Lifelong Machine Learning Literature Survey

PhD Qualifying Examination


Title: "Lifelong Machine Learning Literature Survey"

by

Mr. Lianghao LI


Abstract:

In the past decades, there have been significant advances in machine learning 
theory and algorithms. However, there is a big missing part in the most of 
learning algorithms proposed in the past decades if we compare them to how 
human learn to solve problems. That is the human's continuously study 
capability. Most of the machine learning theory and algorithms still focus on 
solving the problem for one particular task, such as text classification, image 
classification, image segmentation, etc. But human learns to solve various 
problems one after the other, continuously. For example, a musician might learn 
how to play many different instruments, studies how to compose and perform 
different music year over year. Because of the continuously study capability, 
that musician should be able to learn how to play guitar very quick if she 
already knows how to play piano and compose music. We call a machine learning 
system which has this kind of continuously study capability a lifelong learning 
system. Another important reason that why lifelong learning will be a key area 
in machine learning is that a large amount of labeled data from a large number 
of tasks become available. That is largely driven by the prevalence of deep 
learning which requires a lot of labeled data to learn more complex models with 
deep architectures. The huge amount of labeled data from many different tasks 
unlock lifelong learning system research and also push the community to start 
thinking about beyond the traditional one-hot model training. In this survey, 
we review in detail about the lifelong learning paradigm and related works.


Date:			Monday, 14 May 2018

Time:                  	3:00pm - 5:00pm

Venue:                  Room 3494
                         Lifts 25/26

Committee Members:	Prof. Qiang Yang (Supervisor)
 			Prof. Dit-Yan Yeung (Chairperson)
 			Prof. Lei Chen
 			Dr. Yangqiu Song


**** ALL are Welcome ****