Projects

Adaptive Theory of Mind

My long-term research goal is to model how agents should adapt their ToM (depth and width) when interacting with other agents in an open world, open agent domain. Consider how humans use their mentalization - we form beliefs about others’ feelings and try to predict how our actions will affect them - but in some cases (like when buying a coffee), it makes no sense to “predict” the Barista’s next move - it’s both predictable and meaningless. We have a neural regulation mechanism that activates ToM where needed and only when needed - avoiding over or under usage.

I am very interested in how cognitive hierarchies influence agent behavior and decision-making processes in MARL environments. This involves studying how agents model and predict the actions of other agents and the impact of these models on overall system performance.

Deception in Multi-Agent Systems

In this context, a key area of my research is the emergence of deceptive strategies in multi-agent systems. I investigate how and why agents develop deceptive behaviors and the implications of these behaviors for the stability and efficiency of multi-agent interactions. In addition, my work also explains how anti-social behaviors arise from deceptive environments. Lastly, I propose a model-free solution to cope with deceptive opponents with bounded computational resources.

Relevant Papers:

• A (dis-) information theory of revealed and unrevealed preferences
• Between prudence and paranoia: Theory of Mind gone right, and wrong
• A (dis-) information Theory of Revealed and Unrevealed Preferences: Emerging Deception and Skepticism via Theory of Mind
• (Mal)adaptive Mentalizing in the Cognitive Hierarchy, and Its Link to Paranoia
• Aleph-IPOMDP: Mitigating Deception in a Cognitive Hierarchy with Off-Policy Counterfactual Anomaly Detection
• Theory-of-Mind in Partially Observed, Mixed-Motive Games