Yuki Ichihara
Starting in Aug 2026, I will join MBZUAI as a Ph.D. student, supervised by Junpei Komiyama.
My interests include Best-of-N and Minimum Bayes Risk decoding, language model alignment, and RL-based fine-tuning methods.
News
I will join MBZUAI as a Ph.D. student, supervised by Junpei Komiyama.
MO-GRPO was accepted to TACL.
Publications
- Reliable Chain-of-Thought via Prefix Consistency Paper Code
- CITE: Anytime-Valid Statistical Inference in LLM Self-Consistency Paper
- Consensus Group Relative Policy Optimization for Text Generation Paper Code
- MO-GRPO: Mitigating Reward Hacking of Group Relative Policy Optimization on Multi-Objective Problems Paper Code
- Auto-Weighted Group Relative Preference Optimization for Multi-Objective Text Generation Tasks Paper Code
- Theoretical Guarantees for Minimum Bayes Risk Decoding Paper
- Evaluation of Best-of-N Sampling Strategies for Language Model Alignment Paper
- A Policy Gradient Primal-Dual Algorithm for Constrained MDPs with Uniform PAC Guarantees Paper
Background
Education
MBZUAI, Ph.D. (incoming), supervised by Junpei Komiyama
Nara Institute of Science and Technology, Ph.D. program in Engineering (withdrew to re-enroll at MBZUAI)
Nara Institute of Science and Technology, M.S. in Engineering
Tokyo University of Science, B.S. in Electrical Engineering
Research Experience
JST SPRING Fellowship, Doctoral fellowship
CyberAgent AI Lab, Research Intern
ATR, Research Assistant
AIST, Research Assistant
Reviewing
Reviewer, ARR; ICML (nominated as Gold Reviewer)
Contact
For research discussions or collaboration, email is the best way to reach me.