Installation

NNV requires MATLAB 2023a or newer with several toolboxes. This guide covers installation on all platforms.

Prerequisites

MATLAB 2023a or newer with the following toolboxes:

Toolbox

Used For

Computer Vision

Image processing operations

Control Systems

Control system plant models

Deep Learning

Neural network layers and operations

Image Processing

Image manipulation and analysis

Optimization

LP solving (linprog)

Parallel Computing

Multi-core reachability analysis

Statistics and Machine Learning

Statistical operations

Symbolic Math

Symbolic computation

System Identification

System modeling utilities

Clone the Repository

NNV depends on several submodules (CORA, NNMT, HyST, onnx2nnv), so you must clone recursively:

git clone --recursive https://github.com/verivital/nnv.git

Ubuntu Quick Install

On Ubuntu, a one-step installation script is provided:

chmod +x install_ubuntu.sh
./install_ubuntu.sh

Then skip to Verify Installation.

Windows / macOS Installation

  1. Install MATLAB (2023a or newer) with the toolboxes listed above.

  2. Install the ONNX support package: Deep Learning Toolbox Converter for ONNX Model Format

    Note

    Support packages can be installed from MATLAB’s HOME tab > Add-Ons > Get Add-Ons, then search for the package name.

  3. Open MATLAB, navigate to the NNV root directory, and run:

    install

    Note

    If you restart MATLAB, rerun install.m or startup_nnv.m to restore paths. Alternatively, run savepath after installation (may require admin privileges).

Optional Packages

VGG16/VGG19 support (for CNN verification on these architectures):

MATLAB Verification Library (used in CAV 2023 comparison):

Framework converters (load models from TensorFlow/PyTorch):

Conformal Prediction (requires Python 3.12+):

# From the NNV root directory:
python -m venv .venv
# Windows: .venv\Scripts\activate
# macOS/Linux: source .venv/bin/activate
pip install -r requirement.txt

See Conformal Prediction Setup for detailed Python setup instructions.

Verify Installation

Run the diagnostic tool to check your setup:

check_nnv_setup()

This checks:

  • MATLAB version compatibility

  • Required toolbox availability (critical, important, and optional tiers)

  • Submodule paths (CORA, NNMT, HyST)

  • Python environment for conformal prediction (if configured)

You should see a Ready! confirmation when all critical dependencies are satisfied.

You can also check the NNV version programmatically:

NNVVERSION()  % Returns 'NNV v3.0.0'

Execution Without Installation

NNV can be executed online without installing MATLAB through CodeOcean:

Uninstallation

To remove NNV from your MATLAB path:

cd code/nnv
uninstall