Author ORCID Identifier

0000-0002-4628-3566

Date of Award

8-31-2026

Document Type

Open Access Thesis

Degree Name

Doctor of Philosophy (PhD)

Department

Physics, Applied

First Advisor

Rahul V. Kulkarni

Abstract

Reinforcement learning (RL), the study of optimal decision-making over long timescales in stochastic systems, has recently seen remarkable advances due in large part to the efforts of the deep learning community. RL has witnessed great success in solving problems in video games, robotics, biological control, and language modeling. However, a unified statistical mechanics framework to understand and develop the corresponding algorithms is lacking. To address this issue, we begin by showing that the reinforcement learning problem can be formulated and solved using the tools of statistical mechanics. Drawing on physical principles of free energy minimization and invariance, we address important problems in the field related to reward shaping, bounds, exploration, and curriculum learning. We showcase our new framework's ability to solve open problems in average-reward reinforcement learning and compare the new algorithms to existing approaches. Next, we discuss a recent mapping between RL and generative modeling: Generative Flow Networks (GFlowNets). GFlowNets aim to generate samples from intractable distributions, a universal problem in complex physical systems. Currently, this mapping is viewed as a mathematical trick, without a theoretical framework. By treating the GFlowNet as a physical system, we use tools from equilibrium and non-equilibrium statistical mechanics to unify existing objectives and derive new algorithms for the field. Given their broad applicability, we discuss the implications for RL and GFlowNet algorithms in high-dimensional applications. The work in this dissertation aims to strengthen the growing bridge between artificial intelligence and physics, providing a unifying framework for understanding and solving temporal optimization problems.

Comments

Free and open access to this work is made available to the UMass Boston community by ScholarWorks at UMass Boston. Those not on campus and those without a UMass Boston campus username and password may gain access to this work through Interlibrary Loan. If you have a UMass Boston campus username and password and would like to download this work from off-campus, click on the “Off-Campus Users” button.

Download

Share

COinS