Events Calendar

Computer engineering seminar

Speaker: Dr. Farinaz Koushanfar, Rice University

Talk Title: Engineering scalable privacy-preserving big and dense data analytics

Abstract: Data analytics on massive and often sensitive contents regularly arise in various contemporary settings ranging from cloud computing and social networking, to online services, mobile applications, and distributed processing. In this talk, I present novel computer engineering-based solutions that uniquely enable efficient and scalable explorations of the underlying patterns and dependencies present across a complex dataset, with a focus on sensitive privacy-preserving applications. The first part of the talk addresses the challenge of minimizing the computing, storage and communication overhead of the learning algorithms down to the limits of data subspaces and underlying heterogeneous platform. I demonstrate data-aware, domain-specific methodologies that are applicable to a broad class of iterative matrix-based learning algorithms and particularly efficient for challenging datasets with dense dependencies. The new techniques and methods enable optimizing for hardware acceleration as well as real-time stream processing, while they simultaneously benefit the privacy-preserving computing by pushing the limits of costly data analytics to the theoretical bounds. The second portion of the talk discusses novel scalable engineering solutions for privacy preserving computing by Yao's Garbled Circuit (GC) allowing two parties to jointly compute a function while keeping their inputs private. In contrast with the existing (software based) GC methods, I illustrate how scalable and efficient GC computation can be achieved by leveraging a new folded function description and logic synthesis methods along with our created custom libraries and constraints.

Evaluation results of our methodologies show significant improvements in memory footprint, network bandwidth, and the overall computing cost in terms of time and energy (power) compared with the prior art, often by orders of magnitude. Our scalable privacy-preserving approach enables us to implement functions that have not been reported before, small enough that they befit mobile/embedded devices. To facilitate automated end-to-end implementation, we provide a number of user-friendly APIs supported by our custom libraries. I discuss how our new findings will enable practically addressing several known classical challenges as well as exciting applications such as scalable privacy-preserving classification of visual content, secure data mining, and search.

Biography: Farinaz Koushanfar is currently an Associate Professor with the Department of Electrical and Computer Engineering, Rice University, where she directs the Adaptive Computing and Embedded Systems (ACES) Lab. She also serves as the: principal director of the TI DSP Leadership University program; and, as the associate partner of the Intel Collaborative Research Institute for Secure Computing. She received her Ph.D. degree in Electrical Engineering and Computer Science from University of California Berkeley. Her research interests include embedded/cyber-physical systems (CPS) security, hardware trust, adaptive and customizable embedded systems design, and secure function evaluation. Professor Koushanfar received a number of awards and honors for her research, mentorship, and teaching including the PECASE from president Obama, ACM SIGDA Outstanding New Faculty Award, NAS Kavli fellowship, Cisco IoT Security Grand Challenge Award, Young faculty/CAREER awards from NSF, DARPA, ONR, ARO, MIT Technology Review TR-35, and a Best Student Paper Award at ACM SIGMOBILE (Mobicom).

Host: Prof. Massoud Pedram

Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248

Audiences: Everyone Is Invited

Contact: Annie Yu