Prediction with Expert Evaluators' Advice

Alexey Chernov, Vladimir Vovk

Research output: Chapter in Book/Conference proceeding with ISSN or ISBNConference contribution with ISSN or ISBNResearchpeer-review

Abstract

We introduce a new protocol for prediction with expert advice in which each expert evaluates the learner’s and his own performance using a loss function that may change over time and may be different from the loss functions used by the other experts. The learner’s goal is to perform better or not much worse than each expert, as evaluated by that expert, for all experts simultaneously. If the loss functions used by the experts are all proper scoring rules and all mixable, we show that the defensive forecasting algorithm enjoys the same performance guaranteeas that attainable by the Aggregating Algorithm in the standard setting and known to be optimal. This result is also applied to the case of “specialist” experts. In this case, the defensive forecasting algorithm reduces to a simple modification of the Aggregating Algorithm.
Original languageEnglish
Title of host publication20th International Conference, ALT 2009
Place of PublicationBerlin
Pages8-22
Number of pages15
Volume5809
DOIs
Publication statusPublished - 31 Dec 2009
Event20th International Conference, ALT 2009 - Porto, Portugal, October 3-5, 2009
Duration: 31 Dec 2009 → …

Publication series

NameLecture Notes in Computer Science

Conference

Conference20th International Conference, ALT 2009
Period31/12/09 → …

Fingerprint

Evaluator
Prediction
Loss function
Scoring rules

Bibliographical note

© Springer-Verlag Berlin Heidelberg 2009

Cite this

Chernov, A., & Vovk, V. (2009). Prediction with Expert Evaluators' Advice. In 20th International Conference, ALT 2009 (Vol. 5809, pp. 8-22). (Lecture Notes in Computer Science). Berlin. https://doi.org/10.1007/978-3-642-04414-4_6
Chernov, Alexey ; Vovk, Vladimir. / Prediction with Expert Evaluators' Advice. 20th International Conference, ALT 2009. Vol. 5809 Berlin, 2009. pp. 8-22 (Lecture Notes in Computer Science).
@inproceedings{407d6d832c3f4f4fb9b86f7e1d4c99f7,
title = "Prediction with Expert Evaluators' Advice",
abstract = "We introduce a new protocol for prediction with expert advice in which each expert evaluates the learner’s and his own performance using a loss function that may change over time and may be different from the loss functions used by the other experts. The learner’s goal is to perform better or not much worse than each expert, as evaluated by that expert, for all experts simultaneously. If the loss functions used by the experts are all proper scoring rules and all mixable, we show that the defensive forecasting algorithm enjoys the same performance guaranteeas that attainable by the Aggregating Algorithm in the standard setting and known to be optimal. This result is also applied to the case of “specialist” experts. In this case, the defensive forecasting algorithm reduces to a simple modification of the Aggregating Algorithm.",
author = "Alexey Chernov and Vladimir Vovk",
note = "{\circledC} Springer-Verlag Berlin Heidelberg 2009",
year = "2009",
month = "12",
day = "31",
doi = "10.1007/978-3-642-04414-4_6",
language = "English",
volume = "5809",
series = "Lecture Notes in Computer Science",
pages = "8--22",
booktitle = "20th International Conference, ALT 2009",

}

Chernov, A & Vovk, V 2009, Prediction with Expert Evaluators' Advice. in 20th International Conference, ALT 2009. vol. 5809, Lecture Notes in Computer Science, Berlin, pp. 8-22, 20th International Conference, ALT 2009, 31/12/09. https://doi.org/10.1007/978-3-642-04414-4_6

Prediction with Expert Evaluators' Advice. / Chernov, Alexey; Vovk, Vladimir.

20th International Conference, ALT 2009. Vol. 5809 Berlin, 2009. p. 8-22 (Lecture Notes in Computer Science).

Research output: Chapter in Book/Conference proceeding with ISSN or ISBNConference contribution with ISSN or ISBNResearchpeer-review

TY - GEN

T1 - Prediction with Expert Evaluators' Advice

AU - Chernov, Alexey

AU - Vovk, Vladimir

N1 - © Springer-Verlag Berlin Heidelberg 2009

PY - 2009/12/31

Y1 - 2009/12/31

N2 - We introduce a new protocol for prediction with expert advice in which each expert evaluates the learner’s and his own performance using a loss function that may change over time and may be different from the loss functions used by the other experts. The learner’s goal is to perform better or not much worse than each expert, as evaluated by that expert, for all experts simultaneously. If the loss functions used by the experts are all proper scoring rules and all mixable, we show that the defensive forecasting algorithm enjoys the same performance guaranteeas that attainable by the Aggregating Algorithm in the standard setting and known to be optimal. This result is also applied to the case of “specialist” experts. In this case, the defensive forecasting algorithm reduces to a simple modification of the Aggregating Algorithm.

AB - We introduce a new protocol for prediction with expert advice in which each expert evaluates the learner’s and his own performance using a loss function that may change over time and may be different from the loss functions used by the other experts. The learner’s goal is to perform better or not much worse than each expert, as evaluated by that expert, for all experts simultaneously. If the loss functions used by the experts are all proper scoring rules and all mixable, we show that the defensive forecasting algorithm enjoys the same performance guaranteeas that attainable by the Aggregating Algorithm in the standard setting and known to be optimal. This result is also applied to the case of “specialist” experts. In this case, the defensive forecasting algorithm reduces to a simple modification of the Aggregating Algorithm.

U2 - 10.1007/978-3-642-04414-4_6

DO - 10.1007/978-3-642-04414-4_6

M3 - Conference contribution with ISSN or ISBN

VL - 5809

T3 - Lecture Notes in Computer Science

SP - 8

EP - 22

BT - 20th International Conference, ALT 2009

CY - Berlin

ER -

Chernov A, Vovk V. Prediction with Expert Evaluators' Advice. In 20th International Conference, ALT 2009. Vol. 5809. Berlin. 2009. p. 8-22. (Lecture Notes in Computer Science). https://doi.org/10.1007/978-3-642-04414-4_6