A Survey on Federated Transfer Learning

PhD Qualifying Examination


Title: "A Survey on Federated Transfer Learning"

by

Mr. Xueyang WU


Abstract:

The current wave of artificial intelligence (AI) applications depends on 
large-scale data, while in practice required data distribute on many parties 
separately, and each party usually has a small amount of data. For large and 
wealthy companies, the most popular and easy solution to this problem is to 
collect data from individuals or to purchase labeled data from data providers. 
However, such solutions will go to the end due to the trend of increasing 
concerns about data privacy and data security. Nowadays, AI applications face 
the problem of using data, more specifically, how to utilize diverse and 
fragment data that are separately distributed in different parties or clients.

Federated learning is proposed to address the problem of privately isolated 
small data learning, whose main idea is to compose a federation of data in 
which all parties virtually assemble their data without security and privacy 
problems. For statistical learning, federated learning is still facing four 
main challenges, which has become an enormous obstacle for the broad 
utilization of federated learning. From the other aspect, learning from diverse 
and fragment data distribution has been studied for decades, which is 
summarized as transfer learning. The target of transfer learning is to utilize 
the knowledge from an abundant dataset to help the learning on small local 
task.

In this survey, we first introduce federated learning and then discuss its 
statistical challenges in detail. We propose a precise categorization of 
algorithms addressing such challenges in federated learning, named federated 
transfer learning, and investigate existed researches related to this topic and 
categorize them under our framework. Further, we study the methodologies of 
federated transfer learning and provide a concise guideline for researchers 
designing federated transfer learning algorithms according to their application 
and purpose. Finally, we also explore the current and prospective applications 
about federated transfer learning.


Date:			Tuesday, 9 April 2019

Time:                  	2:00pm - 4:00pm

Venue:                  Room 5501
                         Lifts 25/26

Committee Members:	Prof. Qiang Yang (Supervisor)
 			Prof. Cunsheng Ding (Chairperson)
 			Dr. Kai Chen
 			Dr. Xiaojuan Ma


**** ALL are Welcome ****