Implementation Details Of Reluplex: An Efficient SMT Solver for Verifying Deep Neural Networks

How to formally verify the bounds of your Neural Network

Reluplex is an algorithm submitted to CAV in 2017 by Stanford University [1]. Reluplex accepts as input a neural network and a set of constraints on the inputs and outputs of the network. The network must be composed of only linear layers and ReLU activations. The bounds may restrict any arbitrary number of input or output nodes to a single value or a range of …

algorithm constraints deep learning formal-verification implementation inputs neom network networks neural network neural networks set solver stanford stanford university thoughts-and-theory university verify