Doctoral Research Award. Scalability in Decision-Focused Learning: State of the Art, Challenges, and Beyond

Abstract

This presentation will explore recent advancements in Decision-Focused Learning (DFL), an emerging approach in artificial intelligence (AI) that integrates machine learning (ML) prediction with combinatorial optimization to train ML models for optimal decision-making. DFL predicts the unknown parameters of combinatorial optimization problems by focusing on the outcomes obtained using these predicted parameters. This presentation will start by providing an overview of various DFL techniques and introduce a taxonomy that categorizes these methods based on their distinct features. It will then highlight the scalability challenge, a major bottleneck for real-world DFL applications. The presentation will summarize existing strategies developed to address this issue and conclude by exploring potential future directions in the field