Towards Global Neural Network Abstractions with Locally-Exact Reconstruction
This addresses the challenge of explaining and certifying neural network behavior for safety-critical applications, representing a novel advancement beyond existing slack or local methods.
The paper tackles the problem of neural network interpretability and safety by proposing a global abstraction technique that provides sound over-approximation bounds over the entire input domain while ensuring exact reconstructions for local inputs, achieving orders of magnitude tighter bounds than state-of-the-art global methods.
Neural networks are a powerful class of non-linear functions. However, their black-box nature makes it difficult to explain their behaviour and certify their safety. Abstraction techniques address this challenge by transforming the neural network into a simpler, over-approximated function. Unfortunately, existing abstraction techniques are slack, which limits their applicability to small local regions of the input domain. In this paper, we propose Global Interval Neural Network Abstractions with Center-Exact Reconstruction (GINNACER). Our novel abstraction technique produces sound over-approximation bounds over the whole input domain while guaranteeing exact reconstructions for any given local input. Our experiments show that GINNACER is several orders of magnitude tighter than state-of-the-art global abstraction techniques, while being competitive with local ones.