Abstract

The previous chapters demonstrated how the Conformal Predictions (CP) framework can be adapted to traditional machine learning problems including active learning, feature selection, anomaly detection, change detection, model selection, and quality estimation. In this chapter, we describe three other adaptations of the CP framework, each of which is non-traditional in its own way. The task of obtaining a reliability value for the classification of a data instance has been the focus of a number of studies. In Sections 9.2 and 9.3, we describe two methods that use the idea of a metaclassifier to associate reliability values with output predictions from a base classifier. In particular, in Section 9.2, we describe the Metaconformal Predictors, where a base classifier is combined with a metaclassifier that is trained on metadata generated from the data instances and the classification results of the base classifier to associate reliability values on the classification of data instances. In Section 9.3, we describe the Single-Stacking Conformal Predictors, where an ensemble classifier consisting of the base classifier and the metaclassifier is constructed to compute reliability values on the classification outputs. The difference between the metaconformal and the single-stacking approaches is the manner in which the metadata are constructed and the way in which the reliability values are estimated.

Original languageEnglish (US)
Title of host publicationConformal Prediction for Reliable Machine Learning: Theory, Adaptations and Applications
PublisherElsevier Inc.
Pages167-185
Number of pages19
ISBN (Print)9780123985378
DOIs
StatePublished - Apr 2014

Fingerprint

Classifiers
Metadata
Learning systems
Feature extraction

Keywords

  • Ensemble Classifiers
  • Metaclassification
  • Metaconformal Predictors
  • ROC Curves
  • Single-Stacking Conformal Predictors
  • Time Series Analysis

ASJC Scopus subject areas

  • Computer Science(all)

Cite this

Balasubramanian, V. N., Lade, P., Demakethepalli Venkateswara, H., Smirnov, E., & Panchanathan, S. (2014). Other Adaptations. In Conformal Prediction for Reliable Machine Learning: Theory, Adaptations and Applications (pp. 167-185). Elsevier Inc.. https://doi.org/10.1016/B978-0-12-398537-8.00009-2

Other Adaptations. / Balasubramanian, Vineeth N.; Lade, Prasanth; Demakethepalli Venkateswara, Hemanth; Smirnov, Evgueni; Panchanathan, Sethuraman.

Conformal Prediction for Reliable Machine Learning: Theory, Adaptations and Applications. Elsevier Inc., 2014. p. 167-185.

Research output: Chapter in Book/Report/Conference proceedingChapter

Balasubramanian, VN, Lade, P, Demakethepalli Venkateswara, H, Smirnov, E & Panchanathan, S 2014, Other Adaptations. in Conformal Prediction for Reliable Machine Learning: Theory, Adaptations and Applications. Elsevier Inc., pp. 167-185. https://doi.org/10.1016/B978-0-12-398537-8.00009-2
Balasubramanian VN, Lade P, Demakethepalli Venkateswara H, Smirnov E, Panchanathan S. Other Adaptations. In Conformal Prediction for Reliable Machine Learning: Theory, Adaptations and Applications. Elsevier Inc. 2014. p. 167-185 https://doi.org/10.1016/B978-0-12-398537-8.00009-2
Balasubramanian, Vineeth N. ; Lade, Prasanth ; Demakethepalli Venkateswara, Hemanth ; Smirnov, Evgueni ; Panchanathan, Sethuraman. / Other Adaptations. Conformal Prediction for Reliable Machine Learning: Theory, Adaptations and Applications. Elsevier Inc., 2014. pp. 167-185
@inbook{97f52b2894814fada97e9fc5c28f75e7,
title = "Other Adaptations",
abstract = "The previous chapters demonstrated how the Conformal Predictions (CP) framework can be adapted to traditional machine learning problems including active learning, feature selection, anomaly detection, change detection, model selection, and quality estimation. In this chapter, we describe three other adaptations of the CP framework, each of which is non-traditional in its own way. The task of obtaining a reliability value for the classification of a data instance has been the focus of a number of studies. In Sections 9.2 and 9.3, we describe two methods that use the idea of a metaclassifier to associate reliability values with output predictions from a base classifier. In particular, in Section 9.2, we describe the Metaconformal Predictors, where a base classifier is combined with a metaclassifier that is trained on metadata generated from the data instances and the classification results of the base classifier to associate reliability values on the classification of data instances. In Section 9.3, we describe the Single-Stacking Conformal Predictors, where an ensemble classifier consisting of the base classifier and the metaclassifier is constructed to compute reliability values on the classification outputs. The difference between the metaconformal and the single-stacking approaches is the manner in which the metadata are constructed and the way in which the reliability values are estimated.",
keywords = "Ensemble Classifiers, Metaclassification, Metaconformal Predictors, ROC Curves, Single-Stacking Conformal Predictors, Time Series Analysis",
author = "Balasubramanian, {Vineeth N.} and Prasanth Lade and {Demakethepalli Venkateswara}, Hemanth and Evgueni Smirnov and Sethuraman Panchanathan",
year = "2014",
month = "4",
doi = "10.1016/B978-0-12-398537-8.00009-2",
language = "English (US)",
isbn = "9780123985378",
pages = "167--185",
booktitle = "Conformal Prediction for Reliable Machine Learning: Theory, Adaptations and Applications",
publisher = "Elsevier Inc.",
address = "United States",

}

TY - CHAP

T1 - Other Adaptations

AU - Balasubramanian, Vineeth N.

AU - Lade, Prasanth

AU - Demakethepalli Venkateswara, Hemanth

AU - Smirnov, Evgueni

AU - Panchanathan, Sethuraman

PY - 2014/4

Y1 - 2014/4

N2 - The previous chapters demonstrated how the Conformal Predictions (CP) framework can be adapted to traditional machine learning problems including active learning, feature selection, anomaly detection, change detection, model selection, and quality estimation. In this chapter, we describe three other adaptations of the CP framework, each of which is non-traditional in its own way. The task of obtaining a reliability value for the classification of a data instance has been the focus of a number of studies. In Sections 9.2 and 9.3, we describe two methods that use the idea of a metaclassifier to associate reliability values with output predictions from a base classifier. In particular, in Section 9.2, we describe the Metaconformal Predictors, where a base classifier is combined with a metaclassifier that is trained on metadata generated from the data instances and the classification results of the base classifier to associate reliability values on the classification of data instances. In Section 9.3, we describe the Single-Stacking Conformal Predictors, where an ensemble classifier consisting of the base classifier and the metaclassifier is constructed to compute reliability values on the classification outputs. The difference between the metaconformal and the single-stacking approaches is the manner in which the metadata are constructed and the way in which the reliability values are estimated.

AB - The previous chapters demonstrated how the Conformal Predictions (CP) framework can be adapted to traditional machine learning problems including active learning, feature selection, anomaly detection, change detection, model selection, and quality estimation. In this chapter, we describe three other adaptations of the CP framework, each of which is non-traditional in its own way. The task of obtaining a reliability value for the classification of a data instance has been the focus of a number of studies. In Sections 9.2 and 9.3, we describe two methods that use the idea of a metaclassifier to associate reliability values with output predictions from a base classifier. In particular, in Section 9.2, we describe the Metaconformal Predictors, where a base classifier is combined with a metaclassifier that is trained on metadata generated from the data instances and the classification results of the base classifier to associate reliability values on the classification of data instances. In Section 9.3, we describe the Single-Stacking Conformal Predictors, where an ensemble classifier consisting of the base classifier and the metaclassifier is constructed to compute reliability values on the classification outputs. The difference between the metaconformal and the single-stacking approaches is the manner in which the metadata are constructed and the way in which the reliability values are estimated.

KW - Ensemble Classifiers

KW - Metaclassification

KW - Metaconformal Predictors

KW - ROC Curves

KW - Single-Stacking Conformal Predictors

KW - Time Series Analysis

UR - http://www.scopus.com/inward/record.url?scp=84902237651&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84902237651&partnerID=8YFLogxK

U2 - 10.1016/B978-0-12-398537-8.00009-2

DO - 10.1016/B978-0-12-398537-8.00009-2

M3 - Chapter

SN - 9780123985378

SP - 167

EP - 185

BT - Conformal Prediction for Reliable Machine Learning: Theory, Adaptations and Applications

PB - Elsevier Inc.

ER -