Abstract
The development of deep learning models drastically improves the quality of image retrieval systems, and have been widely deployed in real systems or serve as a component in other applications such as person reidentification. However, recent works have shown that these systems are vulnerable to adversarial attacks where a small perturbation on the query image could drastically change the returned results. In this talk, the speaker will survey and introduce several adversarial attack methods to deep image retrieval systems under different settings, including white-box and black-box attacks, and different retrieval system architectures. Experimental results on several benchmark datasets and real image search engines demonstrate the effectiveness of such attacks.
About the speaker
Prof. Wang Wei obtained his PhD in Computer Science from HKUST in 2004. He then joined the University of New South Wales as a Lecturer and is currently the Professor of Computer Science & Engineering there. He is also Visiting Professor of Computer Science and Engineering and Red Bird Visiting Scholar at HKUST.
Prof. Wang’s research revolves around query processing and optimization, and novel data management applications. His research interests include Similarity Query Processing, Artificial Intelligence, Knowledge Graphs, and Security for AI Models.
Prof. Wang received the Best Student Paper Award at the International Conference on Database Systems for Advanced Applications (DASFAA) in 2016, and was the best paper nominee of the annual IEEE International Conference on Data Engineering (ICDE) in 2010 and 2018. He served organizational and review roles in major international conferences and journals. He has published over a hundred research papers and is currently an Associate Editor of IEEE Transactions on Knowledge and Data Engineering (TKDE), and program committee members in various first-tier conferences.
|
