TOWARDS BETTER UNDERSTANDING OF DEEP LEARNING WITH VISUALIZATION

PhD Qualifying Examination


Title: "TOWARDS BETTER UNDERSTANDING OF DEEP LEARNING WITH VISUALIZATION"

by

Mr. Haipeng ZENG


Abstract:

Deep learning can learn representations of data for different kinds of 
tasks by using computational models with multiple processing layers. 
Remarkable progress has been made in detection and classification tasks in 
recent years. However, there is still no clear understanding of the inner 
working mechanisms. Usually, to get a better deep learning model, people 
have to undergo a substantial amount of trial-and-error procedures, which 
is very inconvenient and time-consuming. Consequently, there has been a 
dramatical interest in using visualization to help people better 
understand and train deep learning models intuitively. Existing work 
mainly focuses on three aspects, i.e., feature visualization, relationship 
visualization and process visualization, which show the clear advantages 
in helping understand the reasoning behind deep learning models.

In this survey, we first introduce the background and characteristics of 
deep learning and then give a comprehensive review of how visualization 
techniques are used to help understand and train deep learning models. 
Finally, we conclude the survey with a discussion of future research 
directions.


Date:			Thursday, 10 November 2016

Time:                  	3:00pm - 5:00pm

Venue:                  Room 3494
                         Lifts 25/26

Committee Members:	Prof. Huamin Qu (Supervisor)
 			Prof. Cunsheng Ding (Chairperson)
 			Dr. Yangqiu Song
 			Prof. Chi-Keung Tang


**** ALL are Welcome ****