eCommons

 

Online Learning Algorithms For Sequence Prediction, Importance Weighted Classification, And Active Learning

Other Titles

Abstract

This thesis studies three problems in online learning. For all the problems the proposed solutions are simple yet non-trivial adaptations of existing online machine learning algorithms. For the task of sequential prediction, a modified multiplicative update algorithm that produces small and accurate models is proposed. This algorithm makes no assumption about the complexity of the source that produces the given sequence. For the task of online learning when examples have varying importances, the proposed algorithm is a version of gradient descent in continuous time. Finally, for the task of efficient online active learning, the implementation we provide makes use of many shortcuts. These include replacing a batch learning algorithm with an online one, as well as a creative use of the aforementioned continuous time gradient descent to compute the desirability of asking for the label of a given example. As this thesis shows, online machine learning algorithms can be easily adapted to many new problems.

Journal / Series

Volume & Issue

Description

Sponsorship

Date Issued

2012-08-20

Publisher

Keywords

machine learning; online learning; active learning

Location

Effective Date

Expiration Date

Sector

Employer

Union

Union Local

NAICS

Number of Workers

Committee Chair

Kozen, Dexter Campbell

Committee Co-Chair

Committee Member

Hooker, Giles J.
Joachims, Thorsten
Kleinberg, Robert David

Degree Discipline

Computer Science

Degree Name

Ph. D., Computer Science

Degree Level

Doctor of Philosophy

Related Version

Related DOI

Related To

Related Part

Based on Related Item

Has Other Format(s)

Part of Related Item

Related To

Related Publication(s)

Link(s) to Related Publication(s)

References

Link(s) to Reference(s)

Previously Published As

Government Document

ISBN

ISMN

ISSN

Other Identifiers

Rights

Rights URI

Types

dissertation or thesis

Accessibility Feature

Accessibility Hazard

Accessibility Summary

Link(s) to Catalog Record