Research Director, Fujitsu Limited | Project Researcher, RIKEN AIP
Statistical mechanics of learning · combinatorial optimization · large language model compression
Ph.D. from The University of Tokyo. Research Director at Fujitsu Limited and Project Researcher at the RIKEN Center for Advanced Intelligence Project (AIP). Research spans the theoretical analysis of machine learning via statistical mechanics and high-dimensional statistics, GPU-based combinatorial optimization, and large language model compression (quantization, pruning) and architecture design.
Publishes at NeurIPS, ICML, ICLR, ACL, and AISTATS, and serves as a reviewer for NeurIPS, ICLR, ICML, and AISTATS.
| Venue | Title |
|---|---|
| ACL 2026 (Oral) | PHOTON: Hierarchical Autoregressive Modeling for Lightspeed and Memory-Efficient Language Generation |
| NeurIPS 2025 | Quantization Error Propagation: Revisiting Layer-Wise Post-Training Quantization |
| ICLR 2025 | Optimization by Parallel Quasi-Quantum Annealing with Gradient-Based Sampling |
| NeurIPS 2024 | Controlling Continuous Relaxation for Combinatorial Optimization |
| AISTATS 2024 | Learning Dynamics in Linear VAE: Posterior Collapse Threshold, Superfluous Latent Space Pitfalls, and Speedup with KL Annealing |
Full list with abstracts and links: yuma-ichikawa.github.io/#publications
| Project | Description |
|---|---|
| OneComp | One-line post-training quantization library for LLMs (GPTQ, DBF, QEP, AutoBit) |
| QQA4CO | GPU toolkit for combinatorial and spin-glass optimization via quasi-quantum annealing (ICLR 2025) |
| StatPhysMLSimPlayground | Unified replica-method and online-learning-theory simulation package |