Deep Transfer Learning: Generalization on Clean and Adversarial Data

PhD Thesis Proposal Defence


Title: "Deep Transfer Learning: Generalization on Clean and Adversarial 
Data"

by

Miss Yinghua ZHANG


Abstract:

Machine learning, especially deep learning, has made remarkable progress 
in the past few years, yet the success of deep learning systems heavily 
relies on massive labeled data, while labeled data are usually scarce in 
real-world applications. Transfer learning, which leverages the knowledge 
in well-annotated source domain(s) and helps learning in a low-resource 
target domain, can effectively reduce the dependency on labeled data. In 
this proposal, we study the generalization ability of deep transfer 
learning models on clean and adversarial data and build deep transfer 
learning models that are effective and robust.

The two works, parameter transfer unit and Fisher deep domain adaptation, 
address two common transfer learning settings, inductive transfer learning 
and transductive transfer learning, respectively. Our proposed methods 
address the challenges of these two settings and improve the transfer 
performance on clean data.

While most transfer learning research works focus on the transfer 
performance on clean data, transfer learning models are under the threat 
of adversarial attacks, and such risks have been less studied. To fill the 
gap, we systematically evaluate the robustness of transfer learning models 
under white-box and black-box Fast Gradient Sign Method (FGSM) attacks via 
empirical experiments.

The empirical evaluations on the robustness of transfer learning models 
indicate that adversarial attacks towards transfer learning models raise 
newly-rising challenges, and there is much room for exploration. Some 
directions of future works are discussed in the last part.


Date:			Thursday, 28 January 2021

Time:                  	10:00am - 12:00noon

Zoom Meeting: 
https://hkust.zoom.us/j/95094232900?pwd=aGhoay9oNEtyUHJuN2tXU2FjTW1adz09

Committee Members:	Prof. Qiang Yang (Supervisor)
 			Dr. Yangqiu Song (Supervisor)
  			Dr. Kai Chen (Chairperson)
 			Dr. Qifeng Chen


**** ALL are Welcome ****