Algorithms for Verifying Deep Neural Networks
By Changliu Liu, Tomer Arnon, Christopher Lazarus, Clark W. Barrett, Mykel J. Kochenderfer. 2019
478 citations recorded by the source corpus.
Deep neural networks are widely used for nonlinear function approximation with applications ranging from computer vision to control. Although these networks involve the composition of simple arithmetic operations, it can be very challenging to verify whether a particular network satisfies certain input-output properties. This article surveys methods that have emerged recently for soundly verifying such properties. These methods borrow insights from reachability analysis, optimization, and search. We discuss fundamental differences and connections between existing algorithms. In addition, we provide pedagogical implementations of existing methods and compare them on a set of benchmark problems.
Subfields: Artificial Intelligence
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Alexey Ignatiev
- Joao Marques-Silva
- Nina Narodytska
- Charles Dawson
- Chuchu Fan
- Hongsheng Li
- Hao Shao
- Yu Liu
Citations in the corpus
- Safe Control With Learned Certificates: A Survey of Neural Lyapunov, Barrier, and Contraction Methods for Robotics and Control (2022, background): “It is possible that recent advances in neural network verification (see [87] for a relevant survey) or adversarial analysis and training of neural networks [88] may improve the scalability of verifier architectures; however, most approaches are currently limited to small neural networks.”
- On Relating Explanations and Adversarial Examples (2019, background): “For safety-critical applications such uncertainty might not be tolerable, therefore, a new trend is emerging focusing on methods with provable guarantees [25, 4, 33, 40, 10, 14, 26, 30].”
- Safety-Enhanced Autonomous Driving Using Interpretable Sensor Fusion Transformer (2022, background): “There are efforts taking a bypass, verifying the functioning conditions of neural network models instead of directly understanding them [9, 10, 11].”