Active Learning using Dirichlet Processes for Rare Class Discovery and Classification

Research output: Contribution to conferencePaper

7 Citations (Scopus)
54 Downloads (Pure)

Abstract

Real-world classification problems, such as visual surveillance and network intrusion detection, often contain common yet uninteresting background classes and rare but interesting classes, that need to be both discovered and classified. Active learning offers a suitable solution to joint rare class discovery and classification, by minimising the manual labelling of training data. A novel active learning approach is proposed, which automatically balances the competing goals of new class discovery and improving classification. Crucially it is free of tuneable parameters. Using Dirichlet processes a new active learning criterion is formulated, based on first computing the probability that unlabelled exemplars are from a new class, in addition to existing classes, and subsequently the probability of misclassification, which is then used for query selection. The proposed approach works with any probabilistic classification model and its effectiveness is demonstrated on multiple problems.
Original languageEnglish
Publication statusPublished - 2011
EventBritish Machine Vision Conference - University of Dundee, Dundee, UK United Kingdom
Duration: 29 Aug 20112 Sep 2011
Conference number: 22

Conference

ConferenceBritish Machine Vision Conference
Abbreviated titleBMVC
CountryUK United Kingdom
CityDundee
Period29/08/112/09/11

Fingerprint

Intrusion detection
Labeling
Problem-Based Learning

Cite this

Fincham Haines, T., & Xiang, T. (2011). Active Learning using Dirichlet Processes for Rare Class Discovery and Classification. Paper presented at British Machine Vision Conference, Dundee, UK United Kingdom.

Active Learning using Dirichlet Processes for Rare Class Discovery and Classification. / Fincham Haines, Tom; Xiang, Tao.

2011. Paper presented at British Machine Vision Conference, Dundee, UK United Kingdom.

Research output: Contribution to conferencePaper

Fincham Haines, T & Xiang, T 2011, 'Active Learning using Dirichlet Processes for Rare Class Discovery and Classification' Paper presented at British Machine Vision Conference, Dundee, UK United Kingdom, 29/08/11 - 2/09/11, .
Fincham Haines T, Xiang T. Active Learning using Dirichlet Processes for Rare Class Discovery and Classification. 2011. Paper presented at British Machine Vision Conference, Dundee, UK United Kingdom.
Fincham Haines, Tom ; Xiang, Tao. / Active Learning using Dirichlet Processes for Rare Class Discovery and Classification. Paper presented at British Machine Vision Conference, Dundee, UK United Kingdom.
@conference{13b9092dccc8481bbd0767a9b824865a,
title = "Active Learning using Dirichlet Processes for Rare Class Discovery and Classification",
abstract = "Real-world classification problems, such as visual surveillance and network intrusion detection, often contain common yet uninteresting background classes and rare but interesting classes, that need to be both discovered and classified. Active learning offers a suitable solution to joint rare class discovery and classification, by minimising the manual labelling of training data. A novel active learning approach is proposed, which automatically balances the competing goals of new class discovery and improving classification. Crucially it is free of tuneable parameters. Using Dirichlet processes a new active learning criterion is formulated, based on first computing the probability that unlabelled exemplars are from a new class, in addition to existing classes, and subsequently the probability of misclassification, which is then used for query selection. The proposed approach works with any probabilistic classification model and its effectiveness is demonstrated on multiple problems.",
author = "{Fincham Haines}, Tom and Tao Xiang",
year = "2011",
language = "English",
note = "British Machine Vision Conference, BMVC ; Conference date: 29-08-2011 Through 02-09-2011",

}

TY - CONF

T1 - Active Learning using Dirichlet Processes for Rare Class Discovery and Classification

AU - Fincham Haines, Tom

AU - Xiang, Tao

PY - 2011

Y1 - 2011

N2 - Real-world classification problems, such as visual surveillance and network intrusion detection, often contain common yet uninteresting background classes and rare but interesting classes, that need to be both discovered and classified. Active learning offers a suitable solution to joint rare class discovery and classification, by minimising the manual labelling of training data. A novel active learning approach is proposed, which automatically balances the competing goals of new class discovery and improving classification. Crucially it is free of tuneable parameters. Using Dirichlet processes a new active learning criterion is formulated, based on first computing the probability that unlabelled exemplars are from a new class, in addition to existing classes, and subsequently the probability of misclassification, which is then used for query selection. The proposed approach works with any probabilistic classification model and its effectiveness is demonstrated on multiple problems.

AB - Real-world classification problems, such as visual surveillance and network intrusion detection, often contain common yet uninteresting background classes and rare but interesting classes, that need to be both discovered and classified. Active learning offers a suitable solution to joint rare class discovery and classification, by minimising the manual labelling of training data. A novel active learning approach is proposed, which automatically balances the competing goals of new class discovery and improving classification. Crucially it is free of tuneable parameters. Using Dirichlet processes a new active learning criterion is formulated, based on first computing the probability that unlabelled exemplars are from a new class, in addition to existing classes, and subsequently the probability of misclassification, which is then used for query selection. The proposed approach works with any probabilistic classification model and its effectiveness is demonstrated on multiple problems.

M3 - Paper

ER -