Partition-Based Formulations for Mixed-Integer Optimization of Trained ReLU Neural Networks

Neural Information Processing Systems 

This paper introduces a class of mixed-integer formulations for trained ReLU neural networks.