rEward Decomposition

Reward decomposition (RD) is a technique in reinforcement learning that breaks down complex reward signals into simpler, more interpretable components, aiming to improve both the efficiency of learning and the understandability of agent behavior. Current research focuses on applying RD to diverse problems, including optimizing recommender systems, aligning large language models, and enhancing the explainability of robotic agents, often employing reinforcement learning algorithms and integrating RD with other explanation methods like counterfactual analysis. This work is significant because it addresses the critical need for transparency and interpretability in complex AI systems, leading to more robust, efficient, and trustworthy AI applications across various domains.

Papers

September 20, 2024

RPAF: A Reinforcement Prediction-Allocation Framework for Cache Allocation in Large-Scale Recommender Systems
Shuo Su, Xiaoshuang Chen, Yao Wang, Yulin Wu, Ziqiang Zhang, Kaiqiao Zhan, Ben Wang, Kun Gai
Reinforcement Learning Modern Recommender System Cache Context Caching Strategy Large Scale Recommender System rEward Decomposition Adaptive Streaming

May 28, 2024

Hybrid Preference Optimization: Augmenting Direct Preference Optimization with Auxiliary Objectives
Anirudhan Badrinath, Prabhat Agarwal, Jiajing Xu
Large Language Model Language Model Direct Preference Optimization Preference Optimization Auxiliary Objective rEward Decomposition

December 30, 2023

Causal State Distillation for Explainable Reinforcement Learning
Wenhao Lu, Xufeng Zhao, Thilo Fryen, Jae Hee Lee, Mengdi Li, Sven Magg, Stefan Wermter
Reinforcement Learning Reward Function Causal Learning Explainable Reinforcement Learning rEward Decomposition Causal Feature Distillation

December 18, 2023

Explaining Reinforcement Learning Agents Through Counterfactual Action Outcomes
Yotam Amitai, Yael Septon, Ofra Amir
Reinforcement Learning Local Explanation Explainable Reinforcement Learning Counterfactual Feature rEward Decomposition

July 14, 2023

Global path preference and local response: A reward decomposition approach for network path choice analysis in the presence of locally perceived attributes
Yuki Oyama
Rich Attribute Pedestrian Behavior Choice Model Local Level rEward Decomposition

June 21, 2023

Inverse Constraint Learning and Generalization by Transferable Reward Decomposition
Jaehwi Jang, Minjae Song, Daehyung Park
Strong Generalization Task Transferability Constraint Learning Policy Constraint rEward Decomposition Adaptivity Constraint

April 25, 2023

A Closer Look at Reward Decomposition for High-Level Robotic Explanations
Wenhao Lu, Xufeng Zhao, Sven Magg, Martin Gromniak, Mengdi Li, Stefan Wermter
Reinforcement Learning Action Space Glance Annotation High Level Explainable Reinforcement Learning Robot Explanation rEward Decomposition

October 21, 2022

Integrating Policy Summaries with Reward Decomposition for Explaining Reinforcement Learning Agents
Yael Septon, Tobias Huber, Elisabeth André, Ofra Amir
Reinforcement Learning Reinforcement Learning Agent Based Policy Global Explanation rEward Decomposition

June 13, 2022

Provably Efficient Offline Reinforcement Learning with Trajectory-Wise Reward
Tengyu Xu, Yue Wang, Shaofeng Zou, Yingbin Liang
Offline Reinforcement Learning Proxy Reward Trajectory Wise Reward rEward Decomposition Reward Redistribution

rEward Decomposition

Papers

RPAF: A Reinforcement Prediction-Allocation Framework for Cache Allocation in Large-Scale Recommender Systems

Hybrid Preference Optimization: Augmenting Direct Preference Optimization with Auxiliary Objectives

Causal State Distillation for Explainable Reinforcement Learning

Explaining Reinforcement Learning Agents Through Counterfactual Action Outcomes

Global path preference and local response: A reward decomposition approach for network path choice analysis in the presence of locally perceived attributes

Inverse Constraint Learning and Generalization by Transferable Reward Decomposition

A Closer Look at Reward Decomposition for High-Level Robotic Explanations

Integrating Policy Summaries with Reward Decomposition for Explaining Reinforcement Learning Agents

Provably Efficient Offline Reinforcement Learning with Trajectory-Wise Reward