Sequential Counterfactual Risk Minimization
This work addresses the logged bandit feedback problem for machine learning practitioners, offering an incremental extension to CRM with potential benefits in sequential decision-making scenarios.
The paper tackles the problem of improving policies using offline logged bandit feedback by extending Counterfactual Risk Minimization (CRM) to allow multiple deployments and new data acquisition, resulting in improved performance in terms of excess risk and regret rates as demonstrated empirically in discrete and continuous action settings.
Counterfactual Risk Minimization (CRM) is a framework for dealing with the logged bandit feedback problem, where the goal is to improve a logging policy using offline data. In this paper, we explore the case where it is possible to deploy learned policies multiple times and acquire new data. We extend the CRM principle and its theory to this scenario, which we call "Sequential Counterfactual Risk Minimization (SCRM)." We introduce a novel counterfactual estimator and identify conditions that can improve the performance of CRM in terms of excess risk and regret rates, by using an analysis similar to restart strategies in accelerated optimization methods. We also provide an empirical evaluation of our method in both discrete and continuous action settings, and demonstrate the benefits of multiple deployments of CRM.