A Survey of Online Learning Algorithms

PhD Qualifying Examination


Title: "A Survey of Online Learning Algorithms"

Mr. Weike Pan


Abstract:

Machine learning algorithms have been widely used in diverse domains
ranging from engineering, medical science, earth science, social science
to economics. But, most existed algorithms are offline and can not be
easily adapted to the online learning task. Online learning, both
prediction and clustering, is a task to make a decision and update the
model simultaneously on-the-fly with the sequentially arriving data.
Online prediction and clustering algorithms have a wide spectrum of
applications, e.g. email spam filtering, personalized content
recommendation, social network analysis, etc.

In this survey, we first review three traditional online prediction
algorithms: the classical perceptron algorithm, incremental SVM, and
online learning with kernels. And then we focus on the recently developed
online convex programming techniques, including the greedy projection
approach, the primal-dual framework, the sparse gradient descent
algorithm, etc. Finally, we study several online clustering algorithms,
including re-clustering, sequential clustering, incremental clustering and
evolutionary clustering, etc.

Inspired from the relatively sophisticated online prediction algorithms,
some possible research points for online clustering are also discussed.


Date:     		Monday, 30 March 2009

Time:                   2:00p.m.-4:00p.m.

Venue:                  Room 3501
 			lifts 25-26

Committee Members:      Dr. James Kwok (Supervisor)
 			Prof. Dit-Yan Yeung (Chairperson)
 			Prof. Qiang Yang
 			Dr. Nevin Zhang


**** ALL are Welcome ****