A graph adversarial learning toolbox based on PyTorch and DGL.

Last update: Jan 05, 2023

Related tags

Deep Learning GraphWar

Overview

GraphWar: Arms Race in Graph Adversarial Learning

NOTE: GraphWar is still in the early stages and the API will likely continue to change.

🚀 Installation

Please make sure you have installed PyTorch and Deep Graph Library(DGL).

# Comming soon
pip install -U graphwar

# Recommended
git clone https://github.com/EdisonLeeeee/GraphWar.git && cd GraphWar
pip install -e . --verbose

where -e means "editable" mode so you don't have to reinstall every time you make changes.

Get Started

Assume that you have a dgl.DGLgraph instance g that describes you dataset.

A simple targeted attack

from graphwar.attack.targeted import RandomAttack
attacker = RandomAttack(g)
attacker.attack(1, num_budgets=3) # attacking target node `1` with `3` edges 
attacked_g = attacker.g()
edge_flips = attacker.edge_flips()

A simple untargeted attack

from graphwar.attack.untargeted import RandomAttack
attacker = RandomAttack(g)
attacker.attack(num_budgets=0.05) # attacking the graph with 5% edges perturbations
attacked_g = attacker.g()
edge_flips = attacker.edge_flips()

Implementations

In detail, the following methods are currently implemented:

Attack

Targeted Attack

Methods	Venue
RandomAttack	A simple random method that chooses edges to flip randomly.
DICEAttack	Marcin Waniek et al. 📝 Hiding Individuals and Communities in a Social Network, Nature Human Behavior'16
Nettack	Daniel Zügner et al. 📝 Adversarial Attacks on Neural Networks for Graph Data, KDD'18
FGAttack	Jinyin Chen et al. 📝 Fast Gradient Attack on Network Embedding, arXiv'18 Jinyin Chen et al. 📝 Link Prediction Adversarial Attack Via Iterative Gradient Attack, IEEE Trans'20 Hanjun Dai et al. 📝 Adversarial Attack on Graph Structured Data, ICML'18
GFAttack	Heng Chang et al. 📝 A Restricted Black - box Adversarial Framework Towards Attacking Graph Embedding Models, AAAI'20
IGAttack	Huijun Wu et al. 📝 Adversarial Examples on Graph Data: Deep Insights into Attack and Defense, IJCAI'19
SGAttack	Jintang Li et al. 📝 Adversarial Attack on Large Scale Graph, TKDE'21

Untargeted Attack

Methods	Venue
RandomAttack	A simple random method that chooses edges to flip randomly
DICEAttack	Marcin Waniek et al. 📝 Hiding Individuals and Communities in a Social Network, Nature Human Behavior'16
FGAttack	Jinyin Chen et al. 📝 Fast Gradient Attack on Network Embedding, arXiv'18 Jinyin Chen et al. 📝 Link Prediction Adversarial Attack Via Iterative Gradient Attack, IEEE Trans'20 Hanjun Dai et al. 📝 Adversarial Attack on Graph Structured Data, ICML'18
Metattack	Daniel Zügner et al. 📝 Adversarial Attacks on Graph Neural Networks via Meta Learning, ICLR'19
PGD, MinmaxAttack	Kaidi Xu et al. 📝 Topology Attack and Defense for Graph Neural Networks: An Optimization Perspective, IJCAI'19

Defense

Model-Level

Methods	Venue
MedianGCN	Liang Chen et al. 📝 Understanding Structural Vulnerability in Graph Convolutional Networks, IJCAI'21
RobustGCN	Dingyuan Zhu et al. 📝 Robust Graph Convolutional Networks Against Adversarial Attacks, KDD'19

Data-Level

Methods	Venue
JaccardPurification	Huijun Wu et al. 📝 Adversarial Examples on Graph Data: Deep Insights into Attack and Defense, IJCAI'19

More details of literatures and the official codes can be found in Awesome Graph Adversarial Learning.

Comments

Benchmark Results of Attack Performance

Hi, thanks for sharing the awesome repo with us! I recently run the attack sample code but the resultpgd_attack.py and random_attack.py under examples/attack/untargeted, but the accuracies of both evasion and poison attack seem not to decrease.

I'm pretty confused by the attack results. For CV models, pgd attack easily decreases the accuracy to nearly random guesses, but the results of GreatX seem not to consent with CV models. Is it because the number of the perturbed edges is too small?

Here are the results of pgd_attack.py

Processing...
Done!
Training...
100/100 [==============================] - Total: 874.37ms - 8ms/step- loss: 0.0524 - acc: 0.996 - val_loss: 0.625 - val_acc: 0.815
Evaluating...
1/1 [==============================] - Total: 1.82ms - 1ms/step- loss: 0.597 - acc: 0.843
Before attack
 Objects in BunchDict:
╒═════════╤═══════════╕
│ Names   │   Objects │
╞═════════╪═══════════╡
│ loss    │  0.59718  │
├─────────┼───────────┤
│ acc     │  0.842555 │
╘═════════╧═══════════╛
PGD training...: 100%|███████████████████████████████████████████████████████████████████████| 200/200 [00:02<00:00, 69.74it/s]
Bernoulli sampling...: 100%|██████████████████████████████████████████████████████████████████| 20/20 [00:00<00:00, 804.86it/s]
Evaluating...
1/1 [==============================] - Total: 2.11ms - 2ms/step- loss: 0.603 - acc: 0.842
After evasion attack
 Objects in BunchDict:
╒═════════╤═══════════╕
│ Names   │   Objects │
╞═════════╪═══════════╡
│ loss    │  0.603293 │
├─────────┼───────────┤
│ acc     │  0.842052 │
╘═════════╧═══════════╛
Training...
100/100 [==============================] - Total: 535.83ms - 5ms/step- loss: 0.124 - acc: 0.976 - val_loss: 0.728 - val_acc: 0.779
Evaluating...
1/1 [==============================] - Total: 1.74ms - 1ms/step- loss: 0.766 - acc: 0.827
After poisoning attack
 Objects in BunchDict:
╒═════════╤═══════════╕
│ Names   │   Objects │
╞═════════╪═══════════╡
│ loss    │  0.76604  │
├─────────┼───────────┤
│ acc     │  0.826962 │
╘═════════╧═══════════╛

Here are the results of random_attack.py

Training...
100/100 [==============================] - Total: 600.92ms - 6ms/step- loss: 0.0615 - acc: 0.984 - val_loss: 0.626 - val_acc: 0.811
Evaluating...
1/1 [==============================] - Total: 1.93ms - 1ms/step- loss: 0.564 - acc: 0.832
Before attack
 Objects in BunchDict:
╒═════════╤═══════════╕
│ Names   │   Objects │
╞═════════╪═══════════╡
│ loss    │  0.564449 │
├─────────┼───────────┤
│ acc     │  0.832495 │
╘═════════╧═══════════╛
Peturbing graph...: 253it [00:00, 4588.44it/s]
Evaluating...
1/1 [==============================] - Total: 2.14ms - 2ms/step- loss: 0.585 - acc: 0.826
After evasion attack
 Objects in BunchDict:
╒═════════╤═══════════╕
│ Names   │   Objects │
╞═════════╪═══════════╡
│ loss    │  0.584646 │
├─────────┼───────────┤
│ acc     │  0.826459 │
╘═════════╧═══════════╛
Training...
100/100 [==============================] - Total: 530.04ms - 5ms/step- loss: 0.0767 - acc: 0.98 - val_loss: 0.574 - val_acc: 0.791
Evaluating...
1/1 [==============================] - Total: 1.77ms - 1ms/step- loss: 0.695 - acc: 0.813
After poisoning attack
 Objects in BunchDict:
╒═════════╤═══════════╕
│ Names   │   Objects │
╞═════════╪═══════════╡
│ loss    │  0.695349 │
├─────────┼───────────┤
│ acc     │  0.81338  │
╘═════════╧═══════════╛

opened by ziqi-zhang 2

SG Attack example cannot run as expected on cuda
Hello, I got some error when I run SG Attack's example code on cuda device:

Traceback (most recent call last): File "src/test.py", line 50, in <module> attacker.attack(target) File "/greatx/attack/targeted/sg_attack.py", line 212, in attack subgraph = self.get_subgraph(target, target_label, best_wrong_label) File "/greatx/attack/targeted/sg_attack.py", line 124, in get_subgraph self.label == best_wrong_label)[0].cpu().numpy() RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:3 and cpu!

I found self.label is on cuda device, but best_wrong_label is on cpu. https://github.com/EdisonLeeeee/GreatX/blob/73eac351fdae842dbd74967622bd0e573194c765/greatx/attack/targeted/sg_attack.py#L123-L124

I remove line94 .cpu(), everything is going well and no error report

https://github.com/EdisonLeeeee/GreatX/blob/73eac351fdae842dbd74967622bd0e573194c765/greatx/attack/targeted/sg_attack.py#L94-L96

I found there is a commit that adds .cpu end of line 94, so I dont know it's a bug or something else🤨
opened by beiyanpiki 1
problem with metattack

Thanks for this wonderful repo. However, when I run the metattack example ,the result is not promising Here is my result when attack Cora with metattack Training... 100/100 [====================] - Total: 520.68ms - 5ms/step- loss: 0.0713 - acc: 0.996 - val_loss: 0.574 - val_acc: 0.847 Evaluating... 1/1 [====================] - Total: 2.01ms - 2ms/step- loss: 0.522 - acc: 0.847 Before attack ╒═════════╤═══════════╕ │ Names │ Objects │ ╞═════════╪═══════════╡ │ loss │ 0.521524 │ ├─────────┼───────────┤ │ acc │ 0.846579 │ ╘═════════╧═══════════╛ Peturbing graph...: 100%|██████████| 253/253 [01:00<00:00, 4.17it/s]Evaluating... 1/1 [====================] - Total: 2.08ms - 2ms/step- loss: 0.528 - acc: 0.844 After evasion attack ╒═════════╤═══════════╕ │ Names │ Objects │ ╞═════════╪═══════════╡ │ loss │ 0.528431 │ ├─────────┼───────────┤ │ acc │ 0.844064 │ ╘═════════╧═══════════╛ Training... 32/100 [=====>..............] - ETA: 0s- loss: 0.212 - acc: 0.956 - val_loss: 0.634 - val_acc: 0.807 100/100 [====================] - Total: 407.58ms - 4ms/step- loss: 0.0601 - acc: 0.996 - val_loss: 0.704 - val_acc: 0.787 Evaluating... 1/1 [====================] - Total: 1.66ms - 1ms/step- loss: 0.711 - acc: 0.819 After poisoning attack ╒═════════╤═══════════╕ │ Names │ Objects │ ╞═════════╪═══════════╡ │ loss │ 0.710625 │ ├─────────┼───────────┤ │ acc │ 0.818913 │ ╘═════════╧═══════════╛

opened by shanzhiq 1
Fix a venue of FeaturePropagation

Rossi's FeaturePropagation paper was submitted to ICLR'21 but was rejected. So how about updating with arXiv? This paper has not yet been accepted by other conferences.
bug documentation

opened by jeongwhanchoi 1
Add ratio for `attacker.data()`, `attacker.edge_flips()`, and `attacker.feat_flips()`
This PR allows specifying an argument ratio for attacker.edge_flips(), and attacker.feat_flips(), which determines how many generated perturbations were used for further evaluation/visualization. Correspondingly, attacker.data() holds edge_ratio and feat_ratio for these two methods when constructing perturbed graph.

Example

attacker = ... attacker.reset() attacker.attack(...) # Case1: Only 50% of generated edge perturbations were used trainer.evaluate(attacker.data(edge_ratio=0.5), mask=...) # Case2: Only 50% of generated feature perturbations were used trainer.evaluate(attacker.data(feat_ratio=0.5), mask=...) # NOTE: both arguments can be used simultaneously
enhancement
opened by EdisonLeeeee 0

Releases(0.1.0)

0.1.0(Jun 9, 2022)
GraphWar 0.1.0 🎉

The first major release, built upon PyTorch and PyTorch Geometric (PyG).

About GraphWar

GraphWar is a graph adversarial learning toolbox based on PyTorch and PyTorch Geometric (PyG). It implements a wide range of adversarial attacks and defense methods focused on graph data. To facilitate the benchmark evaluation on graphs, we also provide a set of implementations on popular Graph Neural Networks (GNNs).

Usages

For more details, please refer to the documentation and examples.

How fast can we train and evaluate your own GNN?

Take GCN as an example:

from graphwar.nn.models import GCN from graphwar.training import Trainer from torch_geometric.datasets import Planetoid dataset = Planetoid(root='.', name='Cora') # Any PyG dataset is available! data = dataset[0] model = GCN(dataset.num_features, dataset.num_classes) trainer = Trainer(model, device='cuda:0') trainer.fit({'data': data, 'mask': data.train_mask}) trainer.evaluate({'data': data, 'mask': data.test_mask})

A simple targeted manipulation attack

from graphwar.attack.targeted import RandomAttack attacker = RandomAttack(data) attacker.attack(1, num_budgets=3) # attacking target node `1` with `3` edges attacked_data = attacker.data() edge_flips = attacker.edge_flips()

A simple untargeted (non-targeted) manipulation attack

from graphwar.attack.untargeted import RandomAttack attacker = RandomAttack(data) attacker.attack(num_budgets=0.05) # attacking the graph with 5% edges perturbations attacked_data = attacker.data() edge_flips = attacker.edge_flips()

We will continue to develop this project and introduce more state-of-the-art implementations of papers in the field of graph adversarial attacks and defenses.
Source code(tar.gz)
Source code(zip)
graphwar-0.1.0-py3-none-any.whl(155.84 KB)

Owner

Jintang Li

Ph.D. student @ Sun Yat-sen University (SYSU), China.

GitHub Repository

A modular, open and non-proprietary toolkit for core robotic functionalities by harnessing deep learning

A modular, open and non-proprietary toolkit for core robotic functionalities by harnessing deep learning Website • About • Installation • Using OpenDR

304 Dec 28, 2022

Code for Universal Semi-Supervised Semantic Segmentation models paper accepted in ICCV 2019

USSS_ICCV19 Code for Universal Semi Supervised Semantic Segmentation accepted to ICCV 2019. Full Paper available at https://arxiv.org/abs/1811.10323.

68 Nov 24, 2022

Semi-supevised Semantic Segmentation with High- and Low-level Consistency

Semi-supevised Semantic Segmentation with High- and Low-level Consistency This Pytorch repository contains the code for our work Semi-supervised Seman

123 Dec 30, 2022

PyTorch implementation of paper A Fast Knowledge Distillation Framework for Visual Recognition.

FKD: A Fast Knowledge Distillation Framework for Visual Recognition Official PyTorch implementation of paper A Fast Knowledge Distillation Framework f

129 Dec 24, 2022

KaziText is a tool for modelling common human errors.

KaziText KaziText is a tool for modelling common human errors. It estimates probabilities of individual error types (so called aspects) from grammatic

3 Nov 24, 2022

[NeurIPS'21] "AugMax: Adversarial Composition of Random Augmentations for Robust Training" by Haotao Wang, Chaowei Xiao, Jean Kossaifi, Zhiding Yu, Animashree Anandkumar, and Zhangyang Wang.

112 Nov 07, 2022

Unofficial pytorch implementation for Self-critical Sequence Training for Image Captioning. and others.

An Image Captioning codebase This is a codebase for image captioning research. It supports: Self critical training from Self-critical Sequence Trainin

906 Jan 03, 2023

Chainer Implementation of Semantic Segmentation using Adversarial Networks

Semantic Segmentation using Adversarial Networks Requirements Chainer (1.23.0) Differences Use of FCN-VGG16 instead of Dilated8 as Segmentor. Caution

99 Jun 28, 2022

[NeurIPS 2021]: Are Transformers More Robust Than CNNs? (Pytorch implementation & checkpoints)

Are Transformers More Robust Than CNNs? Pytorch implementation for NeurIPS 2021 Paper: Are Transformers More Robust Than CNNs? Our implementation is b

145 Dec 01, 2022

My solutions for Stanford University course CS224W: Machine Learning with Graphs Fall 2021 colabs (GNN, GAT, GraphSAGE, GCN)

machine-learning-with-graphs My solutions for Stanford University course CS224W: Machine Learning with Graphs Fall 2021 colabs Course materials can be

7 Dec 14, 2022

A graph adversarial learning toolbox based on PyTorch and DGL.

Related tags

Overview

GraphWar: Arms Race in Graph Adversarial Learning

🚀 Installation

Get Started

A simple targeted attack

A simple untargeted attack

Implementations

Attack

Targeted Attack

Untargeted Attack

Defense

Model-Level

Data-Level

Comments

Benchmark Results of Attack Performance

SG Attack example cannot run as expected on cuda

problem with metattack

Fix a venue of FeaturePropagation

Add ratio for `attacker.data()`, `attacker.edge_flips()`, and `attacker.feat_flips()`

Releases(0.1.0)

0.1.0(Jun 9, 2022)

GraphWar 0.1.0 🎉

About GraphWar

Usages

How fast can we train and evaluate your own GNN?

A simple targeted manipulation attack

A simple untargeted (non-targeted) manipulation attack

Owner

Jintang Li

A modular, open and non-proprietary toolkit for core robotic functionalities by harnessing deep learning

Code for Universal Semi-Supervised Semantic Segmentation models paper accepted in ICCV 2019

Semi-supevised Semantic Segmentation with High- and Low-level Consistency

PyTorch implementation of paper A Fast Knowledge Distillation Framework for Visual Recognition.

KaziText is a tool for modelling common human errors.

[NeurIPS'21] "AugMax: Adversarial Composition of Random Augmentations for Robust Training" by Haotao Wang, Chaowei Xiao, Jean Kossaifi, Zhiding Yu, Animashree Anandkumar, and Zhangyang Wang.

Unofficial pytorch implementation for Self-critical Sequence Training for Image Captioning. and others.

Chainer Implementation of Semantic Segmentation using Adversarial Networks

[NeurIPS 2021]: Are Transformers More Robust Than CNNs? (Pytorch implementation & checkpoints)

My solutions for Stanford University course CS224W: Machine Learning with Graphs Fall 2021 colabs (GNN, GAT, GraphSAGE, GCN)

Context Axial Reverse Attention Network for Small Medical Objects Segmentation

Sparse-dense operators implementation for Paddle

Official implementation for “Unsupervised Low-Light Image Enhancement via Histogram Equalization Prior”

KDD CUP 2020 Automatic Graph Representation Learning: 1st Place Solution

This is the code of NeurIPS'21 paper "Towards Enabling Meta-Learning from Target Models".

[NeurIPS 2020] Official repository for the project "Listening to Sound of Silence for Speech Denoising"

Code for BMVC2021 paper "Boundary Guided Context Aggregation for Semantic Segmentation"

ncnn is a high-performance neural network inference framework optimized for the mobile platform

Pytorch cuda extension of grid_sample1d

Repo for the ACMMM20 submission: "Personalized breath based biometric authentication with wearable multimodality".