Bethany R. Leffler

Rutgers, The State University of New Jersey, Massachusetts Institute of Technology, Rutgers Sexual and Reproductive Health and Rights

Papers

Total Citations

193

H-Index

About

Bethany R. Leffler is a leading researcher in reinforcement learning, with a focus on developing algorithms that enable agents to learn efficiently in complex, continuous, and stochastic environments. Her work addresses the fundamental challenge of balancing exploration and exploitation, particularly when an agent must generalize from limited experience. Leffler’s major contributions include the formalization of the adaptive *k*-meteorologists problem within the KWIK framework, providing provably efficient algorithms for structure learning and feature selection. She also pioneered the concept of relocatable action models, allowing agents to transfer knowledge across related states by representing transitions as state-independent outcomes. Her CORL algorithm (Continuous-state Offset-dynamics Reinforcement Learner) tackles the difficult problem of learning in domains with switching dynamics, such as robot navigation on varying terrain, and provides theoretical guarantees of performance. With over 190 citations across her most influential works, Leffler’s research has significantly advanced the theoretical foundations of model-based reinforcement learning, demonstrating how typed parametric models and latent structure can dramatically improve sample efficiency. Her work remains essential reading for anyone interested in provably efficient exploration and generalization in autonomous systems.