Hindsight Value Function for Variance Reduction in Stochastic Dynamic Environment

Guo, Jiaming; Zhang, Rui; Zhang, Xishan; Peng, Shaohui; Yi, Qi; Du, Zidong; Hu, Xing; Guo, Qi; Chen, Yunji

doi:10.24963/ijcai.2021/341

Hindsight Value Function for Variance Reduction in Stochastic Dynamic Environment

Jiaming Guo, Rui Zhang, Xishan Zhang, Shaohui Peng, Qi Yi, Zidong Du, Xing Hu, Qi Guo, Yunji Chen

Download BibTeX Cite

Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence. Main Track. Pages 2476-2482.

https://doi.org/10.24963/ijcai.2021/341

Keywords

Machine Learning: Deep Learning
Machine Learning: Deep Reinforcement Learning
Uncertainty in AI: Sequential Decision Making

Abstract

Policy gradient methods are appealing in deep reinforcement learning but suffer from high variance of gradient estimate. To reduce the variance, the state value function is applied commonly. However, the effect of the state value function becomes limited in stochastic dynamic environments, where the unexpected state dynamics and rewards will increase the variance. In this paper, we propose to replace the state value function with a novel hindsight value function, which leverages the information from the future to reduce the variance of the gradient estimate for stochastic dynamic environments. Particularly, to obtain an ideally unbiased gradient estimate, we propose an information-theoretic approach, which optimizes the embeddings of the future to be independent of previous actions. In our experiments, we apply the proposed hindsight value function in stochastic dynamic environments, including discrete-action environments and continuous-action environments. Compared with the standard state value function, the proposed hindsight value function consistently reduces the variance, stabilizes the training, and improves the eventual policy.

Jiaming Guo, Rui Zhang, Xishan Zhang, Shaohui Peng, Qi Yi, Zidong Du, Xing Hu, Qi Guo, and Yunji Chen. 2021. Hindsight Value Function for Variance Reduction in Stochastic Dynamic Environment. In Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, IJCAI-21. IJCAI, 2476–2482. https://doi.org/10.24963/ijcai.2021/341

@inproceedings{ijcai2021p341,
  title     = {Hindsight Value Function for Variance Reduction in Stochastic Dynamic Environment},
  author    = {Guo, Jiaming and Zhang, Rui and Zhang, Xishan and Peng, Shaohui and Yi, Qi and Du, Zidong and Hu, Xing and Guo, Qi and Chen, Yunji},
  booktitle = {Proceedings of the Thirtieth International Joint Conference on
               Artificial Intelligence, {IJCAI-21}},
  publisher = {International Joint Conferences on Artificial Intelligence Organization},
  editor    = {Prof. Zhi-Hua Zhou},
  pages     = {2476--2482},
  year      = {2021},
  month     = {8},
  note      = {Main Track},
  doi       = {10.24963/ijcai.2021/341},
  url       = {https://doi.org/10.24963/ijcai.2021/341},
}

TY  - CPAPER
TI  - Hindsight Value Function for Variance Reduction in Stochastic Dynamic Environment
AU  - Guo, Jiaming
AU  - Zhang, Rui
AU  - Zhang, Xishan
AU  - Peng, Shaohui
AU  - Yi, Qi
AU  - Du, Zidong
AU  - Hu, Xing
AU  - Guo, Qi
AU  - Chen, Yunji
T2  - Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence, IJCAI-21
PB  - International Joint Conferences on Artificial Intelligence Organization
PY  - 2021
SP  - 2476
EP  - 2482
DO  - 10.24963/ijcai.2021/341
UR  - https://doi.org/10.24963/ijcai.2021/341
ER  -