paper-conference

Train Once, Get a Family: State-Adaptive Balances for Offline-to-Online Reinforcement Learning
Offline-to-online reinforcement learning (RL) is a training paradigm that combines pre-training on a pre-collected dataset with …
Shenzhi Wang, Qisen Yang, Jiawei Gao, Matthieu Gaetan Lin, Hao Chen, Liwei Wu, Ning Jia, Shiji Song, Gao Huang
PDF Cite Arxiv
Train Once, Get a Family: State-Adaptive Balances for Offline-to-Online Reinforcement Learning
Boosting Offline Reinforcement Learning with Action Preference Query
Compared to online fine-tuning, querying the preferences between pre-collected and learned actions can be equally or even more helpful to the erroneous estimate problem.
Qisen Yang, Shenzhi Wang, Matthieu Gaetan Lin, Shiji Song, Gao Huang
PDF Cite Arxiv
Boosting Offline Reinforcement Learning with Action Preference Query
Efficient Knowledge Distillation from Model Checkpoints
We observe that an intermediate model often serves as a better teacher compared to the fully converged model, although the former has much lower accuracy. This phenomenon can be partially explained by the information bottleneck principle.
Chaofei Wang, Qisen Yang, Rui Huang, Shiji Song, Gao Huang
PDF Cite Arxiv
Efficient Knowledge Distillation from Model Checkpoints