PRIMA: general and precise neural network certification via scalable convex hull approximations

A video of a short presentation of our work on PRIMA, a general and precise neural network certification framework leveraging scalable convex hull approximations. Formal verification of neural networks is critical for their safe adoption in real-world applications. However, designing a precise and scalable verifier remains an open and difficult challenge. In this paper, we take a major step forward in addressing this challenge and present a new verification framework, called Prima. Prima is both (i) general: it handles any non-linear activation function, and (ii) precise: it computes precise convex abstractions involving multiple neurons via novel convex hull approximation algorithms that leverage concepts from computational geometry. The algorithms have polynomial complexity, yield fewer constraints, and minimize precision loss.