End-to-end learning for decision optimization via constraint-enforcing approximators
Date:
We developed a novel neural network architecture that can learn to approximately solve optimization problems. We incorporate this network into an end-to-end framework which can effectively learn forecasts producing near-optimal decisions.