Multi-Armed Bandits and Reinforcement Learning: Advancing Decision Making in E-Commerce and Beyond

in conjunction with KDD 2023

MARBLE: Multi-ARmed Bandits and Reinforcement LEarning

In this workshop, we aim to have in-depth discussions on the visions, challenges, and emerging directions of reinforcement learning (RL) and multi-armed bandits (MAB), as well as their applications in e-commerce and other areas. Reinforcement learning has seen unprecedented advances over the last decade, including the design of new algorithms and an improved understanding of RL theory and its boundaries. E-commerce, on the other hand, is one of the fastest growing domains in industry, providing numerous challenging problems for sequential decision making algorithms, including recommendation, advertising, personalization, pricing, forecasting, and supply chain optimization. While e-commerce has existed for more than 20 years, RL and MAB only recently began to influence its modeling and infrastructure.

Progress in RL and MAB methodology has led to remarkable empirical results in a number of applied domains, yet a gap exists between the core RL research community and these application-focused communities. Today, only the simplest multi-armed bandit methods are used, while more advanced RL techniques are rarely implemented for real-world problems. Such problems, including those in e-commerce, tend to have unique characteristics, including special constraints due to the business mechanisms, system/infra requirements, and customer impact considerations. For example, the exploration-exploitation trade-off is no longer a simple decision, because the frequency and magnitude of changes will inevitably impact customers’ trust and shopping experiences. Some of those constraints are fundamental and hard to circumvent, creating new challenges and opening up interesting directions in both theoretical development and real-world impact. We believe that increased knowledge-sharing between theoreticians, empirical researchers, and practitioners will help to refine and focus the trajectory of the field and benefit all communities.

This workshop aims to stimulate discussions, especially those at the boundaries between computer science, marketing science, operations research, statistics, and econometrics, and bringing together researchers from academia and frontline practitioners. Professors and students in universities, researchers from research labs and tech companies, applied scientists, and machine learning engineers from the industry are all potential audiences and participants. Not only will this workshop serve as a medium for in-depth discussion, it will help interested researchers outside of the field gain a high-level view of the current state-of-the-art and potential directions of multi-armed bandits and reinforcement learning.

News and Updates

April 2023: MarbleKDD submission website is open. Please see Call for Paper for more submission details and visit this link for your submission!
March 2023: MarbleKDD details added to the website.
May 2023: MarbleKDD deadline is extended to June 6