Feat/gnnfingers implementation

[irtiza1999]

📦 Pull Request Template

Thank you for your contribution! Please complete the checklist and provide relevant details below to help us review your PR effectively.

---

📋 Summary

This PR integrates GNNFingers (A Fingerprinting Framework for Verifying Ownerships of Graph Neural Networks) into the PyGIP framework, providing comprehensive model ownership verification capabilities for Graph Neural Networks. The implementation has been reorganized into a dedicated examples/ folder for better project structure and easier review.

🧪 Related Issues

---

✅ Checklist

• My code follows the project's coding style
• I have tested the changes and verified that they work
• I have added necessary documentation (if applicable)
• I have linked related issues above (if any)
• The PR is made from a feature branch, not main

---

🧠 Additional Context (Optional)

Key Features Added

• GNNFingers Defense Framework: Complete integration of the fingerprinting-based ownership verification system
• Multi-Task Support: Node classification, graph classification, link prediction, and graph matching
• PyGIP Dataset Adapter: Seamless integration with existing PyGIP datasets (Cora, PROTEINS, etc.)
• Comprehensive Testing Suite: Full command-line interface with training, testing, and verification capabilities
• Model Piracy Detection: Production-ready single model verification system
• Automatic Fallbacks: Intelligent error handling and dataset/model adaptation

---

Major Components

• Core Defense Implementation: models/defense/gnn_fingers_defense.py
• GNN Model Architectures: models/defense/gnn_fingers_models.py
• Dataset Integration: datasets/gnn_fingers_datasets.py, datasets/gnnfingers_adapter.py
• Utility Functions: utils/gnn_fingers_utils.py
• Examples and Experiments: examples/ folder with comprehensive test scripts

---

Technical Details

• Fingerprint Construction: Graph-based fingerprints for each task type
• Universal Verification: Binary classifier (Univerifier) for ownership detection
• Joint Optimization: Alternating training algorithm for fingerprints and verifier
• Robust Attack Resistance: Handles fine-tuning, distillation, and obfuscation attacks

Performance Metrics

Task Type	Dataset	Model Architecture	AUC	Accuracy
Node Classification	Cora	GCN	0.88	0.90
Node Classification	Cora	Graphsage	0.89	0.85
Node Classification	Citeseer	GCN	0.97	0.90
Node Classification	Citeseer	Graphsage	0.86	0.85
Link Prediction	Cora	GCN	0.69	0.70
Link Prediction	Cora	Graphsage	0.87	0.85
Link Prediction	Citeseer	GCN	0.52	0.55
Link Prediction	Citeseer	Graphsage	0.61	0.75
Graph Matching	AIDS	GCNMean	0.40	0.50
Graph Matching	AIDS	GCNDiff	0.48	0.50
Graph Matching	AIDS	SimGNN	0.22	0.50
Graph Matching	PROTEINS	GCNMean	0.46	0.50
Graph Matching	PROTEINS	GCNDiff	0.42	0.50
Graph Matching	PROTEINS	SimGNN	0.44	0.50
Graph Classification	PROTEINS	GCNMean	1.00	1.00
Graph Classification	PROTEINS	GCNDiff	1.00	1.00
Graph Classification	PROTEINS	GraphsageMean	1.00	1.00
Graph Classification	PROTEINS	GraphsageDiff	1.00	1.00
Graph Classification	AIDS	GCNMean	1.00	0.95
Graph Classification	AIDS	GCNDiff	1.00	1.00
Graph Classification	AIDS	GraphsageMean	1.00	1.00
Graph Classification	AIDS	GraphsageDiff	1.00	1.00

---

🧪 Testing and Examples

Examples Folder Structure

examples/
├── run_gnnfingers_experiments.py    # Main experiment runner
├── adapter_demo.py                   # GNNFingers adapter demonstration
├── test_adapter.py                   # Adapter functionality testing
├── test_examples_setup.py           # Setup verification script
└── README.md                         # Comprehensive documentation

Core Functionality Tests

Setup Verification

# Test that examples folder is properly configured
python examples/test_examples_setup.py

Dataset Adapter Tests

# Test PyGIP dataset adaptation functionality
python examples/test_adapter.py

Adapter Demo

# Run the complete adapter demonstration
python examples/adapter_demo.py

Individual Experiment Commands

Quick experiment (32 fingerprints, 50 epochs) - recommended for testing

python examples/run_gnnfingers_experiments.py --task node_classification --dataset Cora --model GCN --quick

Full experiment (64 fingerprints, 100 epochs) - production quality

python examples/run_gnnfingers_experiments.py --task graph_classification --dataset PROTEINS --model GCNMean

Batch Operation Commands

Run all 22 experiments in quick mode (testing)

python examples/run_gnnfingers_experiments.py --all --quick

Run all 22 experiments in full mode (production)

python examples/run_gnnfingers_experiments.py --all --full

Specific Task Examples

Node Classification

python examples/run_gnnfingers_experiments.py --task node_classification --dataset Cora --model GCN --quick
python examples/run_gnnfingers_experiments.py --task node_classification --dataset Citeseer --model Graphsage --quick

Graph Classification

python examples/run_gnnfingers_experiments.py --task graph_classification --dataset PROTEINS --model GCNMean --quick
python examples/run_gnnfingers_experiments.py --task graph_classification --dataset AIDS --model GraphsageDiff --quick

Link Prediction

python examples/run_gnnfingers_experiments.py --task link_prediction --dataset Cora --model GCN --quick
python examples/run_gnnfingers_experiments.py --task link_prediction --dataset Citeseer --model Graphsage --quick

Graph Matching

python examples/run_gnnfingers_experiments.py --task graph_matching --dataset AIDS --model GCNMean --quick
python examples/run_gnnfingers_experiments.py --task graph_matching --dataset PROTEINS --model SimGNN --quick

---