DualOPSD: Adaptive Privileged Teachers for On-Policy Self-Distillation

Yutong Chen, Guangfu Guo, Zhichao Xu, Kunpeng Liu

arXiv:2608.26019

DualOPSD turns a fixed privileged teacher into an adaptive teaching policy. The student first learns from privileged supervision; the teacher then moves toward the updated student distribution on the same trajectory, improving future supervision without another rollout.

On Qwen3-8B, DualOPSD improves avg@12 over OPSD by 23.61, 13.89, and 10.00 points on AIME 2024, AIME 2025, and HMMT 2025, respectively.

CoScale-RL: Efficient Post-Training by Co-Scaling Data and Computation

Yutong Chen, Jiandong Gao, Ji Wu

arXiv:2601.14695

CoScale-RL improves the data and computational efficiency of reinforcement learning by co-scaling solution diversity and rollout computation, then using re-distillation to preserve efficiency.

Investigating the Effectiveness of Small-Scale Fine-Tuning in R1-Style Reinforcement Learning

Yutong Chen, Jiandong Gao, Ji Wu

arXiv:2505.17988

An analytical and empirical study of sample efficiency in supervised fine-tuning and reinforcement learning, introducing re-distillation as an efficient bridge between the two.

Expressive Speech-Driven Facial Animation with Controllable Emotions

Yutong Chen, Junhong Zhao, Wei-Qiang Zhang

2023 IEEE International Conference on Multimedia and Expo Workshops (ICMEW)

A deep learning approach that generates expressive speech-driven facial animation while enabling continuous control over emotion type and intensity.