This page lists our software packages for the verification of systems using neural networks.
Neural System Verify is a software package for performing reachability analysis on neural agent-environment systems implemented by ReLU feed-forward neural networks. Further information can be found in the corresponding paper (published at KR 2018).
Click here to download the source code: NSVerify
Recurrent Neural System Verify is an extension of NSVerify to handle recurrent neural networks. Further information can be found in the corresponding paper (published at AAAI 2019).
Click here to download the source code: RNSVerify
Verification of neural Multi-Agent Systems is a software package for performing MILP-based verification for agent-environment systems where the agent is controlled by ReLU-activated feed-forward neural networks, the environment is non-deterministic with branching temporal evolutions, and specifications are expressed in a bounded variant of CTL. This toolkit was previously known as NANESVerify. Further information can be found in the corresponding paper (published at AAMAS 2020).
Click here to download the source code: VenMAS-ND
An extension supporting MILP-based verification for multi-agent systems where the perception system of each agent is controlled by neural networks. Further information can be found in the corresponding paper (published at KR 2020).