The increasing use of deep neural networks for safety-critical applications, such as autonomous driving and flight control, raises concerns about their safety and reliability. Formal verification can address these concerns by guaranteeing that a deep learning system operates as intended, but the state of the art is limited to small systems. In this work-in-progress report we give an overview of our work on mitigating this difficulty, by pursuing two complementary directions: devising scalable verification techniques, and identifying design choices that result in deep learning systems that are more amenable to verification.
@article{arxiv.1801.05950,
title = {Toward Scalable Verification for Safety-Critical Deep Networks},
author = {Lindsey Kuper and Guy Katz and Justin Gottschlich and Kyle Julian and Clark Barrett and Mykel Kochenderfer},
journal= {arXiv preprint arXiv:1801.05950},
year = {2018}
}