Efficient Neural Network Validation with Affine Forms
Asma Soualah, Matthieu Martel · 2022
Affine forms, which are an extension of interval arithmetic, have been successfully used to assess the robustness of neural networks as well as to explain their decisions. They are well-suited to this application domain since they capture affine relations between variables and since neural networks mostly perform affine computations. However, a drawback is that affine forms are time and memory consuming, making the verification of industrial-size neural network difficult or impossible. In this article, we study the impact of noisy symbol merging on accuracy and time, in the context of neural network validation. Intuitively, we use less noisy symbols and therefore need less time and memory to analyze neural networks. This is confirmed by our experimental results carried out in a image recognition setting.