Certified Robustness of Graph Convolution Networks for Graph Classification under Topological Attacks

Hongwei Jin, University of Illinois at Chicago
Computing Abstraction

Description: Graph convolutional networks (GCNs) have become effective models for graph classification. Like many deep networks, GCNs are vulnerable to adversarial attacks on graph topology and node attributes. Recently, a few effective attack and defense algorithms have been developed, but certificates of robustness against topological perturbations are currently available only for PageRank and label/feature propagation, while none has been designed for GCNs. In this talk, Hongwei will present the first algorithm for certifying the robustness of GCNs to topological attacks in the application of graph classification. Their method is based on Lagrange dualization and convex envelope, which result in tight approximation bounds that are efficiently computable by dynamic programming.  To handle the issue of permutation invariance on graphs, Hongwei will also briefly introduce a tractable variant of Gromov-Wasserstein distance for comparing graphs.

Join on your computer or mobile app

Click here to join the meeting

Or call in (audio only)

+1 630-556-7958,,358758746#   United States, Big Rock

Phone Conference ID: 358 758 746#