Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
110 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
44 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Topology Attack and Defense for Graph Neural Networks: An Optimization Perspective (1906.04214v3)

Published 10 Jun 2019 in cs.LG, cs.CR, cs.SI, and stat.ML

Abstract: Graph neural networks (GNNs) which apply the deep neural networks to graph data have achieved significant performance for the task of semi-supervised node classification. However, only few work has addressed the adversarial robustness of GNNs. In this paper, we first present a novel gradient-based attack method that facilitates the difficulty of tackling discrete graph data. When comparing to current adversarial attacks on GNNs, the results show that by only perturbing a small number of edge perturbations, including addition and deletion, our optimization-based attack can lead to a noticeable decrease in classification performance. Moreover, leveraging our gradient-based attack, we propose the first optimization-based adversarial training for GNNs. Our method yields higher robustness against both different gradient based and greedy attack methods without sacrificing classification accuracy on original graph.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (7)
  1. Kaidi Xu (85 papers)
  2. Hongge Chen (20 papers)
  3. Sijia Liu (204 papers)
  4. Pin-Yu Chen (311 papers)
  5. Tsui-Wei Weng (51 papers)
  6. Mingyi Hong (172 papers)
  7. Xue Lin (92 papers)
Citations (414)

Summary

  • The paper introduces an optimization-based method to craft topology attacks, significantly exposing vulnerabilities in GNN models.
  • It details a defensive strategy that integrates robust training with optimization techniques to mitigate adversarial manipulations.
  • Experimental results demonstrate improved resilience in GNNs under various attack scenarios, highlighting practical implications for real-world applications.

Overview of "IJCAI--19 Formatting Instructions"

The paper "IJCAI--19 Formatting Instructions" provides detailed guidelines for authors preparing manuscripts for the IJCAI-19 Proceedings. Authored by representatives from the Department of Computer Science, Bar-Ilan University, this document emphasizes adherence to specific formatting standards to ensure the uniformity of electronically submitted manuscripts in PDF format.

Summary

The paper outlines a meticulous framework for manuscript preparation, starting from basic requirements such as the maximum permissible length of a paper, which is limited to six pages plus one additional page for references. Key specifications include formatting dimensions for the document, the use of Times Roman font, and the layout of sections and headings.

Moreover, the document addresses specific requirements depending on the submission track, signposting authors to visit the conference’s website to confirm the exact length restrictions applicable to their submission. The paper meticulously details the spacing, font sizes, and alignment requirements for various parts of the manuscript, including the abstract, text, headings, and references.

Technical Details

Additional formatting instructions are provided for:

  • Word Processing Software: The availability of LaTeX macros and a Word template is highlighted, enabling the document to maintain consistency irrespective of the software used by authors.
  • Illustrations: Authors are encouraged to embed illustrations directly within the document where they are discussed, restricting them to black-and-white to ensure consistency in print and avoiding superimposing text on complex patterns.
  • Tables and Formulas: Recommendations include using the booktabs package in LaTeX for more readable tables and resizing equations to fit within the two-column format without compromising readability.
  • Citations and References: The paper details the preferred citation style, advocating clarity and consistency, crucial for scholarly referencing.

Implications and Future Directions

The implications of these formatting instructions are primarily towards ensuring quality and uniformity in conference proceedings, facilitating easier readability and comprehension across various topics presented at the conference. By standardizing manuscript preparation, the guidelines help streamline the publication process, enabling a smooth transition from manuscript submission to the production of the proceedings.

This document exemplifies the meticulous organizational structure necessary for leading conferences like IJCAI. It also provides a blueprint that future iterations must consider for evolving scholarly communication standards, potentially addressing adjustments for newer formatting technologies or digital dissemination strategies.

In conclusion, while the essay does not introduce novel intellectual content or claim significant theoretical contributions, its role in maintaining the rigor and clarity essential for high-profile academic conferences is undeniable. As the field moves forward, similar documents will likely continue adapting to integrate novel publishing paradigms without sacrificing the quality benchmarks established by prior editions.

X Twitter Logo Streamline Icon: https://streamlinehq.com