English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Recent research on vulnerabilities of deep reinforcement learning (RL) hasshown that adversarial policies adopted by an adversary agent can influence atarget RL agent (victim agent) to perform poorly in a multi-agent environment.In existing studies, adversarial policies are directly trained based onexperiences of interacting with the victim agent. There is a key shortcoming ofthis approach; knowledge derived from historical interactions may not beproperly generalized to unexplored policy regions of the victim agent, makingthe trained adversarial policy significantly less effective. In this work, wedesign a new effective adversarial policy learning algorithm that overcomesthis shortcoming. The core idea of our new algorithm is to create a newimitator to imitate the victim agent's policy while the adversarial policy willbe trained not only based on interactions with the victim agent but also basedon feedback from the imitator to forecast victim's intention. By doing so, wecan leverage the capability of imitation learning in well capturing underlyingcharacteristics of the victim policy only based on sample trajectories of thevictim. Our victim imitation learning model differs from prior models as theenvironment's dynamics are driven by adversary's policy and will keep changingduring the adversarial policy training. We provide a provable bound toguarantee a desired imitating policy when the adversary's policy becomesstable. We further strengthen our adversarial policy learning by making ourimitator a stronger version of the victim. Finally, our extensive experimentsusing four competitive MuJoCo game environments show that our proposedadversarial policy learning algorithm outperforms state-of-the-art algorithms.

Imitating Opponent to Win: Adversarial Policy Imitation Learning in Two-player Competitive Games