Nara Institute of Science and Technology — Reinforcement Learning & Language Modeling
@inproceedings{
kitamura2024a,
title={A Policy Gradient Primal-Dual Algorithm for Constrained {MDP}s with Uniform {PAC} Guarantees},
author={Toshinori Kitamura and Tadashi Kozuno and Masahiro Kato and YUKI ICHIHARA and Soichiro Nishimori and Akiyoshi Sannai and Sho Sonoda and Wataru Kumagai and Yutaka Matsuo},
booktitle={First Reinforcement Learning Safety Workshop},
year={2024},
url={https://openreview.net/forum?id=EWXKo2fj4h}
}
}
@article{ichihara2025evaluation,
title={Evaluation of Best-of-N Sampling Strategies for Language Model Alignment},
author={Yuki Ichihara and Yuu Jinnai and Tetsuro Morimura and Kenshi Abe and Kaito Ariu and Mitsuki Sakamoto and Eiji Uchibe},
journal={Transactions on Machine Learning Research},
issn={2835-8856},
year={2025},
url={https://openreview.net/forum?id=H4S4ETc8c9},
note={}
}
@inproceedings{ichihara-etal-2025-theoretical,
title = "Theoretical Guarantees for Minimum {B}ayes Risk Decoding",
author = "Ichihara, Yuki and
Jinnai, Yuu and
Ariu, Kaito and
Morimura, Tetsuro and
Uchibe, Eiji",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.acl-long.793/",
doi = "10.18653/v1/2025.acl-long.793",
pages = "16262--16284",
ISBN = "979-8-89176-251-0",
abstract = "Minimum Bayes Risk (MBR) decoding optimizes output selection by maximizing the expected utility value of an underlying human distribution. While prior work has shown the effectiveness of MBR decoding through empirical evaluation, few studies have analytically investigated why the method is effective. As a result of our analysis, we show that, given the size $n$ of the reference hypothesis set used in computation, MBR decoding approaches the optimal solution with high probability at a rate of $\mathcal{O}(n^{-\frac{1}{2}})$, under certain assumptions, even though the language space $\mathcal{Y}$ is significantly larger $|\mathcal{Y}| \gg n$.This result helps to theoretically explain the strong performance observed in several prior empirical studies on MBR decoding. In addition, we provide the performance gap for maximum-a-posteriori (MAP) decoding and compare it to MBR decoding. The result of this paper indicates that MBR decoding tends to converge to the optimal solution faster than MAP decoding in several cases."
}
@inproceedings{ichihara-etal-2025-awgrpo,
title = "Auto-Weighted Group Relative Preference Optimization for Multi-Objective Text Generation Tasks",
author = "Yuki, Ichihara and Yuu, Jinnai",
booktitle = "Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing: Industry Track",
year = "2025",
}
@misc{ichihara2025mogrpo,
title={MO-GRPO: Mitigating Reward Hacking of Group Relative Policy Optimization on Multi-Objective Problems},
author={Ichihara, Yuki and Jinnai, Yuu and Morimura, Tetsuro and Sakamoto, Mitsuki and Mitsuhashi, Ryota and Uchibe, Eiji},
year={2025},
eprint={2509.22047},
archivePrefix={arXiv},
url={https://doi.org/10.48550/arXiv.2509.22047}
}
My research centers on reinforcement learning for language models, and test-time compute (Best-of-N / Minimum Bayes Risk decoding). I aim to bridge theoretical guarantees with practical pipelines for real-world text generation tasks.