Vulnerability Analysis of Neural Networks

The Hong Kong University of Science and Technology
Department of Computer Science and Engineering

Final Year Thesis Oral Defense

Title: "Vulnerability Analysis of Neural Networks"

by

KUO Tzu-yang

Abstract:

Through efforts from researchers, many research papers have demonstrated 
that machine learning models are inherently vulnerable to adversarial 
samples, where maliciously crafted inputs can trigger target machine 
learning systems to misbehave, by various adversarial attack techniques on 
machine learning systems. Many research papers have proposed defense 
methods against emerging adversarial attacks. However, many of the defense 
mechanisms were not able to capture various types of adversarial attacks 
perfectly. In the thesis project, we leverage internal information in the 
forwarding process to construct a defense mechanism capturing various 
types of adversarial attacks simultaneously.


Date            : 15 May 2020 (Friday)

Time            : 14:00 - 14:40

Zoom Meeting    : https://hkust.zoom.us/j/113843145

Advisor         : Prof. CHEUNG Shing-Chi

2nd Reader      : Prof. YEUNG Dit-Yan