SIRNN- A Math Library for Secure RNN Inference

Explore a comprehensive 15-minute IEEE presentation on SIRNN, a mathematical library designed for secure RNN inference. Delve into the world of secure inference, RNN applications, and the challenges they present. Discover the SIRNN hierarchy, its building blocks, and communication optimization techniques. Learn about fixed-point representation and various mathematical implementations. Examine the library's performance through evaluations using the Google 30 Benchmark and Heads Benchmark. Gain valuable insights into the future of secure machine learning inference in this informative talk.