Brandon Reagen
Assistant Professor – Department of Electrical and Computer Engineering – New York University
Invited Talk: Characterizing the Performance of End-to-End Private Inference
Abstract: As privacy and security continue to increase in importance new techniques are needed to uphold strong guarantees to users surrounding how their data is used. One emerging solution is privacy-preserving computation (PPC), which includes a set of cryptographic primitives that enable computation directly on encrypted data (e.g., homomorphic encryption and garbled circuits). A key motivation and benchmark for these primitives has been machine learning: its wide use and amenability to PPC constraints makes it a natural starting point. In this talk I will present characterization results for running private inferences on end-to-end systems in a classic client-cloud setup. The data paints a clear picture motivating research in three areas: computation, communication, and storage. To conclude, optimizations will be presented to address these overheads.
Biography: Brandon Reagen is an Assistant Professor at New York University in the Tandon school of engineering. He received his Ph.D from Harvard in 2018 and undergraduate degree from the University of Massachusetts, Amherst in 2012. His research focuses on novel hardware accelerator designs for high performance and low power with applications in privacy preserving computation, including homomorphic encryption and secure multi-party computation.