Adversarial learning and a problem of denial-of-service defense in the Cloud

Time: Tuesday, May 21, 2019 - 11:00am - 12:00pm
Type: Seminar Series
Presenter: George Kesidis; Pennsylvania State University
Room/Office: Room 335
Location:
17 Hillhouse Avenue
New Haven, CT 06511
United States

Department of Electrical Engineering & The Yale Institute for Network Science (YINS) Seminar

"Adversarial learning and a problem of denial-of-service defense in the Cloud"

Professor George Kesidis
Pennsylvania State University

Abstract: The past decade has seen increasing interest in security problems of learning systems. Though many well-cited papers do not acknowledge related prior work (e.g. robust learning, steganography), envision attacks requiring omniscient adversaries, and ignore obvious (even commonly deployed) defenses, some instances of adversarial-learning problems are novel and of legitimate concern. In this talk, we will first overview the main problem areas of adversarial learning (test-time evasions - TTEs, reverse engineering, data poisoning), and then describe a state-of-the-art TTE attack and defense for DNN classifiers of images. The defense uses an anomaly detector based on DNN-internal neuron activations. In the second half of the talk, we will consider problems of security in cloud computing systems against denial-of-service attacks. After overviewing attack-defense scenarios, we will focus a strategy shuffling client-to-server assignments leading to quarantine of clients engaging in "low volume" denial-of-service attacks.

This research is in collaboration with David J. Miller, Angelos Stavrou, Daniel Fleck, and students. This research is supported by DARPA XD3 and AFOSR DDDAS grants, Cisco URP gifts, and gifts of AWS credits.

Bio: George Kesidis received his MS (1990, machine learning and stochastic optimization) and PhD (1992, performance evaluation and networking) in EECS from U.C. Berkeley. Following eight years as a professor of ECE at the University of Waterloo, he has been a professor of EE and CSE at the Pennsylvania State University since 2000. His research interests include problems in networking, cyber security, machine learning, and performance evaluation. In the past 15 years, his research in these areas has been supported by over a dozen NSF grants and several Cisco Systems URP gifts. Currently his research is supported by grants from NSF CSR (cost-effective cloud computing), NSF NeTS (wireless network access), DARPA XD3 (cloud-side DDoS defense), AFOSR DDDAS (data-driven active learning), and Cisco URP gifts (machine learning, network security). With D.J. Miller he co-founded a small start-up on machine learning applications to security problems. His web site is http://www.cse.psu.edu/~gik2

Hosted by: Professor Leandros Tassiulas

Tuesday, May 21, 2019 at 11:00am
Location: 17 Hillhouse Avenue, Room 335