Bibliographic record
Abstract
FastLloyd: Federated, Accurate, Secure, and Tunable k-Means Clustering with Differential Privacy FastLloyd is an approach to privacy-preserving k-means clustering in horizontally federated settings. It offersstate-of-the-art utility while providing formal privacy guarantees through differential privacy, and achieves orders ofmagnitude faster performance compared to previous privacy-preserving clustering methods. Overview This repository implements the FastLloyd protocol described in the paper "FastLloyd: Federated, Accurate, Secure, andTunable k-Means Clustering with Differential Privacy". FastLloyd addresses the challenging problem of collaborativeclustering across multiple data owners without compromising privacy, through: 1. A novel differentially private k-means algorithm with radius constraints2. A lightweight secure aggregation protocol for federated settings Installation Requirements - Python 3.8 or higher- Open MPI (for multiparty communication)- Required Python packages listed in `env.yml` Setup 1. Download the repoistory 2. Extract the data and navigate to code directory: mkdir -p datafor f in *.tar.xz; do tar --extract --xz --file="$f" --directory=datadonemv data D-Diaa-FastLloyd-879df3a/cd D-Diaa-FastLloyd-879df3a/ 3. Create and activate the conda environment: conda env create -f env.ymlconda activate fastlloyd Usage Running Experiments FastLloyd supports multiple experiment types: 1. **Accuracy**: Evaluate clustering quality across different privacy settings python experiments.py --exp_type "accuracy" 2. **Scale**: Analyze scalability with dataset size, dimensions, and number of clusters python experiments.py --exp_type "scale" 3. **Timing**: Measure communication and computation time mpirun -np 3 python experiments.py --exp_type "timing" You can also use the provided scripts to run multiple experiment types: bash scripts/experiment_runner.sh # For accuracy and scale experimentsbash scripts/timing_runner.sh # For timing experiments with varying numbers of clients Visualization The repository includes several visualization tools in the `plots` directory: - `per_dataset.py`: Creates performance visualizations for individual datasets- `scale_heatmap.py`: Generates heatmaps to analyze scalability- `synthetic_bar.py`: Creates bar plots comparing performance on synthetic datasets- `ablation_plots.py`: Creates plots for ablation studies- `timing_analysis.py`: Analyzes and reports execution timing data Customization You can customize various aspects of the experiments through the argument parser in `experiments.py`: python experiments.py --exp_type "test" --datasets "mnist" "adult" --method "diagonal_then_frac" --alpha 0.8 --post "fold" --results_folder "my_results" Key parameters include: - `--exp_type`: Type of experiment to run (accuracy, scale, timing, test)- `--datasets`: Datasets to use for the experiment- `--method`: Maximum distance method to use- `--alpha`: Maximum distance parameter- `--post`: Post-processing method for centroids- `--results_folder`: Folder to store results Citation If you use FastLloyd in your research, please cite the paper: @inproceedings{diaa2025fastlloyd, title={FastLloyd: Federated, Accurate, Secure, and Tunable $k$-Means Clustering with Differential Privacy}, author={A. Diaa and T. Humphries and F. Kerschbaum}, eventtitle = {The 34th {USENIX} Security Symposium}, year={2025},}
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.012 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.002 | 0.002 |
| Science and technology studies | 0.002 | 0.001 |
| Scholarly communication | 0.005 | 0.006 |
| Open science | 0.007 | 0.010 |
| Research integrity | 0.003 | 0.004 |
| Insufficient payload (model declined to judge) | 0.139 | 0.138 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".