Reproduce “High-Resolution Convolutional Neural Networks on Homomorphically Encrypted Data via Sharding Ciphertexts” 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
Vivian Maloney, Richard F. Obrecht, Vikram Saraph, Prathibha Rama, Kate Tallaksen
Researchers developed methods to efficiently and accurately evaluate high-resolution Deep Convolutional Neural Networks on homomorphically encrypted data. Their approach enabled ResNet-50 to achieve 80.2% accuracy on ImageNet-1k and demonstrated speedups of 4.6x to 6.5x compared to prior state-of-the-art encrypted inference techniques.
Key insights
- Large, high-resolution image data can be effectively managed within FHE ciphertext limitations through sharding without compromising performance or privacy.
- Simplifying data packing to a standard row-major format and optimizing homomorphic operations (like strided convolution via downsampling) leads to substantial latency reductions.
- Regularizing the input distribution to activation functions during training allows for more efficient, low-degree polynomial approximations of non-linearities in FHE.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.