Privacy-Preserving Graph-Based Machine Learning with Fully Homomorphic Encryption for Collaborative Anti-Money Laundering¶
Authors: Fabrianne Effendi; Anupam Chattopadhyay
Year: 2024
Venue: arXiv preprint
type: application collection: ppml status: auto
Links: arXiv:2411.02926
Summary¶
Auto-extracted summary (unreviewed)
Builds privacy-preserving pipelines for collaborative anti-money-laundering detection using TFHE through Zama Concrete ML: a GNN (GIN) pipeline made FHE-compatible via quantization and pruning, and a graph-feature XGBoost pipeline that uses a Graph Feature Preprocessor to add fan-in/fan-out, cycle and vertex-statistic features. On a downsized balanced AMLworld HI-Small subset the XGBoost model reaches over 99% accuracy, F1, precision and recall identically in clear and FHE-encrypted inference, and graph features raise F1 by 8% on the imbalanced subset, at the cost of significant FHE inference overhead.
Abstract¶
Combating money laundering has become increasingly complex with the rise of cybercrime and digitalization of financial transactions. Graph-based machine learning techniques have emerged as promising tools for Anti-Money Laundering (AML) detection, capturing intricate relationships within money laundering networks. However, the effectiveness of AML solutions is hindered by data silos within financial institutions, limiting collaboration and overall efficacy. This research presents a novel privacy-preserving approach for collaborative AML machine learning, facilitating secure data sharing across institutions and borders while preserving privacy and regulatory compliance. Leveraging Fully Homomorphic Encryption (FHE), computations are directly performed on encrypted data, ensuring the confidentiality of financial data. Notably, FHE over the Torus (TFHE) was integrated with graph-based machine learning using Zama Concrete ML. The research contributes two key privacy-preserving pipelines. First, the development of a privacy-preserving Graph Neural Network (GNN) pipeline was explored. Optimization techniques like quantization and pruning were used to render the GNN FHE-compatible. Second, a privacy-preserving graph-based XGBoost pipeline leveraging Graph Feature Preprocessor (GFP) was successfully developed. Experiments demonstrated strong predictive performance, with the XGBoost model consistently achieving over 99% accuracy, F1-score, precision, and recall on the balanced AML dataset in both unencrypted and FHE-encrypted inference settings. On the imbalanced dataset, the incorporation of graph-based features improved the F1-score by 8%. The research highlights the need to balance the trade-off between privacy and computational efficiency.
Source: arxiv
Extracted details¶
| Field | Value | Source | Confidence |
|---|---|---|---|
| contribution | Privacy-preserving GNN (GIN) pipeline with quantization and pruning for TFHE compatibility | llm | 0.85 |
| contribution | Privacy-preserving graph-feature XGBoost pipeline with Graph Feature Preprocessor | llm | 0.85 |
| contribution | Evaluation of clear vs FHE-encrypted inference on balanced and imbalanced AML datasets | llm | 0.85 |
| schemes | TFHE | llm | 0.85 |
| libraries | Concrete ML 1.4.1 | llm | 0.85 |
| libraries | XGBoost | llm | 0.85 |
| libraries | Graph Feature Preprocessor (GFP) | llm | 0.85 |
| libraries | ONNX | llm | 0.85 |
| models | XGBoost | llm | 0.85 |
| models | GNN (GIN) | llm | 0.85 |
| datasets | AMLworld HI-Small (modified balanced and imbalanced subsets) | llm | 0.85 |
| tasks | anti-money laundering detection | llm | 0.85 |
| tasks | graph-based transaction classification | llm | 0.85 |
| hardware | Intel Xeon E5-1630 v4 3.70GHz, 8 cores (CPU only) | llm | 0.85 |
| threat model | collaborative institutions evaluating models on encrypted financial data | llm | 0.85 |
| interaction | non-interactive FHE | llm | 0.85 |
| code url | https://github.com/fabecode/GraphML-FHE (URL found in the full text) | regex | 0.50 |
| code url | https://github.com/IBM/AMLSim.43 (URL found in the full text) | regex | 0.50 |
| code url | https://github.com/IBM/Multi-GNN.46 (URL found in the full text) | regex | 0.50 |
| code url | https://github.com/fabecode/GraphML-FHE | llm | 0.85 |
| limitations | FHE-encrypted inference incurs significant time overhead | llm | 0.85 |
| limitations | Dataset downsized because of FHE computational constraints | llm | 0.85 |
| limitations | Concrete ML offers no GPU support | llm | 0.85 |
| limitations | GNN FHE compatibility only explored; scalability to larger datasets not assessed | llm | 0.85 |
| key result | accuracy (clear and FHE): 0.9972 (Bayesian-optimized XGBoost, balanced Dataset 1, basic features) | llm | 0.85 |
| key result | F1 (clear and FHE): 0.9978 (balanced Dataset 1, basic features) | llm | 0.85 |
| key result | F1 improvement from graph features: +8% (imbalanced Dataset 2 (5.72% illicit)) | llm | 0.85 |
Keyword mentions¶
Counts of vocabulary terms in the full text (first page of occurrence in brackets).
- scheme: TFHE 31 [p.1], MPC 13 [p.5], CKKS 2 [p.5]
- library: Concrete (Zama) 43 [p.1], Concrete ML 26 [p.1], Zama 7 [p.1], PyTorch 6 [p.7], ONNX 3 [p.7]
- model: GNN 65 [p.1], Gradient boosting / XGBoost 41 [p.1], CNN 6 [p.5], ResNet 1 [p.5], AlexNet 1 [p.19], RNN / LSTM 1 [p.18], Random forest 1 [p.4]
- hardware: GPU 2 [p.14], Intel Xeon 1 [p.14]
- technique: Quantization 40 [p.1], Bootstrapping 12 [p.5], Convolution 12 [p.3], Programmable bootstrapping 7 [p.6], Pruning 7 [p.1], Federated learning 2 [p.5], Non-interactive 2 [p.21]
- task: Finance 37 [p.1]
- topic: Homomorphic encryption 131 [p.1], Private / secure inference 12 [p.1], Open source 5 [p.3], Privacy-preserving machine learning 3 [p.5]
Files¶
| Path (under papers/) | Collection | Pages | Size | SHA-256 | Note |
|---|---|---|---|---|---|
references/ppml/inference/effendi-2024_privacy-preserving-graph-based-ml-with-fhe.pdf |
ppml/inference | 25 | 807.8 KB | 677e5caa2a70 |
was Effendi and Chattopadhyay - 2024 - Privacy-Preserving Graph-Based Machine Learning with Fully Homomorphic Encryption for Collaborative.pdf |
Database key: privacy-preserving-graph-based-machine-learning-with-fully-homomorphic · created 2026-11-05 · updated 2026-11-05 · edit with python3 tools/paperdb.py set privacy-preserving-graph-based-machine-learning-with-fully-homomorphic COLUMN VALUE