event

Ph.D. Proposal by Xinyu Xing

Primary tabs

Ph.D. Thesis Proposal Announcement

Title: Exploiting and Protecting Intelligent Systems

Xinyu Xing
Ph.D. Student
School of Computer Science
College of Computing
Georgia Institute of Technology

Date: Thursday, October 9, 2014
Time: 10:00 AM - 12:00 PM EST
Location: Klaus 3126

Committee:
Dr.Wenke Lee (Advisor, School of Computer Science, Georgia Institute of Technology)
Dr. Nick Feamster (Co-advisor, School of Computer Science, Georgia Institute of Technology)
Dr. Mustaque Ahamad (School of Computer Science, Georgia Institute of Technology)
Dr. Hongyuan Zha (School of Computational Science and Engineering, Georgia Institute of Technology)
Dr. Michael Bailey (College of Engineering, University of Michigan)

Abstract:
Modern computer applications widely utilize intelligent systems to appeal to the specific interests, viewpoints, and contexts of individual users. Ideally, the intelligent systems allow applications to highlight information uniquely relevant to each of their users, thereby increasing user satisfaction. Unfortunately, as we observe in recent literatures, the intelligent systems employed by popular applications have not been hardened against attacks. In this dissertation, I summarize existing attacks against intelligent systems into two classes -- the one compromising the output of an intelligent system and the one exploiting its intelligent algorithms. Then, I propose a new class of attacks exploiting the input data of an intelligent system.

Aiming at the aforementioned attacks, I also propose three defense mechanisms to secure different intelligent systems -- ad networks, text classification tools and personalized search. First, to harden ad networks, I develop Expector, a system that automatically inspects and identifies browser extensions that abuse their privileges and manipulate ads that ad networks place on a user webpage. Next, I propose a framework, PrivClass for text classification tools. It allows users to share their probability beliefs with others while minimizing data leakage. Finally, I present Bobble, a Web browser extension that alerts a user the search results potentially manipulated by adversaries.

Status

  • Workflow Status:Published
  • Created By:Danielle Ramirez
  • Created:10/07/2014
  • Modified By:Fletcher Moore
  • Modified:10/07/2016

Categories

Target Audience