Rare-event Simulation for Neural Network and Random Forest Predictors

We study rare-event simulation for a class of problems where the target hitting sets interest are defined via modern machine learning tools such as neural networks and random forests. This problem is motivated from fast emerging studies on safety evaluation intelligent systems, robustness quantification models, other potential applications to large-scale in which can be used approximate complex set boundaries. investigate an importance sampling scheme that integrates dominating point machinery large deviations sequential mixed integer programming locate underlying points. Our approach works range network architectures including fully connected layers, rectified linear units, normalization, pooling convolutional forests built standard decision trees. provide efficiency guarantees numerical demonstration our using classification model UCI Machine Learning Repository.