IBCB: Efficient Inverse Batched Contextual Bandit for Behavioral Evolution History
Yi Xu, Weiran Shen, Xiao Zhang, Jun Xu
TL;DR
This work tackles learning from expert behavior that evolves over time when rewards are inaccessible in batched contextual bandits. It introduces IBCB, an inverse batched contextual bandit framework that recasts the problem as a quadratic program solvable by OSQP, unifying deterministic and randomized policies and enabling simultaneous recovery of policy and reward parameters from evolution data. The approach delivers significantly faster training and improved generalization, including robustness to out-of-distribution and contradictory data, and accommodates fairness-aware extensions. Empirical results on synthetic data and the MovieLens ML-100K dataset show strong performance gains over baselines across online and batch testing scenarios, underscoring the practical impact for streaming decision systems and recommender settings.
Abstract
Traditional imitation learning focuses on modeling the behavioral mechanisms of experts, which requires a large amount of interaction history generated by some fixed expert. However, in many streaming applications, such as streaming recommender systems, online decision-makers typically engage in online learning during the decision-making process, meaning that the interaction history generated by online decision-makers includes their behavioral evolution from novice expert to experienced expert. This poses a new challenge for existing imitation learning approaches that can only utilize data from experienced experts. To address this issue, this paper proposes an inverse batched contextual bandit (IBCB) framework that can efficiently perform estimations of environment reward parameters and learned policy based on the expert's behavioral evolution history. Specifically, IBCB formulates the inverse problem into a simple quadratic programming problem by utilizing the behavioral evolution history of the batched contextual bandit with inaccessible rewards. We demonstrate that IBCB is a unified framework for both deterministic and randomized bandit policies. The experimental results indicate that IBCB outperforms several existing imitation learning algorithms on synthetic and real-world data and significantly reduces running time. Additionally, empirical analyses reveal that IBCB exhibits better out-of-distribution generalization and is highly effective in learning the bandit policy from the interaction history of novice experts.