Decontamination of Mutually Contaminated Models
International audience A variety of machine learning problems are characterized by data sets that are drawn from multiple different convex combinations of a fixed set of base distributions. We call this a mutual contamination model. In such problems, it is often of interest to recover these base dis...
Main Authors: | , |
---|---|
Other Authors: | , , , , |
Format: | Conference Object |
Language: | English |
Published: |
HAL CCSD
2014
|
Subjects: | |
Online Access: | https://hal.archives-ouvertes.fr/hal-03371264 https://hal.archives-ouvertes.fr/hal-03371264/document https://hal.archives-ouvertes.fr/hal-03371264/file/blanchard14-supp-pdfjam.pdf |
id |
ftccsdartic:oai:HAL:hal-03371264v1 |
---|---|
record_format |
openpolar |
spelling |
ftccsdartic:oai:HAL:hal-03371264v1 2023-05-15T16:50:19+02:00 Decontamination of Mutually Contaminated Models Blanchard, Gilles Scott, Clayton Institut für Mathematik Potsdam Universität Potsdam University of Michigan Ann Arbor University of Michigan System Samuel Kaski, Jukka Corander Reykjavik, Iceland 2014 https://hal.archives-ouvertes.fr/hal-03371264 https://hal.archives-ouvertes.fr/hal-03371264/document https://hal.archives-ouvertes.fr/hal-03371264/file/blanchard14-supp-pdfjam.pdf en eng HAL CCSD hal-03371264 https://hal.archives-ouvertes.fr/hal-03371264 https://hal.archives-ouvertes.fr/hal-03371264/document https://hal.archives-ouvertes.fr/hal-03371264/file/blanchard14-supp-pdfjam.pdf http://creativecommons.org/licenses/by/ info:eu-repo/semantics/OpenAccess Seventeenth International Conference on Artificial Intelligence and Statistics (AISTATS 2014) https://hal.archives-ouvertes.fr/hal-03371264 Seventeenth International Conference on Artificial Intelligence and Statistics (AISTATS 2014), 2014, Reykjavik, Iceland. pp.1-9 https://proceedings.mlr.press/v33/blanchard14.html [INFO.INFO-AI]Computer Science [cs]/Artificial Intelligence [cs.AI] [INFO.INFO-LG]Computer Science [cs]/Machine Learning [cs.LG] [MATH.MATH-ST]Mathematics [math]/Statistics [math.ST] info:eu-repo/semantics/conferenceObject Conference papers 2014 ftccsdartic 2021-10-16T22:24:24Z International audience A variety of machine learning problems are characterized by data sets that are drawn from multiple different convex combinations of a fixed set of base distributions. We call this a mutual contamination model. In such problems, it is often of interest to recover these base distributions, or otherwise discern their properties. This work focuses on the problem of classification with multiclass label noise, in a general setting where the noise proportions are unknown and the true class distributions are nonseparable and potentially quite complex. We develop a procedure for decontamination of the contaminated models from data, which then facilitates the design of a consistent discrimination rule. Our approach relies on a novel method for estimating the error when projecting one distribution onto a convex combination of others, where the projection is with respect to a statistical distance known as the separation distance. Under sufficient conditions on the amount of noise and purity of the base distributions, this projection procedure successfully recovers the underlying class distributions. Connections to novelty detection, topic modeling, and other learning problems are also discussed. Conference Object Iceland Archive ouverte HAL (Hyper Article en Ligne, CCSD - Centre pour la Communication Scientifique Directe) |
institution |
Open Polar |
collection |
Archive ouverte HAL (Hyper Article en Ligne, CCSD - Centre pour la Communication Scientifique Directe) |
op_collection_id |
ftccsdartic |
language |
English |
topic |
[INFO.INFO-AI]Computer Science [cs]/Artificial Intelligence [cs.AI] [INFO.INFO-LG]Computer Science [cs]/Machine Learning [cs.LG] [MATH.MATH-ST]Mathematics [math]/Statistics [math.ST] |
spellingShingle |
[INFO.INFO-AI]Computer Science [cs]/Artificial Intelligence [cs.AI] [INFO.INFO-LG]Computer Science [cs]/Machine Learning [cs.LG] [MATH.MATH-ST]Mathematics [math]/Statistics [math.ST] Blanchard, Gilles Scott, Clayton Decontamination of Mutually Contaminated Models |
topic_facet |
[INFO.INFO-AI]Computer Science [cs]/Artificial Intelligence [cs.AI] [INFO.INFO-LG]Computer Science [cs]/Machine Learning [cs.LG] [MATH.MATH-ST]Mathematics [math]/Statistics [math.ST] |
description |
International audience A variety of machine learning problems are characterized by data sets that are drawn from multiple different convex combinations of a fixed set of base distributions. We call this a mutual contamination model. In such problems, it is often of interest to recover these base distributions, or otherwise discern their properties. This work focuses on the problem of classification with multiclass label noise, in a general setting where the noise proportions are unknown and the true class distributions are nonseparable and potentially quite complex. We develop a procedure for decontamination of the contaminated models from data, which then facilitates the design of a consistent discrimination rule. Our approach relies on a novel method for estimating the error when projecting one distribution onto a convex combination of others, where the projection is with respect to a statistical distance known as the separation distance. Under sufficient conditions on the amount of noise and purity of the base distributions, this projection procedure successfully recovers the underlying class distributions. Connections to novelty detection, topic modeling, and other learning problems are also discussed. |
author2 |
Institut für Mathematik Potsdam Universität Potsdam University of Michigan Ann Arbor University of Michigan System Samuel Kaski, Jukka Corander |
format |
Conference Object |
author |
Blanchard, Gilles Scott, Clayton |
author_facet |
Blanchard, Gilles Scott, Clayton |
author_sort |
Blanchard, Gilles |
title |
Decontamination of Mutually Contaminated Models |
title_short |
Decontamination of Mutually Contaminated Models |
title_full |
Decontamination of Mutually Contaminated Models |
title_fullStr |
Decontamination of Mutually Contaminated Models |
title_full_unstemmed |
Decontamination of Mutually Contaminated Models |
title_sort |
decontamination of mutually contaminated models |
publisher |
HAL CCSD |
publishDate |
2014 |
url |
https://hal.archives-ouvertes.fr/hal-03371264 https://hal.archives-ouvertes.fr/hal-03371264/document https://hal.archives-ouvertes.fr/hal-03371264/file/blanchard14-supp-pdfjam.pdf |
op_coverage |
Reykjavik, Iceland |
genre |
Iceland |
genre_facet |
Iceland |
op_source |
Seventeenth International Conference on Artificial Intelligence and Statistics (AISTATS 2014) https://hal.archives-ouvertes.fr/hal-03371264 Seventeenth International Conference on Artificial Intelligence and Statistics (AISTATS 2014), 2014, Reykjavik, Iceland. pp.1-9 https://proceedings.mlr.press/v33/blanchard14.html |
op_relation |
hal-03371264 https://hal.archives-ouvertes.fr/hal-03371264 https://hal.archives-ouvertes.fr/hal-03371264/document https://hal.archives-ouvertes.fr/hal-03371264/file/blanchard14-supp-pdfjam.pdf |
op_rights |
http://creativecommons.org/licenses/by/ info:eu-repo/semantics/OpenAccess |
_version_ |
1766040489336766464 |