Learning to Design Convex Relaxations for Nonconvex Optimization

Name: Learning to Design Convex Relaxations for Nonconvex Optimization
Uploaded: 2026-10-18T17:34:35.545584+00:00
Duration: 331 s
Description: AutoConvexRelax learns how to design convex relaxations for hard nonconvex optimization problems. It models the problem as a graph and trains a policy to select symbolic rewrites sequentially. This learned process finds tighter bounds than traditional methods by intelligently choosing which relaxation technique, like McCormick or SDP lifting, to apply at each step.

AutoConvexRelax learns how to design convex relaxations for hard nonconvex optimization problems. It models the problem as a graph and trains a policy to select symbolic rewrites sequentially. This learned process finds tighter bounds than traditional methods by intelligently choosing which relaxation technique, like McCormick or SDP lifting, to apply at each step.

Open the full episode