MD-ML: Super Fast Privacy-Preserving Machine Learning for Malicious Security with Dishonest Majority

Watch a 12-minute conference presentation from USENIX Security '24 exploring MD-ML, a groundbreaking framework for privacy-preserving machine learning (PPML) that operates with malicious security under dishonest majority conditions. Learn how researchers from Shanghai Jiao Tong University developed novel protocols that significantly improve the efficiency of secure multi-party computation in machine learning applications. Discover the framework's innovative features, including single-element-per-party dot product communication and cost-effective multiply-then-truncate operations. Examine benchmark results comparing MD-ML's performance against existing solutions across various neural network architectures like LeNet, AlexNet, and ResNet-18, demonstrating substantial speed improvements of up to 157.7x faster online execution in WAN environments.