Timezone: »

 
Poster
Beyond Rewards: a Hierarchical Perspective on Offline Multiagent Behavioral Analysis
Shayegan Omidshafiei · Andrei Kapishnikov · Yannick Assogba · Lucas Dixon · Been Kim

Tue Nov 29 02:00 PM -- 04:00 PM (PST) @ Hall J #520
in Poster Session 2 »

Each year, expert-level performance is attained in increasingly-complex multiagent domains, where notable examples include Go, Poker, and StarCraft II. This rapid progression is accompanied by a commensurate need to better understand how such agents attain this performance, to enable their safe deployment, identify limitations, and reveal potential means of improving them. In this paper we take a step back from performance-focused multiagent learning, and instead turn our attention towards agent behavior analysis. We introduce a model-agnostic method for discovery of behavior clusters in multiagent domains, using variational inference to learn a hierarchy of behaviors at the joint and local agent levels. Our framework makes no assumption about agents' underlying learning algorithms, does not require access to their latent states or policies, and is trained using only offline observational data. We illustrate the effectiveness of our method for enabling the coupled understanding of behaviors at the joint and local agent level, detection of behavior changepoints throughout training, discovery of core behavioral concepts, demonstrate the approach's scalability to a high-dimensional multiagent MuJoCo control domain, and also illustrate that the approach can disentangle previously-trained policies in OpenAI's hide-and-seek domain.

Keywords: Reinforcement Learning · interpretability · multiagent reinforcement learning · behavioral analysis · multiagent systems
[ Paper]  [ Poster]  [ OpenReview] 

Author Information

Shayegan Omidshafiei (Google)
Andrei Kapishnikov (Google)
Yannick Assogba (Google Research)
Lucas Dixon (Research, Google)
Lucas Dixon

I'm interested in visualization, understanding machine learning, and language as a boundary object for AI systems - for example natural language preferences for recommenders. My PhD and postdocs were at the University of Edinburgh, and then at Google I worked in Google and then founded the engineering and research efforts at Jigsaw. I've worked on a diverse range of topics including network security, foundations of mathematics and logic, graphical languages for quantum information, ML for NLP, and data visualization. I helped make uProxy/Outline, Project Shield, DigitalAttackMap, Syria Defection Tracker, and unfiltered.news. I led Jigsaw's early research into disinformation, and was the technical lead of the Conversation AI/Perspective efforts, and then cofounded Keen, an Area 120 project.

Been Kim (Google Brain)

More from the Same Authors