https://github.com/kautenja/robust-graph-convolutional-networks-against-adversarial-attacks-implementation

A Keras implementation of the paper "Robust Graph Convolutional Networks Against Adversarial Attacks"
https://github.com/kautenja/robust-graph-convolutional-networks-against-adversarial-attacks-implementation

adversarial-attacks gaussian-graph-convolution graph-convolutional-networks keras-layer

Last synced: 6 months ago
JSON representation

A Keras implementation of the paper "Robust Graph Convolutional Networks Against Adversarial Attacks"

• Host: GitHub
• URL: https://github.com/kautenja/robust-graph-convolutional-networks-against-adversarial-attacks-implementation
• Owner: Kautenja
• Created: 2019-11-08T19:16:47.000Z (almost 6 years ago)
• Default Branch: master
• Last Pushed: 2023-03-24T22:13:40.000Z (over 2 years ago)
• Last Synced: 2025-02-13T22:37:17.601Z (8 months ago)
• Topics: adversarial-attacks, gaussian-graph-convolution, graph-convolutional-networks, keras-layer
• Language: Jupyter Notebook
• Homepage:
• Size: 16.4 MB
• Stars: 2
• Watchers: 3
• Forks: 0
• Open Issues: 2
• Metadata Files:
  • Readme: README.md

Awesome Lists containing this project

README

          
# Robust Graph Convolutional Networks Against Adversarial Attacks (Implementation)

This repository implements the paper

[_Robust Graph Convolutional Networks Against Adversarial Attacks_](https://doi.org/10.1145/3292500.3330851).

The following attacks are implemented and tested:

- [x] Random attack: generate fake edges

- [ ] RL-S2V: adversarial attack based on reinforcement learning

- [ ] NETTACK: adversarial perturbations for direct and influence attacks

The following baselines are tested:

- [x] [Graph Convolutional Networks](https://arxiv.org/abs/1609.02907)

- [ ] [Graph Attention Networks](https://arxiv.org/abs/1710.10903)

## Usage

The [random-attack.ipynb](random-attack.ipynb) notebook contains code for

training and evaluating the GCN and RGCN models under a non-targeted random

attack. [random-attack-plot.ipynb](random-attack-plot.ipynb) contains logic to

generate the plot of noise ratio to accuracy between GCN and RGCN during a

non-targeted random attack.

## Results

### Random Attack

The plot below shows the results of the RGCN model compared to the GCN model

under a non-targeted random attack from a ratio of noise to clean edges from 0

to 1 in 0.1 increments. Each increment is generated as the mean value of test

accuracies over 10 training runs for each model.

![Random Attack Results](img/plot.png)

### RL-S2V Attack

TODO

### NETTACK

TODO

## Scalability Test

TODO