Reproduce “Comparison of Fully Homomorphic Encryption and Garbled Circuit Techniques in Privacy-Preserving Machine Learning Inference” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kalyan Cheerla, Lotfi Ben Othmane, Kirill Morozov
A study conducted by researchers at the University of North Texas empirically compares Fully Homomorphic Encryption (FHE) and Garbled Circuits (GC) for privacy-preserving machine learning inference on a two-layer neural network. The evaluation quantifies their respective trade-offs in execution time, memory usage, communication, and output accuracy, offering practical insights into their applicability for secure neural network computations.
Key insights
- FHE offers non-interactive inference and strong confidentiality for both input and model structure, while GC is interactive and inherently leaks the model's structure.
- The choice between FHE and GC involves trade-offs: GC provides significantly faster execution and lower memory footprint, whereas FHE incurs higher computational, memory, and communication overheads.
- Numerical approximations for non-linearities and the inherent properties of cryptographic schemes introduce accuracy deviations, with FHE showing higher output variability compared to GC in the tested neural network.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.