DarKnight: Training ML Models with Private Data
Abstract:
This talk will describe DarKnight, a framework for large DNN training while protecting input privacy and computation integrity. DarKnight relies on cooperative execution between trusted execution environments (TEE) and accelerator, where the TEE provides privacy and integrity verification, while accelerators perform the bulk of linear algebraic computation to optimize the performance. In particular, DarKnight uses a customized data encoding strategy based on matrix masking to create input obfuscation within a TEE. The obfuscated data is then offloaded to GPUs for fast linear algebraic computation. DarKnight's data obfuscation strategy provides provable data privacy and computation integrity in the cloud servers. While prior works tackle inference privacy and cannot be utilized for training, DarKnight's encoding scheme is designed to support both training and inference.
Speaker
Dr. Murali Annavaram
Murali Annavaram is a Professor in the Ming-Hsieh Department of Electrical Engineering and in the department of Computer Science (joint appointment) at the University of Southern California. His research group tackles a wide range of computer system design challenges, relating to energy efficiency, security and privacy. He has been inducted to the hall of fame for three of the prestigious computer architecture conferences ISCA, MICRO and HPCA. He served as a Technical Program Chair for HPCA 2021, and served as the General Co-Chair for ISCA 2018. Prior to his appointment at USC he worked at Intel Microprocessor Research Labs from 2001 to 2007. His work at Intel lead to the first 3D microarchitecture paper, and also influenced Intel’s TurboBoost technology. In 2007 he was a visiting researcher at the Nokia Research Center working on mobile phone-based wireless traffic sensing using virtual trip lines, which later become Nokia Traffic Works product. IN 2020 he was a visiting faculty scientist at Facebook, where he designed the checkpoint systems for distributed training. Murali co-authored Parallel Computer Organization and Design, a widely used textbook to teach both the basic and advanced principles of computer architecture. Murali received the Ph.D. degree in Computer Engineering from the University of Michigan, Ann Arbor, in 2001. He is a Senior Member of IEEE and ACM.