IBCB: Efficient Inverse Batched Contextual Bandit for Behavioral Evolution History
This addresses a problem for streaming applications like recommender systems where traditional imitation learning fails to utilize evolving expert data, offering an incremental improvement with a novel method for a known bottleneck.
The paper tackles the challenge of imitation learning from behavioral evolution history, where data comes from experts evolving from novice to experienced, by proposing an inverse batched contextual bandit (IBCB) framework that efficiently estimates reward parameters and policies. Experimental results show IBCB outperforms existing imitation learning algorithms on synthetic and real-world data, significantly reducing running time and improving out-of-distribution generalization.
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.