EPSRC logo

Details of Grant 

EPSRC Reference: EP/J001384/1
Title: Towards a New Generation of Matrix Learning Methods in Machine Learning
Principal Investigator: Ying, Dr Y
Other Investigators:
Researcher Co-Investigators:
Project Partners:
Department: Computer Science
Organisation: University of Exeter
Scheme: First Grant - Revised 2009
Starts: 01 February 2012 Ends: 26 June 2013 Value (£): 99,568
EPSRC Research Topic Classifications:
Artificial Intelligence
EPSRC Industrial Sector Classifications:
No relevance to Underpinning Sectors
Related Grants:
Panel History:
Panel DatePanel NameOutcome
13 Jul 2011 EPSRC ICT Responsive Mode - July 2011 Announced
Summary on Grant Application Form
Machine Learning (ML) has been a very active field in computer science over the past two decades. The interplay between ML, statistics and numerical optimisation is becoming increasingly fruitful. In particular, the understanding of optimisation within ML is a process that has just begun and has recently received most of the attention. This project will consider challenging optimisation issues in the context of matrix learning in ML. Most of the research in this area has been based on "recycling" knowledge and general-purpose softwares acquired from numerical optimisation research. In other words, their special structures related to specific ML tasks are largely ignored which hampers their applications to large-scale datasets. At the same time, the modern technologies are creating a huge number of high-dimensional and large-scale datasets, which is particularly true in industry, e-commerce, life science and computer vision. We believe that it is now the time to build on the success of the existing approaches to develop a new generation of matrix learning methods for high-dimensional and large-scale data analysis. The main theme in this proposal is to develop a completely new eigenvalue optimisation framework for various matrix learning problems in ML by exploring their special convex structures. This new framework will not only provide new insights into matrix learning problems in ML but also, most importantly, the beautiful mathematics underlying eigenvalue optimisation will greatly facilitate the design of efficient algorithms. More powerful ML methodologies than generic SDP solvers and exiting first-order methods will arise from the innovative interaction between ML and eigenvalue optimization. Therefore, the proposed research theme represents a significant shift in emphasis.

Specifically, this proposal aims to develop a new line of matrix learning methods in ML by exploring their special structures in convex optimisation which includes developing new models, designing efficient optimisation algorithms and rigorously establishing their convergence characteristics. Extensive empirical studies will be carried out to illustrate the potential of the new methods developed and refine the state-of-the-art results. In particular, we will implement the algorithms by means of user-friendly software tools, and apply them on two large and challenging application problems: face identification (e.g. Labeled Faces in the Wild dataset from Yahoo News) and collaborative filtering (prediction of users' preferences to products).

Key Findings
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Potential use in non-academic contexts
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Description This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Date Materialised
Sectors submitted by the Researcher
This information can now be found on Gateway to Research (GtR) http://gtr.rcuk.ac.uk
Project URL:  
Further Information:  
Organisation Website: http://www.ex.ac.uk