Supervised Reinforcement Learning

Supervised reinforcement learning (SRL) combines the strengths of supervised and reinforcement learning paradigms to train agents that learn from both labeled data and rewards. Current research emphasizes improving model interpretability and robustness, particularly in partially observable environments, often using partially supervised approaches that leverage state estimators and incorporate both supervised semantic and unsupervised latent information. This focus addresses challenges like adversarial attacks and the need for efficient, safe, and explainable AI systems in applications ranging from robotics and autonomous driving to video analysis and violence detection. The resulting advancements are significant for improving the reliability and trustworthiness of AI agents in complex real-world scenarios.

Papers

February 14, 2024

Learning Interpretable Policies in Hindsight-Observable POMDPs through Partially Supervised Reinforcement Learning
Michael Lanier, Ying Xu, Nathan Jacobs, Chongjie Zhang, Yevgeniy Vorobeychik
Deep Reinforcement Learning Partially Observable Markov Decision Process Observable Environment Interpretable Policy Supervised Reinforcement Learning

December 28, 2022

Certifying Safety in Reinforcement Learning under Adversarial Perturbation Attacks
Junlin Wu, Hussein Sibai, Yevgeniy Vorobeychik
Reinforcement Learning Adversarial Perturbation Partially Observable Markov Decision Process Adversarial Environment Adversarial Machine Learning Safety Certificate Supervised Reinforcement Learning

September 3, 2022

Sharp bounds on the price of bandit feedback for several models of mistake-bounded online learning
Raymond Feng, Jesse Geneson, Andrew Lee, Espen Slettnes
Reinforcement Learning Full Model Bandit Feedback Curious Price Sharp Bound Supervised Reinforcement Learning Mistake Bound

March 29, 2022

Text-Driven Video Acceleration: A Weakly-Supervised Reinforcement Learning Method
Washington Ramos, Michel Silva, Edson Araujo, Victor Moura, Keller Oliveira, Leandro Soriano Marcolino, Erickson R. Nascimento
Weakly Supervised Text to Video Instructional Video Untrimmed Video Video Context Supervised Reinforcement Learning Document Attention Network

February 4, 2022

Video Violence Recognition and Localization Using a Semi-Supervised Hard Attention Model
Hamid Mohammadi, Ehsan Nazerfard
Semi Supervised Localization Focus Violence Detection Hard Attention Supervised Reinforcement Learning

Supervised Reinforcement Learning

Papers

Learning Interpretable Policies in Hindsight-Observable POMDPs through Partially Supervised Reinforcement Learning

Certifying Safety in Reinforcement Learning under Adversarial Perturbation Attacks

Sharp bounds on the price of bandit feedback for several models of mistake-bounded online learning

Text-Driven Video Acceleration: A Weakly-Supervised Reinforcement Learning Method

Video Violence Recognition and Localization Using a Semi-Supervised Hard Attention Model