David D. Yao

A Continuous-Time Reinforcement Learning Framework for Fine-Tuning Discrete Diffusion Models

Zikun Zhang, Jiayuan Sheng, David D. Yao, Wenpin Tang

We formulate reinforcement learning (RL) in continuous time with discrete state spaces and possibly arbitrary action spaces via a stochastic control approach, where the state dynamics are modeled as a controlled continuous-time Markov chain (CTMC). We consider policy optimization…Read more