Deep Testing of Advanced Learning Systems

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

Final Year Thesis Oral Defense

Title: "Deep Testing of Advanced Learning Systems"

by

CHAN Jung

Abstract:

While deep learning techniques may have evolved from statistical analysis, 
there is, to the best of our knowledge, little research into applying 
statistical methods into the study of deep learning systems. The testing 
of deep neural networks is also a relatively new field, with few up to 
date, comprehensive surveys in the topic.

This thesis presents a comprehensive survey of the field of deep neural 
network testing, with additional detail in related research, like 
adversarial generation. This thesis also explores applying statistical 
analysis of the neuron outputs of certain layers of a deep neural 
networks, with some surprising insights being gleaned. Using this 
statistical information, we implement a novel iteration based on existing 
tools for testing and evaluating deep neural network. This novel tool is 
compared with existing methods, showing that it is more effective within 
certain use cases.


Date            : 2 May 2019 (Thursday)

Time            : 14:30 - 15:10

Venue           : Room 4621 (near lifts 31/32), HKUST

Advisor         : Prof. CHEUNG Shing-Chi

2nd Reader      : Dr. SONG Yangqiu