About
I am Daisuke Nohara, a first-year master’s student at the Institute of Science Tokyo (formerly Tokyo Institute of Technology), conducting research in the Rio Yokota Laboratory. My research focuses on reinforcement learning for improving reasoning in large language models, with particular interests in controlling reasoning behavior and understanding the optimization principles behind stable and efficient RL post-training.
I also enjoy building software systems as a hobby, from low-level Rust engines to self-hosted infrastructure.
Research Interests
My research centers on improving and analyzing reasoning in large language models through reinforcement learning. Specific topics I am currently exploring include:
- Reasoning length control in RL-trained LLMs — Investigating optimal output lengths that balance accuracy and computational cost during inference.
- Reasoning efficiency — Reducing the computational cost of chain-of-thought reasoning while preserving answer quality.
- Optimization principles for LLM reinforcement learning — Studying trust-region design, divergence regularization, and policy-update constraints for stable and efficient post-training.
Publications
First-Author
- On the Optimal Reasoning Length for RL-Trained Language Models
Daisuke Nohara, Taishi Nakamura, Rio Yokota. arXiv:2602.09591, 2026.
Accepted at AdaptFM: Resource-Adaptive Foundation Model Inference (ICML 2026 Workshop).
[arXiv] [OpenReview] [Workshop]
Co-Authored
- Optimal Sparsity of Mixture-of-Experts Language Models for Reasoning Tasks
Taishi Nakamura, Satoki Ishikawa, Masaki Kawamura, Takumi Okamoto, Daisuke Nohara, Jun Suzuki, Rio Yokota. arXiv:2508.18672, 2025.
Accepted at ICLR 2026 (Oral).
[arXiv] [OpenReview]
Technical Writing
- Post-Training a Japanese Speech Foundation Model: Applying Reinforcement Learning to TTS
DubGuild Tech Blog, July 2026.
[Article (Japanese)]
Experience
- Preferred Networks, Inc. — Summer Intern
- Aug 2026 – Present
- National Institute of Informatics (NII) — Research Assistant
- May 2026 – Present
- DubGuild — Research Intern
- Apr 2026 – Present
- Institute of Science Tokyo — Research Assistant
- Sep 2025 – Feb 2026
Education
- Institute of Science Tokyo — Master’s Program in Computer Science
- Apr 2026 – Present
- Institute of Science Tokyo — Department of Computer Science
- Apr 2022 – Mar 2026
Activities
Conference presentations and other research activities. View activities →