Monday, April 26, 2021 – 12:00pm to 1:00pm
Virtual Presentation – ET Remote Access – Zoom
SRINI DEVADAS, Webster Professor of Electrical Engineering and Computer Science https://people.csail.mit.edu/devadas/
DAUnTLeSS: Data Augmentation and Uniform Transformation for Learning with Scalability and Security
We revisit private optimization and learning from an information processing view. Different from the classic cryptographic framework of operation-by-operation obfuscation, a private learning and inference framework via either data-dependent or random transformation on the sample domain is proposed, along with a security analysis framework, termed probably approximately correct (PAC) inference resistance, which bridges the information loss in data processing and prior knowledge.
We study the applications of such a framework from generalized linear regression models to modern learning techniques, such as deep learning. We explore the advantages of this new random transform approach with respect to underlying privacy guarantees, computational efficiency and utility for neural networks.
Hanshen Xiao, PhD candidate at MIT, is the primary author of this work.
Srini Devadas is the Webster Professor of EECS at MIT where he has been on the faculty since 1988. His current research interests are in computer security, computer architecture, and applied cryptography. Devadas received the 2015 ACM/IEEE Richard Newton award, the 2017 IEEE W. Wallace McDowell award and the 2018 IEEE Charles A. Desoer award for his research in secure hardware. He is a Fellow of the ACM and IEEE. He is a MacVicar Faculty Fellow, an Everett Moore Baker and a Bose award recipient, considered MIT’s highest teaching honors.
Faculty Host: Elaine Shi
CyLab seminars are closed to the public and open to CyLab partners and CMU faculty, students and staff.
Zoom Participation. See announcement.