Gradient Ascent for Best Response Regression

Victoria Racher; Christian Borgelt

doi:10.1007/978-3-030-74251-5_12

Gradient Ascent for Best Response Regression

Victoria Racher, Christian Borgelt

Artificial Intelligence and Human Interfaces

Publikation: Beitrag in Buch/Bericht/Konferenzband/Gesetzeskommentar › Konferenzbeitrag › Peer-reviewed

Abstract

Although regression is among the oldest areas of statistics, new approaches may still be found. One recent suggestion is Best Response Regression, where one tries to find a regression function that provides, for as many instances as possible, a better prediction than some reference regression function. In this paper we propose a new method for Best Response Regression that is based on gradient ascent rather than mixed integer programming. We evaluate our approach for a variety of noise (or error) distributions, showing that especially for heavy-tailed distributions best response regression outperforms, on unseen data, ordinary least squares regression, both w.r.t. the sum of squared errors as well as the number of instances for which better predictions are provided.

Originalsprache	Englisch
Titel	Advances in Intelligent Data Analysis XIX
Untertitel	International Symposium on Intelligent Data Analysis
Redakteure/-innen	Pedro Henriques Abreu, Pedro Pereira Rodrigues, Alberto Fernández, João Gama
Erscheinungsort	Heidelberg / Berlin
Herausgeber (Verlag)	Springer Verlag
Kapitel	12
Seiten	141-154
Seitenumfang	14
Band	LNCS 12695
ISBN (elektronisch)	978-3-030-74251-5
ISBN (Print)	978-3-030-74250-8
DOIs	https://doi.org/10.1007/978-3-030-74251-5_12
Publikationsstatus	Veröffentlicht - 13 Apr. 2021
Veranstaltung	International Symposium on Intelligent Data Analysis 2021 - Online, Porto, Portugal Dauer: 26 Apr. 2021 → 26 Apr. 2021 https://ida2021.org/

Publikationsreihe

Name	Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Band	12695 LNCS
ISSN (Print)	0302-9743
ISSN (elektronisch)	1611-3349

Online-Konferenz

Online-Konferenz	International Symposium on Intelligent Data Analysis 2021
Kurztitel	IDA 2021
Land/Gebiet	Portugal
Ort	Porto
Zeitraum	26/04/21 → 26/04/21
Internetadresse	https://ida2021.org/

Bibliographische Notiz

Funding Information:
Acknowledgments. The second author gratefully acknowledges the financial support from Land Salzburg within the WISS 2025 project IDA-Lab (20102-F1901166-KZP and 20204-WISS/225/197-2019).

Publisher Copyright:
© 2021, Springer Nature Switzerland AG.

Systematik der Wissenschaftszweige 2012

102 Informatik

Zugriff auf Dokument

10.1007/978-3-030-74251-5_12

Andere Dateien und Links

Dieses zitieren

Racher, V., & Borgelt, C. (2021). Gradient Ascent for Best Response Regression. in P. H. Abreu, P. P. Rodrigues, A. Fernández, & J. Gama (Hrsg.), Advances in Intelligent Data Analysis XIX: International Symposium on Intelligent Data Analysis (Band LNCS 12695, S. 141-154). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Band 12695 LNCS). Springer Verlag. https://doi.org/10.1007/978-3-030-74251-5_12

Racher, Victoria ; Borgelt, Christian. / Gradient Ascent for Best Response Regression. Advances in Intelligent Data Analysis XIX: International Symposium on Intelligent Data Analysis. Hrsg. / Pedro Henriques Abreu ; Pedro Pereira Rodrigues ; Alberto Fernández ; João Gama. Band LNCS 12695 Heidelberg / Berlin : Springer Verlag, 2021. S. 141-154 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).

@inproceedings{945dd111db3d4d67935504a31f9ef1b7,

title = "Gradient Ascent for Best Response Regression",

abstract = "Although regression is among the oldest areas of statistics, new approaches may still be found. One recent suggestion is Best Response Regression, where one tries to find a regression function that provides, for as many instances as possible, a better prediction than some reference regression function. In this paper we propose a new method for Best Response Regression that is based on gradient ascent rather than mixed integer programming. We evaluate our approach for a variety of noise (or error) distributions, showing that especially for heavy-tailed distributions best response regression outperforms, on unseen data, ordinary least squares regression, both w.r.t. the sum of squared errors as well as the number of instances for which better predictions are provided.",

keywords = "Best response regression, Gradient ascent, Newton–Raphson method, Objective function smoothing, Resilient backpropagation, Newton-Raphson method",

author = "Victoria Racher and Christian Borgelt",

note = "Publisher Copyright: {\textcopyright} 2021, Springer Nature Switzerland AG.; International Symposium on Intelligent Data Analysis 2021, IDA 2021 ; Conference date: 26-04-2021 Through 26-04-2021",

year = "2021",

month = apr,

day = "13",

doi = "10.1007/978-3-030-74251-5_12",

language = "English",

isbn = "978-3-030-74250-8",

volume = "LNCS 12695",

series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",

publisher = "Springer Verlag",

pages = "141--154",

editor = "Abreu, {Pedro Henriques} and Rodrigues, {Pedro Pereira} and Alberto Fern{\'a}ndez and Jo{\~a}o Gama",

booktitle = "Advances in Intelligent Data Analysis XIX",

url = "https://ida2021.org/",

}

Racher, V & Borgelt, C 2021, Gradient Ascent for Best Response Regression. in PH Abreu, PP Rodrigues, A Fernández & J Gama (Hrsg.), Advances in Intelligent Data Analysis XIX: International Symposium on Intelligent Data Analysis. Bd. LNCS 12695, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), Bd. 12695 LNCS, Springer Verlag, Heidelberg / Berlin, S. 141-154, International Symposium on Intelligent Data Analysis 2021, Porto, Portugal, 26/04/21. https://doi.org/10.1007/978-3-030-74251-5_12

Gradient Ascent for Best Response Regression. / Racher, Victoria; Borgelt, Christian.
Advances in Intelligent Data Analysis XIX: International Symposium on Intelligent Data Analysis. Hrsg. / Pedro Henriques Abreu; Pedro Pereira Rodrigues; Alberto Fernández; João Gama. Band LNCS 12695 Heidelberg / Berlin: Springer Verlag, 2021. S. 141-154 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Band 12695 LNCS).

Publikation: Beitrag in Buch/Bericht/Konferenzband/Gesetzeskommentar › Konferenzbeitrag › Peer-reviewed

TY - GEN

T1 - Gradient Ascent for Best Response Regression

AU - Racher, Victoria

AU - Borgelt, Christian

PY - 2021/4/13

Y1 - 2021/4/13

N2 - Although regression is among the oldest areas of statistics, new approaches may still be found. One recent suggestion is Best Response Regression, where one tries to find a regression function that provides, for as many instances as possible, a better prediction than some reference regression function. In this paper we propose a new method for Best Response Regression that is based on gradient ascent rather than mixed integer programming. We evaluate our approach for a variety of noise (or error) distributions, showing that especially for heavy-tailed distributions best response regression outperforms, on unseen data, ordinary least squares regression, both w.r.t. the sum of squared errors as well as the number of instances for which better predictions are provided.

AB - Although regression is among the oldest areas of statistics, new approaches may still be found. One recent suggestion is Best Response Regression, where one tries to find a regression function that provides, for as many instances as possible, a better prediction than some reference regression function. In this paper we propose a new method for Best Response Regression that is based on gradient ascent rather than mixed integer programming. We evaluate our approach for a variety of noise (or error) distributions, showing that especially for heavy-tailed distributions best response regression outperforms, on unseen data, ordinary least squares regression, both w.r.t. the sum of squared errors as well as the number of instances for which better predictions are provided.

KW - Best response regression

KW - Gradient ascent

KW - Newton–Raphson method

KW - Objective function smoothing

KW - Resilient backpropagation

KW - Newton-Raphson method

UR - http://www.scopus.com/inward/record.url?scp=85105865561&partnerID=8YFLogxK

UR - https://www.mendeley.com/catalogue/2a043f54-335a-35aa-82d5-772dab3f160e/

U2 - 10.1007/978-3-030-74251-5_12

DO - 10.1007/978-3-030-74251-5_12

M3 - Conference contribution

SN - 978-3-030-74250-8

VL - LNCS 12695

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 141

EP - 154

BT - Advances in Intelligent Data Analysis XIX

A2 - Abreu, Pedro Henriques

A2 - Rodrigues, Pedro Pereira

A2 - Fernández, Alberto

A2 - Gama, João

PB - Springer Verlag

CY - Heidelberg / Berlin

T2 - International Symposium on Intelligent Data Analysis 2021

Y2 - 26 April 2021 through 26 April 2021

ER -

Racher V, Borgelt C. Gradient Ascent for Best Response Regression. in Abreu PH, Rodrigues PP, Fernández A, Gama J, Hrsg., Advances in Intelligent Data Analysis XIX: International Symposium on Intelligent Data Analysis. Band LNCS 12695. Heidelberg / Berlin: Springer Verlag. 2021. S. 141-154. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). doi: 10.1007/978-3-030-74251-5_12