Conference Tutorials

Tutorial materials from major conferences. These provide structured introductions to NNV’s verification capabilities, suitable for workshops and courses.

AAAI 2026 – VNN-COMP Lab

“The Verification of Neural Networks Competition (VNN-COMP): A Lab for Benchmark Proposers, Verification Tool Participants, and the Broader AI Community” (Lab LH03)

Presented on March 4, 2026 at AAAI-26 in Philadelphia, PA, this half-day lab introduces participants to neural network verification with hands-on interactive demos. NNV contributors Taylor T. Johnson and Ben Wooding are among the organizers. You will learn:

  • Introduction to neural network verification and its motivation

  • VNN-COMP history, rules, and VNN-COMP 2025 results

  • Benchmark formats: VNN-LIB 1.0/2.0 and ONNX

  • How to participate as a tool developer or benchmark proposer

  • Interactive demos via Google Colab notebooks

  • AWS-based evaluation infrastructure overview

  • Slides

  • VNN-LIB Standard

  • VNN-COMP 2026 (co-located with SAIV @ CAV/FLoC 2026, Lisbon, July 24–25)

SPIE 2025 – Medical Imaging Verification

“Robustness Verification of Medical Imaging Neural Networks”

A hands-on tutorial covering verification of medical image classifiers and segmentation networks. You will learn:

  • 2D classification robustness with OrganCMNIST

  • 3D volumetric classification with VolumeStar

  • Segmentation robustness under bias field, gamma correction, and noise

  • When to use conformal prediction for large segmentation models

  • Location: examples/Tutorial/SPIE/

  • See Medical Imaging for the worked examples

DSN 2024 – Dependable Systems

“Tutorial: Safe, Secure, and Trustworthy AI via Formal Verification of Neural Networks and Autonomous CPS with NNV”

This tutorial covers NNV’s core verification workflow with emphasis on safety-critical systems. You will learn:

  • Loading and converting neural networks (MATLAB and ONNX formats)

  • Defining input perturbation sets (Star, ImageStar)

  • Running exact and approximate reachability analysis

  • Verifying closed-loop control systems (NNCS)

  • Adversarial robustness analysis for image classifiers

  • Paper: DOI 10.1109/DSN-S60304.2024.00027

EMSOFT 2023 – Embedded Software

“Tutorial: Neural Network and Autonomous CPS Formal Verification for Trustworthy AI and Safe Autonomy”

An introduction to NNV focusing on embedded and cyber-physical systems. You will learn:

  • Star set reachability fundamentals

  • FFNN and CNN verification

  • Neural network control system verification (ACC benchmark)

  • Interpreting verification results for safety-critical embedded systems

  • Paper: DOI 10.1145/3607890.3608454

  • Location: examples/Tutorial/

Interactive Exercises

The tutorial folder includes hands-on exercises with starter code:

  • ACAS Xu: Tutorial/NN/ACAS Xu/exercise_vnnlib_onnx.m

  • Robustness verification: Tutorial/NN/GTSRB/exercise_verify_robustness.m

  • NNCS reachability: Tutorial/NNCS/ACC/Exercise/exercise_reachability_nncs.m

Note

The ACC training example requires the Simulink toolbox.

Running Tutorials Online

All tutorials can be run without local installation:

  • MATLAB Online: Try NNV on MATLAB Online (a MATLAB license may be required for some examples, but many run as guest). NNV shared folder – click “Open in MATLAB Online” → “Copy Folder” (copying may take 15 minutes to a couple of hours). After copying, run startup_nnv.m to configure paths.

  • CodeOcean: NNV CAV 2023 capsule (runs in browser, no installation needed)

All tutorial source code is available at examples/Tutorial/.