Deep Neural Network Verification Toolbox

Tools for verification and analysis of deep neural networks. Watch our demo video for a quick description!

Getting Started

For more detailed instructions, see our documentation.

Installation

Clone this network:

$ git clone https://github.com/dlshriver/DNNV.git

Create a python virtual environment for this project:

$ ./manage.sh init

To activate the virtual environment and set environment variables correctly for tools installed using the provided manage.sh script, run:

$ . .env.d/openenv.sh

Install any of the supported verifiers (Reluplex, planet, MIPVerify.jl, Neurify, ERAN):

$ ./manage.sh install reluplex planet mipverify neurify eran

Make sure that the project environment is activated when installing verifiers with the manage.sh script. Otherwise some tools may not install correctly.

Additionally, several verifiers make use of the Gurobi solver. This should be installed automatically, but requires a license to be manually activated and available on the host machine. Academic licenses can be obtained for free from the Gurobi website.

Finally, planet has several additional requirements that currently must be installed separately before installation with ./manage.sh: libglpk-dev, qt5-qmake, valgrind, libltdl-dev, protobuf-compiler.

Usage

Properties are specified in our property DSL, extended from Python. A property specification can import python modules, and define variables. The only required component is the property expression, which must appear at the end of the file. An example of a local robustness property is shown below.

from dnnv.properties import *

N = Network("N")
x = Image("path/to/image")
epsilon = Parameter("epsilon", float, default=1.0)

Forall(
    x_,
    Implies(
        ((x - epsilon) < x_ < (x + epsilon)),
        argmax(N(x_)) == argmax(N(x))),
    ),
)

To check whether property holds for some network using the ERAN verifier, run:

$ python -m dnnv network.onnx property.prop --eran

Additionally, if the property defines parameters, using the Parameter keyword, they can be specified on the command line using the option --prop.PARAMETER_NAME, where PARAMETER_NAME is the name of the parameter. For the property defined above, a value for epsilon can be provided with a command line option as follows:

$ python -m dnnv network.onnx property.prop --eran --prop.epsilon=2.0

A set of example networks and properties that can be run with DNNV are available here.

Acknowledgements

This material is based in part upon work supported by the National Science Foundation under grant number 1900676.