Secure Outsourced Matrix Computation and Application to Neural Networks

Explore a practical solution for encrypting matrices homomorphically and performing arithmetic operations on encrypted matrices in this 22-minute ACM conference talk. Delve into a novel matrix encoding method and an efficient evaluation strategy for basic matrix operations including addition, multiplication, and transposition. Learn how to encrypt multiple matrices in a single ciphertext for improved amortized performance. Discover the application of these techniques to neural networks and compare the results with existing methods. Gain insights into recent progress in homomorphic encryption and its functionality in various schemes.