Abstract

"Neural network verification guarantees safety properties in autonomous and safety-critical real-time applications. In this paper, we propose a scalable automated verification framework capable of evaluating constraint bounds across non-linear activation layers in under 200ms..."

Submission Authors