Navigation

Recent Publications

More...

Home

Fast and Robust Matrix Factorization

huilee - Posted on 14 December 2016

Project Description: 

Matrix factorization (MF) is one of the fundamental techniques for analyzing latent relationship between two entities. MF is successful in many tasks such as click prediction for web search, link prediction and community detection for social networks, image alignment, video surveillance, to name a few. Due to its power to handle problems in many real applications, MF has been recognized as one of the Top 10 algorithm. With the recent advent of programmer-friendly parallel processing frameworks, Internet-scale matrix factorizations have become practicable and are of increasing interest to both academic community and industry. The objective of this project is to design and develop scalable and robust parallel matrix factorization for web-scale data. The proposed framework is scalable to huge volume of data and robust to matrices with different distributions.

Researcher name: 
Hui Li
Researcher position: 
PhD
Researcher department: 
Department of Computer Science
Researcher email: 
hli2@cs.hku.hk
Research Project Details
Project Duration: 
12/2016 to 12/2017
Project Significance: 
Matrix factorization (MF) is one of the fundamental techniques for analyzing latent relationship between two entities. MF is successful in many tasks such as click prediction for web search, link prediction and community detection for social networks, image alignment, video surveillance, to name a few. Due to its power to handle problems in many real applications, MF has been recognized as one of the Top 10 algorithm. With the recent advent of programmer-friendly parallel processing frameworks, Internet-scale matrix factorizations have become practicable and are of increasing interest to both academic community and industry. The objective of this project is to design and develop scalable and robust parallel matrix factorization for web-scale data. The proposed framework is scalable to huge volume of data and robust to matrices with different distributions.