projects

Research

1. A Scalable Computational Framework for Activity-Based Dynamic Discrete Choice Models (ongoing)

   Master's Thesis · Hiroshima University · supervised by Prof. Makoto Chikaraishi

   Reachability-based state space pruning and GPU-accelerated backward induction for exact city-scale DDCM estimation. Applied to the Higashi-Hiroshima travel diary dataset.

   APTE 2026  ICMC 2026
2. Structural Inverse Reinforcement Learning for Tractable Welfare Measurement (ongoing)

   Long-term research agenda · JSPS DC1 application

   Identifies conditions under which RL and GNN connections to activity-based DDCM preserve the welfare guarantee, and builds an exact algorithm for spatial transport welfare analysis at city scale.