Analyzing Neural Network Robustness Using Graph Curvature

TL;DR

This paper introduces a novel graph-theoretic approach using neural Ricci curvature to analyze and improve neural network robustness by identifying and minimizing bottleneck edges.

Contribution

It defines neural Ricci curvature and demonstrates its effectiveness in identifying bottleneck edges related to neural network robustness.

Findings

Bottleneck edges are more common in less robust inputs.

Neural Ricci curvature can identify critical edges affecting robustness.

Potential for robust training by reducing bottleneck edges.

Abstract

This paper presents a new look at the neural network (NN) robustness problem, from the point of view of graph theory analysis, specifically graph curvature. Graph curvature (e.g., Ricci curvature) has been used to analyze system dynamics and identify bottlenecks in many domains, including road traffic analysis and internet routing. We define the notion of neural Ricci curvature and use it to identify bottleneck NN edges that are heavily used to ``transport data" to the NN outputs. We provide an evaluation on MNIST that illustrates that such edges indeed occur more frequently for inputs where NNs are less robust. These results will serve as the basis for an alternative method of robust training, by minimizing the number of bottleneck edges.