Algorithm for very fast computation of least absolute value regression

Amin Nobakhti; Hong Wang; Tianyou Chai

doi:10.1109/ACC.2009.5160229

Algorithm for very fast computation of least absolute value regression

Amin Nobakhti, Hong Wang, Tianyou Chai

Research output: Chapter in Book/Conference proceeding › Conference contribution › peer-review

Abstract

The Least Squares (LS) problem has been popular in industrial modeling applications due to its speed, efficiency and simplicity. However, the LS solution is known to be unreliable when the data distribution is not Gaussian and is flat-tailed and such data anomalies occur frequently in the industry. The Least Absolute Value (LAV) problem overcomes these difficulties but at the expense of greatly increasing the complexity of the solution. This was partly addressed when it was shown that the LAV problem could be formulated as a Linear Programme (LP). However, the LP formulation is not suitable for implementation in all types of applications. In this paper, a very fast direct search algorithm is developed to solve the general dimension LAV problem using only elementary operations. The algorithm has been shown to be significantly faster than the LP approach through several experiments. © 2009 AACC.

Original language	English
Title of host publication	Proceedings of the American Control Conference\|Proc Am Control Conf
Pages	14-19
Number of pages	5
DOIs	https://doi.org/10.1109/ACC.2009.5160229
Publication status	Published - 2009
Event	2009 American Control Conference, ACC 2009 - St. Louis, MO Duration: 1 Jul 2009 → …

Conference

Conference	2009 American Control Conference, ACC 2009
City	St. Louis, MO
Period	1/07/09 → …

Access to Document

10.1109/ACC.2009.5160229

Cite this

@inproceedings{7b24ac48e73f44a4954559298c127d89,

title = "Algorithm for very fast computation of least absolute value regression",

abstract = "The Least Squares (LS) problem has been popular in industrial modeling applications due to its speed, efficiency and simplicity. However, the LS solution is known to be unreliable when the data distribution is not Gaussian and is flat-tailed and such data anomalies occur frequently in the industry. The Least Absolute Value (LAV) problem overcomes these difficulties but at the expense of greatly increasing the complexity of the solution. This was partly addressed when it was shown that the LAV problem could be formulated as a Linear Programme (LP). However, the LP formulation is not suitable for implementation in all types of applications. In this paper, a very fast direct search algorithm is developed to solve the general dimension LAV problem using only elementary operations. The algorithm has been shown to be significantly faster than the LP approach through several experiments. {\textcopyright} 2009 AACC.",

author = "Amin Nobakhti and Hong Wang and Tianyou Chai",

year = "2009",

doi = "10.1109/ACC.2009.5160229",

language = "English",

isbn = "9781424445240",

pages = "14--19",

booktitle = "Proceedings of the American Control Conference|Proc Am Control Conf",

note = "2009 American Control Conference, ACC 2009 ; Conference date: 01-07-2009",

}

TY - GEN

T1 - Algorithm for very fast computation of least absolute value regression

AU - Nobakhti, Amin

AU - Wang, Hong

AU - Chai, Tianyou

PY - 2009

Y1 - 2009

N2 - The Least Squares (LS) problem has been popular in industrial modeling applications due to its speed, efficiency and simplicity. However, the LS solution is known to be unreliable when the data distribution is not Gaussian and is flat-tailed and such data anomalies occur frequently in the industry. The Least Absolute Value (LAV) problem overcomes these difficulties but at the expense of greatly increasing the complexity of the solution. This was partly addressed when it was shown that the LAV problem could be formulated as a Linear Programme (LP). However, the LP formulation is not suitable for implementation in all types of applications. In this paper, a very fast direct search algorithm is developed to solve the general dimension LAV problem using only elementary operations. The algorithm has been shown to be significantly faster than the LP approach through several experiments. © 2009 AACC.

AB - The Least Squares (LS) problem has been popular in industrial modeling applications due to its speed, efficiency and simplicity. However, the LS solution is known to be unreliable when the data distribution is not Gaussian and is flat-tailed and such data anomalies occur frequently in the industry. The Least Absolute Value (LAV) problem overcomes these difficulties but at the expense of greatly increasing the complexity of the solution. This was partly addressed when it was shown that the LAV problem could be formulated as a Linear Programme (LP). However, the LP formulation is not suitable for implementation in all types of applications. In this paper, a very fast direct search algorithm is developed to solve the general dimension LAV problem using only elementary operations. The algorithm has been shown to be significantly faster than the LP approach through several experiments. © 2009 AACC.

U2 - 10.1109/ACC.2009.5160229

DO - 10.1109/ACC.2009.5160229

M3 - Conference contribution

SN - 9781424445240

SP - 14

EP - 19

BT - Proceedings of the American Control Conference|Proc Am Control Conf

T2 - 2009 American Control Conference, ACC 2009

Y2 - 1 July 2009

ER -

Algorithm for very fast computation of least absolute value regression

Abstract

Conference

Access to Document

Fingerprint

Cite this