A Survey on Neural Network-Based Unsupervised Text Style Transfer

PhD Qualifying Examination


Title: "A Survey on Neural Network-Based Unsupervised Text Style Transfer"

by

Mr. Dongkyu LEE


Abstract:

Text style transfer aims to transfer a style of an input to a target style 
while preserving the underlying meaning and fluency of the input sentence. 
A style of a corpus can be defined as a prevailing feature/characteristic 
that the instances commonly share, such as sentiment, formality, or tense. 
The task is closely linked to controllable text generation, with style 
being the subject of the control; hence by nature, a wide range of 
applications is feasible. This survey explores the overview of text style 
transfer, in terms of training scheme as well as different model 
architectures. The scope of this work is mainly focused on unsupervised 
text style transfer, in which paired style transfer sentences are not 
present in training. The survey starts by introducing the task definition, 
discusses previous approaches in unsupervised text style transfer 
learning, and highlights challenges in current research.


Date:			Tuesday, 19 April 2022

Time:                  	4:00pm - 6:00pm

Zoom Meeting: 
https://hkust.zoom.us/j/95901560413?pwd=bXVmcEtkWGhqTFFUV3l0dUgrdWtmdz09

Committee Members:	Prof. Nevin Zhang (Supervisor)
 			Prof. Fangzhen Lin (Chairperson)
 			Dr. Minhao CHENG
 			Dr. Brian Mak


**** ALL are Welcome ****