Please join RCB Borealis and CSAIL Alliances for a compelling talk! RBC Borealis will also share more information and answer questions on thier current research, career opportunities, early talent programs, and more.

​Many important problems—like planning routes, assigning resources, or selecting the most useful subset of items—require making discrete choices under constraints.

Although modern machine learning has achieved remarkable success, it still struggles with this type of decision-making problems. This limitation is not accidental: most ML models rely on smooth, differentiable functions, while combinatorial problems live in inherently discrete, constrained spaces.

In this talk, Akbar Rafiey will present a framework for bridging this gap by transforming hard combinatorial problems into continuous formulations that are compatible with gradient-based optimization, while still guaranteeing high-quality discrete solutions. He will describe two recent results: one for problems involving permutations, such as routing and matching, and another for problems involving constrained set selection.