A survey on Explanation methods of Machine Learning

PhD Qualifying Examination


Title: "A survey on Explanation methods of Machine Learning"

by

Mr. King Sun CHAN


Abstract:

With the advent of deep learning techniques, machine learning has made 
significant improvement on its performance in a variety of tasks. 
Widespread adoption of machine learning models in our systems, especially 
those being employed in high stake tasks, has enhanced the need for 
explanation methods to help us understand the mechanism underlying the 
algorithms of these models. Gaining a better understanding of the 
abilities and restrictions of these models, especially, deep neural 
networks, is becoming increasingly important. In some countries, it is a 
legal responsibility of artificial intelligent system providers to give 
explanation on how their systems reach specific decisions to stakeholders.

This survey is to provide an overview of explanation methods (1) for 
analyzing the models after training (post-hoc) (2) applicable to specific 
type of models or to all types of models (3) give explanation to certain 
instances or to the holistic model behaviors. As there are a variety of 
machine learning models architectures such as linear models, convolutional 
neural networks, recurrent neural networks, which cater for different data 
types such as images, audios, texts, tabular data, etc., applicability of 
these explanation methods on various data types for different model 
architectures are highlighted.


Date:  			Wednesday, 15 June 2022

Time:                  	2:00pm - 4:00pm

Zoom Meeting:
https://hkust.zoom.us/j/98485898686?pwd=bHlpZEcrU0JsUExuSzZqTW84Sk1Edz09

Committee Members:	Prof. Shing-Chi Cheung (Supervisor)
 			Dr. Yangqiu Song (Chairperson)
 			Dr. Shuai Wang
 			Prof. Raymond Wong


**** ALL are Welcome ****