no code implementations • 9 Mar 2023 • Chengyang Ying, Zhongkai Hao, Xinning Zhou, Hang Su, Songming Liu, Dong Yan, Jun Zhu
Extensive experiments in both image-based and state-based tasks show that TAD can significantly improve the performance of handling different tasks simultaneously, especially for those with high TDR, and display a strong generalization ability to unseen tasks.
1 code implementation • 15 Sep 2022 • Chengyang Ying, Zhongkai Hao, Xinning Zhou, Hang Su, Dong Yan, Jun Zhu
In this paper, we reveal that the instability is also related to a new notion of Reuse Bias of IS -- the bias in off-policy evaluation caused by the reuse of the replay buffer for evaluation and optimization.
1 code implementation • 12 Jun 2022 • Chengyang Ying, You Qiaoben, Xinning Zhou, Hang Su, Wenbo Ding, Jianyong Ai
Among different adversarial noises, universal adversarial perturbations (UAP), i. e., a constant image-agnostic perturbation applied on every input frame of the agent, play a critical role in Embodied Vision Navigation since they are computation-efficient and application-practical during the attack.
1 code implementation • 9 Jun 2022 • Chengyang Ying, Xinning Zhou, Hang Su, Dong Yan, Ning Chen, Jun Zhu
Though deep reinforcement learning (DRL) has obtained substantial success, it may encounter catastrophic failures due to the intrinsic uncertainty of both transition and observation.
no code implementations • 30 Jun 2021 • You Qiaoben, Chengyang Ying, Xinning Zhou, Hang Su, Jun Zhu, Bo Zhang
In this paper, we provide a framework to better understand the existing methods by reformulating the problem of adversarial attacks on reinforcement learning in the function space.
no code implementations • ICML Workshop AML 2021 • Chengyang Ying, Xinning Zhou, Dong Yan, Jun Zhu
Though deep reinforcement learning (DRL) has obtained substantial success, it may encounter catastrophic failures due to the intrinsic uncertainty caused by stochastic policies and environment variability.
no code implementations • ICML Workshop AML 2021 • You Qiaoben, Xinning Zhou, Chengyang Ying, Jun Zhu
Deep reinforcement learning (DRL) policies are vulnerable to the adversarial attack on their observations, which may mislead real-world RL agents to catastrophic failures.