Towards low-cost, high-accuracy classifiers for linear solver selection

Sanjukta Bhowmick, Brice Toth, Padma Raghavan

Research output: Chapter in Book/Report/Conference proceedingConference contribution

7 Citations (Scopus)

Abstract

The time to solve linear systems depends to a large extent on the choice of the solution method and the properties of the coefficient matrix. Although there are several linear solution methods, in most cases it is impossible to predict apriori which linear solver would be best suited for a given linear system. Recent investigations on selecting linear solvers for a given system have explored the use of classification techniques based on the linear system parameters for solver selection. In this paper, we present a method to develop low-cost high-accuracy classifiers. We show that the cost for constructing a classifier can be significantly reduced by focusing on the computational complexity of each feature. In particular, we filter out low information linear system parameters and then order the remaining parameters to decrease the total computation cost for classification at a prescribed accuracy. Our results indicate that the speedup factor of the time to compute the feature set using our ordering can be as high as 262. The accuracy and computation time of the feature set generated using our method is comparable to a near-optimal one, thus demonstrating the effectiveness of our technique.

Original languageEnglish (US)
Title of host publicationComputational Science - ICCS 2009 - 9th International Conference, Proceedings
Pages463-472
Number of pages10
EditionPART 1
DOIs
StatePublished - Aug 21 2009
Event9th International Conference on Computational Science, ICCS 2009 - Baton Rouge, LA, United States
Duration: May 25 2009May 27 2009

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
NumberPART 1
Volume5544 LNCS
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Conference

Conference9th International Conference on Computational Science, ICCS 2009
CountryUnited States
CityBaton Rouge, LA
Period5/25/095/27/09

Fingerprint

Linear systems
High Accuracy
Classifiers
Linear Systems
Classifier
Costs
Information Systems
Computational complexity
Computational Complexity
Speedup
Filter
Predict
Decrease
Coefficient

ASJC Scopus subject areas

  • Theoretical Computer Science
  • Computer Science(all)

Cite this

Bhowmick, S., Toth, B., & Raghavan, P. (2009). Towards low-cost, high-accuracy classifiers for linear solver selection. In Computational Science - ICCS 2009 - 9th International Conference, Proceedings (PART 1 ed., pp. 463-472). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 5544 LNCS, No. PART 1). https://doi.org/10.1007/978-3-642-01970-8_45

Towards low-cost, high-accuracy classifiers for linear solver selection. / Bhowmick, Sanjukta; Toth, Brice; Raghavan, Padma.

Computational Science - ICCS 2009 - 9th International Conference, Proceedings. PART 1. ed. 2009. p. 463-472 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 5544 LNCS, No. PART 1).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Bhowmick, S, Toth, B & Raghavan, P 2009, Towards low-cost, high-accuracy classifiers for linear solver selection. in Computational Science - ICCS 2009 - 9th International Conference, Proceedings. PART 1 edn, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), no. PART 1, vol. 5544 LNCS, pp. 463-472, 9th International Conference on Computational Science, ICCS 2009, Baton Rouge, LA, United States, 5/25/09. https://doi.org/10.1007/978-3-642-01970-8_45
Bhowmick S, Toth B, Raghavan P. Towards low-cost, high-accuracy classifiers for linear solver selection. In Computational Science - ICCS 2009 - 9th International Conference, Proceedings. PART 1 ed. 2009. p. 463-472. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 1). https://doi.org/10.1007/978-3-642-01970-8_45
Bhowmick, Sanjukta ; Toth, Brice ; Raghavan, Padma. / Towards low-cost, high-accuracy classifiers for linear solver selection. Computational Science - ICCS 2009 - 9th International Conference, Proceedings. PART 1. ed. 2009. pp. 463-472 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); PART 1).
@inproceedings{ab6ef0ea8262489abbab7ec6f1ca725b,
title = "Towards low-cost, high-accuracy classifiers for linear solver selection",
abstract = "The time to solve linear systems depends to a large extent on the choice of the solution method and the properties of the coefficient matrix. Although there are several linear solution methods, in most cases it is impossible to predict apriori which linear solver would be best suited for a given linear system. Recent investigations on selecting linear solvers for a given system have explored the use of classification techniques based on the linear system parameters for solver selection. In this paper, we present a method to develop low-cost high-accuracy classifiers. We show that the cost for constructing a classifier can be significantly reduced by focusing on the computational complexity of each feature. In particular, we filter out low information linear system parameters and then order the remaining parameters to decrease the total computation cost for classification at a prescribed accuracy. Our results indicate that the speedup factor of the time to compute the feature set using our ordering can be as high as 262. The accuracy and computation time of the feature set generated using our method is comparable to a near-optimal one, thus demonstrating the effectiveness of our technique.",
author = "Sanjukta Bhowmick and Brice Toth and Padma Raghavan",
year = "2009",
month = "8",
day = "21",
doi = "10.1007/978-3-642-01970-8_45",
language = "English (US)",
isbn = "3642019692",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
number = "PART 1",
pages = "463--472",
booktitle = "Computational Science - ICCS 2009 - 9th International Conference, Proceedings",
edition = "PART 1",

}

TY - GEN

T1 - Towards low-cost, high-accuracy classifiers for linear solver selection

AU - Bhowmick, Sanjukta

AU - Toth, Brice

AU - Raghavan, Padma

PY - 2009/8/21

Y1 - 2009/8/21

N2 - The time to solve linear systems depends to a large extent on the choice of the solution method and the properties of the coefficient matrix. Although there are several linear solution methods, in most cases it is impossible to predict apriori which linear solver would be best suited for a given linear system. Recent investigations on selecting linear solvers for a given system have explored the use of classification techniques based on the linear system parameters for solver selection. In this paper, we present a method to develop low-cost high-accuracy classifiers. We show that the cost for constructing a classifier can be significantly reduced by focusing on the computational complexity of each feature. In particular, we filter out low information linear system parameters and then order the remaining parameters to decrease the total computation cost for classification at a prescribed accuracy. Our results indicate that the speedup factor of the time to compute the feature set using our ordering can be as high as 262. The accuracy and computation time of the feature set generated using our method is comparable to a near-optimal one, thus demonstrating the effectiveness of our technique.

AB - The time to solve linear systems depends to a large extent on the choice of the solution method and the properties of the coefficient matrix. Although there are several linear solution methods, in most cases it is impossible to predict apriori which linear solver would be best suited for a given linear system. Recent investigations on selecting linear solvers for a given system have explored the use of classification techniques based on the linear system parameters for solver selection. In this paper, we present a method to develop low-cost high-accuracy classifiers. We show that the cost for constructing a classifier can be significantly reduced by focusing on the computational complexity of each feature. In particular, we filter out low information linear system parameters and then order the remaining parameters to decrease the total computation cost for classification at a prescribed accuracy. Our results indicate that the speedup factor of the time to compute the feature set using our ordering can be as high as 262. The accuracy and computation time of the feature set generated using our method is comparable to a near-optimal one, thus demonstrating the effectiveness of our technique.

UR - http://www.scopus.com/inward/record.url?scp=68849090240&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=68849090240&partnerID=8YFLogxK

U2 - 10.1007/978-3-642-01970-8_45

DO - 10.1007/978-3-642-01970-8_45

M3 - Conference contribution

SN - 3642019692

SN - 9783642019692

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

SP - 463

EP - 472

BT - Computational Science - ICCS 2009 - 9th International Conference, Proceedings

ER -