Enhancing Context-Based Meta-Reinforcement Learning Algorithms via an Efficient Task Encoder (Student Abstract)
Abstract
Meta-Reinforcement Learning (meta-RL) algorithms enable agents to adapt to new tasks from small amounts of exploration, based on the experience of similar tasks. Recent studies have pointed out that a good representation of a task is key to the success of off-policy context-based meta-RL. Inspired by contrastive methods in unsupervised representation learning, we propose a new method to learn the task representation based on the mutual information between transition tuples in a trajectory and the task embedding. We also propose a new estimation for task similarity based on Q-function, which can be used to form a constraint on the distribution of the encoded task variables, making the task encoder encode the task variables more effective on new tasks. Experiments on meta-RL tasks show that the newly proposed method outperforms existing meta-RL algorithms.
Cite
Text
Xu et al. "Enhancing Context-Based Meta-Reinforcement Learning Algorithms via an Efficient Task Encoder (Student Abstract)." AAAI Conference on Artificial Intelligence, 2021. doi:10.1609/AAAI.V35I18.17965Markdown
[Xu et al. "Enhancing Context-Based Meta-Reinforcement Learning Algorithms via an Efficient Task Encoder (Student Abstract)." AAAI Conference on Artificial Intelligence, 2021.](https://mlanthology.org/aaai/2021/xu2021aaai-enhancing/) doi:10.1609/AAAI.V35I18.17965BibTeX
@inproceedings{xu2021aaai-enhancing,
title = {{Enhancing Context-Based Meta-Reinforcement Learning Algorithms via an Efficient Task Encoder (Student Abstract)}},
author = {Xu, Feng and Jiang, Shengyi and Yin, Hao and Zhang, Zongzhang and Yu, Yang and Li, Ming and Li, Dong and Liu, Wulong},
booktitle = {AAAI Conference on Artificial Intelligence},
year = {2021},
pages = {15937-15938},
doi = {10.1609/AAAI.V35I18.17965},
url = {https://mlanthology.org/aaai/2021/xu2021aaai-enhancing/}
}