Distributed learning in games under bounded rationality

In multi-agent systems, bounded rationality arises both from the agents’ cognitive and informational limitations and from the uncertainty of the environments in which they interact. This dissertation considers models, solution concepts, and learning dynamics that explicitly embrace bounded rationality as a structural feature. Furthermore, we focus on the inherently distributed nature of these systems, where agents make decisions independently based on local information and individualized models. We study both cooperative and non-cooperative games under these dual perspectives of bounded rationality and distributed decision-making. In the cooperative setting, we show that globally stable and efficient outcomes can emerge from fully distributed dynamics. Focusing on (i) Transferable Utility (TU) coalitional games, (ii) TU B-matchings, and (iii) Non-Transferable Utility (NTU) B-matchings, we extend the classical core solution concept to each structure and design distributed dynamics in which agents rely only on local information and individual payoff aspirations. We prove convergence to the corresponding core, demonstrating that collective order can arise from simple, distributed decision-making. On the non-cooperative side, we advance the empirical evidence equilibrium (EEE) framework as a lens for boundedly rational learning. Within this framework we (i) connect single-agent empirical evidence models to population games, allowing us to leverage evolutionary game theory to prove convergence, (ii) derive explicit contraction and stability condition for weakly-coupled environments, and (iii) introduce exogenous signal games showing empirical evidence equilibria generalize Nash equilibria in perfect-monitoring games. To complement these theoretical contributions, we define independent softmax dynamics as a learning model and illustrate their behavior through simulations across the studied subclasses. Taken together, these results deepen our understanding of how distributed, boundedly rational agents, each acting on limited information and individualized models, can still generate stable and efficient collective outcomes. This perspective reframes uncertainty and bounded rationality as elements that can be constructively incorporated into the design of resilient and adaptive multi-agent systems.

Name thesis:degree_name

Ph.D.

Level thesis:degree_level

Dissertation

Discipline thesis:degree_discipline

Industrial Engineering

Grantor

University of Illinois Urbana-Champaign

Year dc:date

2025

Author dc:creator

• Hamed, Aya

Contributors dc:contributor

• Shamma, Jeff S.
• Etesami, Seyed R.
• Garg, Jugal
• Marden, Jason R.

• Game Theory
• Bounded Rationality
• Learning
• Distributed dynamics
• Cooperative games
• Coalitional games
• Matching
• Stochastic games

Statement dc:rights

• © 2025 Aya Hamed

Language dc:language

en

Handle dc:identifier

https://hdl.handle.net/2142/132744

OAI identifier oai:identifier

oai:www.ideals.illinois.edu:2142/132744