CSIS Seminar

CYSE Faculty Candidate Seminar: Explore Security and Privacy in Deep Image Retrieval Systems

Abstract

The rapid growth of large image and video collection has made content-based image retrieval possible at a large scale, e.g., Google, Pinterest, Bing, TinEye and Alibaba. Powered by deep learning, these techniques have been increasingly built into social networks, e-commerce and fashion design to capture semantic similarities from visual queries for finer results. Unfortunately, by inheriting the backend from deep learning, these systems are also vulnerable to the well-known adversarial examples. In today’s talk, we will look at the security and privacy side in hash-based image retrieval systems. For search accuracy, service providers typically maintain a growing database via automated crawling and indexing of images online. When private or copyrighted images are collected into the database, an adversary can extract those images by querying similar images from the target categories. We will develop a privacy-preserving mechanism based on adversarial examples to “stash” the private images in the deep hash space while maintaining perceptual similarity. Then we will take a look at the security problem when attackers embed malicious images into the database and point them to specific categories to be retrieved by the user. We will develop a practical method to enhance black-box transferability for targeted attacks. We will conclude this talk with a discussion of the ongoing challenges and connections between security and computation, along with the directions for future works across the hardware and software stacks. Zoom Link: https://gmu.zoom.us/j/99398956647

Speaker Bio

Dr. Cong Wang is an Assistant Professor from the Computer Science Department at the Old Dominion University in Norfolk, VA. He is also affiliated with the Center of Cybersecurity Research and Education. Cong’s research focuses on addressing security and performance challenges in Mobile/Cloud Computing, IoT, Machine Learning and Systems. He has published in networking, distributed/mobile computing, AI and security conferences and journals including INFOCOM, IPDPS, ICDCS, CVPR, IJCAI, CCS, ACM MM, PERCOM, IEEE Trans. On Mobile Computing, IEEE Trans. On Computers and IEEE Trans. On Distributed and Paralleled Systems. He received his Ph.D from the Department of Electrical and Computer Engineering, Stony Brook University(16’), B. Eng in Information Engineering from the Chinese University of Hong Kong (08’) and M. Sc in Electrical Engineering from Columbia University (09’). He is the recipient of IEEE PERCOM Mark Weiser Best Paper Award in 2018, COVA-CCI Cybersecurity Research and Innovation Award, ODU’s Richard Cheng Innovative Research Award in 2020, and NSF CAREER Award in 2021.