11/01/2023
By Hossein Haeri

The Miner School of Computer & Information Sciences, Department of Computer Science, invites you to attend a Master’s thesis defense by Hossein Haeri on “Reward Sharing Relational Networks: A Framework for Training Complex Multi-Agent Reinforcement Learning (MARL) Systems."

Candidate Name: Hossein Haeri
Degree: Master’s Science
Defense Date: Wednesday, Nov. 15, 2023
Time: 4 – 6 p.m. (EST)
Location: 208, Shah Hall, North Campus
Thesis/Dissertation Title: “Reward Sharing Relational Networks: A Framework for Training Complex Multi-Agent Reinforcement Learning (MARL) Systems"

Advisor: Reza Ahmadzadeh, Department of Computer Science, University of Massachusetts Lowell

Co-Advisor: Kshitij Jerath, Department of Mechanical & Industrial Engineering, University of Massachusetts Lowell

Committee Member: Maru Cabrera, Department of Computer Science, University of Massachusetts Lowell


Brief Abstract:
In this study, we incorporate 'social' dynamics into the MARL framework using a user-defined relational network, focusing on how agent relations impact emergent behaviors. Using principles from sociology and neuroscience, we introduce the Reward-Sharing Relational Networks (RSRN) to model agent relationships. Here, the weight of network edges represents the extent an agent values another's success. By formulating relational rewards based on RSRN interaction weights, we train a multi-agent system using a reinforcement learning algorithm. We assess the system's efficacy in a 3-agent setup across varied network structures, such as self-interested and communitarian. Our findings underscore the significant role of reward-sharing networks in shaping learned behaviors, suggesting that RSRN offers a versatile framework for understanding emergent behaviors in line with sociological perspectives.