}l, {\"O}mer Faruk and Zhou, Hongkuan and Zeng, Hanqing and Xia, Yinglong and Li, Jianbo and Prasanna, Viktor},
  booktitle = {Proceedings of the SC'23 Workshops of The International Conference on High Performance Computing, Network, Storage, and Analysis},
  pages = {922--931},
  year = {2023},
  doi = {10.1145/3624062.3624169},
  ranking = {09},
}

arXiv

Training Diverse Graph Experts for Ensembles: A Systematic Empirical Study

Gangda Deng, Yuxin Yang, Ömer Faruk Akgül, and 4 more authors

arXiv preprint arXiv:2510.18370, 2025

arXiv Bib Code

@article{deng2025diverse,
  title = {Training Diverse Graph Experts for Ensembles: A Systematic Empirical Study},
  author = {Deng, Gangda and Yang, Yuxin and Akg{\"u}l, {\"O}mer Faruk and Zeng, Hanqing and Xia, Yinglong and Kannan, Rajgopal and Prasanna, Viktor},
  journal = {arXiv preprint arXiv:2510.18370},
  year = {2025},
  ranking = {10},
}

professional service

Reviewer: ICML (Top Reviewer, 2026), NeurIPS (Top Reviewer, 2026), ICLR 2027, UAI 2026, and EMNLP 2026.

Program Committee: AAAI 2027.

other pursuits

Before moving to the United States to pursue my Ph.D., I competed as a member of the Turkish National Taekwondo Team. I won a gold medal at the Balkan Taekwondo Championships and was named the tournament's second-most valuable athlete.