PV1 Verisigperforms reachability analysis on closed-loop systems with neural network controllers

Tool for verifying safety properties of closed-loop systems with NN controllers.

Application domain/field

Internals

They combine ideas from neural network reachability with ideas from hybrid system verification. They approximate neural networks with Taylor models. Moreover, they note that the reachability computation can be parallelised since each neuron in a layer can be analysed independently. Uses Flow*

Comments

The CAV'21 paper refers to a Verisig 2.0 License: GPL v3.0
Neural network

Links

Related papers

Last publication date

15 July 2021

Related tools

Compared to in CAV '21 paper: NNV and ReachNN* (and an older version of Verisig)

ProVerB specific



ProVerB is a part of SLEBoK. Last updated: February 2023.