Front entry of Building 99 on Microsoft Campus
2021年7月1日 2024年2月22日

Cryptography and Privacy Colloquium | Redmond

地点: Microsoft Research Redmond

Secure Inference of Large Language Models and large scale CNNs through Function Secret Sharing

Nishanth Chandran, Microsoft Research India

March 21, 2024 | 9:00 AM | Virtual

Abstract

Secure inference enables a model publisher to offer inference-as-a-service to a model consumer in such a way that the weights (or prompt) of the model are kept hidden from the consumer and the consumer’s input is kept hidden from the publisher. Cryptographically, this is realized through specialized secure two-party computation (2PC) protocols. A recent paradigm for 2PC protocols (in the preprocessing model) have emerged using the technique of function secret sharing. These techniques shift the overheads in 2PC from communication to computation. In this talk, we will cover these recent techniques that have enabled secure inference of ImageNet scale CNNs as well as Large Language Models with only a small overhead over executing them in the clear.

Biography

Nishanth Chandran is a Principal Researcher at Microsoft Research, India. His research interests are cryptography and security. Prior to joining MSRI, Nishanth was a Researcher at AT&T Labs, and before that he was a Post-doctoral Researcher at MSR Redmond. Nishanth is a recipient of the 2010 Chorafas Award for exceptional achievements in research and his research has received coverage in science journals and in the media at venues such as Nature and MIT Technology Review. He has published several papers in top computer science conferences and journals such as Crypto, Eurocrypt, IEEE S&P, CCS, STOC, FOCS, and so on. His work on position-based cryptography was selected as one of the top 3 works and invited to QIP 2011 as a plenary talk. Nishanth has served on the technical program committee of all the top cryptography conferences on several occasions and he holds many US Patents. Nishanth received his Ph.D. in Computer Science from UCLA, M.S. in Computer Science from UCLA, and B.E. in Computer Science and Engineering from Anna University (Hindustan College of Engineering), Chennai. Nishanth is also a top ranking All India Radio South Indian Classical Violinist and has performed at international venues such as the Hollywood Bowl, Los Angeles and the Madras Music Academy, Chennai.

TrustRate: A Decentralized Platform for Hijack-Resistant Anonymous Reviews

Rohit Dwivedula, University of Texas – Austin

April 5, 2024 | 10:30 AM | Virtual

Abstract

Reviews and ratings by users form a central component in several widely used products today (e.g., product reviews, ranking content, etc.), but today’s platforms for managing such reviews are centralized, ad-hoc and vulnerable to various forms of tampering. TrustRate is an end-to-end decentralized, hijack-resistant platform for authentic, anonymous, tamper-proof reviews. With a prototype implementation and evaluation at the scale of thousands of nodes, we demonstrate the efficacy and performance of our platform, towards a new paradigm for building products based on trusted reviews by end users without having to trust a single organization that manages the reviews. 

Biography

Rohit Dwivedula is a first year PhD student at the University of Texas – Austin, advised by Aditya Akella and Daehyeok Kim. Before that, he was a research fellow at Microsoft Research, India and worked on research problems in two areas: (1) systems + privacy and security, and (2) AI infrastructure.