2020/06/08 by Théo Ryffel, Pierre Tholoniat, Ryffel, Théo +5 · 4 citations
Computer Science · #Cryptography and Data Security #Cryptography and Security (cs.CR) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Privacy-Preserving Technologies in Data #Stochastic Gradient Optimization Techniques
paper · pdf · doi:10.48550/arxiv.2006.04593
openalex publication_date 2020/06/08 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
We propose AriaNN, a low-interaction privacy-preserving framework for private\nneural network training and inference on sensitive data. Our semi-honest\n2-party computation protocol (with a trusted dealer) leverages function secret\nsharing, a recent lightweight cryptographic protocol that allows us to achieve\nan efficient online phase. We design optimized primitives for the building\nblocks of neural networks such as ReLU, MaxPool and BatchNorm. For instance, we\nperform private comparison for ReLU operations with a single message of the\nsize of the input during the online phase, and with preprocessing keys close to\n4X smaller than previous work. Last, we propose an extension to support n-party\nprivate federated learning. We implement our framework as an extensible system\non top of PyTorch that leverages CPU and GPU hardware acceleration for\ncryptographic and machine learning operations. We evaluate our end-to-end\nsystem for private inference between distant servers on standard neural\nnetworks such as AlexNet, VGG16 or ResNet18, and for private training on\nsmaller networks like LeNet. We show that computation rather than communication\nis the main bottleneck and that using GPUs together with reduced key size is a\npromising solution to overcome this barrier.\n