LEARNING BILINGUAL RELATIONS FOR MODELING IMPROVISATION

PhD Qualifying Examination


Title: "LEARNING BILINGUAL RELATIONS FOR MODELING IMPROVISATION"

by

Mr. Karteek ADDANKI


Abstract:

In this survey, we investigate various techniques for learning bilingual 
relations and highlight the significance of learning bilingual relations for 
modeling improvisation. Despite improvisation being an integral part of various 
art forms and a prime example of human intelligence and creativity, very little 
research has been done in machine learning toward modeling improvisation. While 
systems have been built for generating improvised output in the domains of 
lyrics and poetry, almost all these efforts employ task and domain specific 
resources without a major emphasis on the nature of representations needed to 
model improvisation efficiently. Also, using domain-specific resources makes it 
expensive to transfer these models to new resource-scarce languages.

In an effort to identify research directions for modeling improvisation 
efficiently in an unsupervised fashion, we survey both symbolic and deep 
learning approaches for learning bilingual relationships and discuss their 
merits in the context of machine improvisation. We preface the survey with a 
discussion of various systems that improvised output in novel domains such as 
poetry and music using statistical learning techniques. As evaluating 
improvised output is particularly challenging which makes it difficult to 
compare the performance of different models, we also briefly discuss the 
challenges in evaluating improvisation.


Date:			Wednesday, 29 April 2015

Time:                  	2:00pm - 4:00pm

Venue:                  Room 3494
                         Lifts 25/26

Committee Members:	Prof. Dekai Wu (Supervisor)
 			Prof. Dit-Yan Yeung (Chairperson)
 			Prof. Qiang Yang
 			Prof. Pascale Fung (ECE)


**** ALL are Welcome ****