Maximum-entropy parameter estimation for the k-NN modified value-difference kernel

I.H.E. Hendrickx; A. van den Bosch

Maximum-entropy parameter estimation for the k-NN modified value-difference kernel

I.H.E. Hendrickx, A. van den Bosch

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

Abstract

We introduce an extension of the modified value-difference kernel of $k$-nn by replacing the kernel's default class distribution matrix with the matrix produced by the maximum-entropy learning algorithm. This hybrid algorithm is tested on fifteen machine learning benchmark tasks, comparing the hybrid to standard $k$-nn classification and maximum-entropy-based classification. Results show that the hybrid typically outperforms the lower-scoring of the two other algorithms, often significantly; in a majority of cases the hybrid yields the highest accuracy of the three algorithms. Error analysis indicates that the hybrid's errors overlap more with $k$-nn than with maximum entropy modeling

Original language	English
Title of host publication	Proceedings of the 16th Belgian-Dutch Conference on Artificial Intelligence (BNAIC 2004), 21-22 october 2004, Groningen, The Netherlands
Editors	R. Verbruggen, N. Taatgen, L. Schomaker
Place of Publication	[s.l]
Publisher	[s.n.]
Pages	19-26
Number of pages	8
Volume	16
Publication status	Published - 2004

Publication series

Name
Volume	16

2 Finished

Algorithm development for memory models of language
Hendrickx, I. H. E., Daelemans, W. M. P. & van den Bosch, A.
1/09/01 → 1/09/05
Project: Research project
Memory models of language
van den Bosch, A.
1/07/01 → 1/07/06
Project: Research project

Cite this

@inproceedings{fed1eb75669f4bb88c8c4cd2d21a6d75,

title = "Maximum-entropy parameter estimation for the k-NN modified value-difference kernel",

abstract = "We introduce an extension of the modified value-difference kernel of $k$-nn by replacing the kernel's default class distribution matrix with the matrix produced by the maximum-entropy learning algorithm. This hybrid algorithm is tested on fifteen machine learning benchmark tasks, comparing the hybrid to standard $k$-nn classification and maximum-entropy-based classification. Results show that the hybrid typically outperforms the lower-scoring of the two other algorithms, often significantly; in a majority of cases the hybrid yields the highest accuracy of the three algorithms. Error analysis indicates that the hybrid's errors overlap more with $k$-nn than with maximum entropy modeling",

author = "I.H.E. Hendrickx and {van den Bosch}, A.",

note = "Pagination: 8",

year = "2004",

language = "English",

volume = "16",

publisher = "[s.n.]",

pages = "19--26",

editor = "R. Verbruggen and N. Taatgen and L. Schomaker",

booktitle = "Proceedings of the 16th Belgian-Dutch Conference on Artificial Intelligence (BNAIC 2004), 21-22 october 2004, Groningen, The Netherlands",

}

Maximum-entropy parameter estimation for the k-NN modified value-difference kernel. / Hendrickx, I.H.E.; van den Bosch, A.

Proceedings of the 16th Belgian-Dutch Conference on Artificial Intelligence (BNAIC 2004), 21-22 october 2004, Groningen, The Netherlands. ed. / R. Verbruggen; N. Taatgen; L. Schomaker. Vol. 16 [s.l] : [s.n.], 2004. p. 19-26.

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

TY - GEN

T1 - Maximum-entropy parameter estimation for the k-NN modified value-difference kernel

AU - Hendrickx, I.H.E.

AU - van den Bosch, A.

N1 - Pagination: 8

PY - 2004

Y1 - 2004

N2 - We introduce an extension of the modified value-difference kernel of $k$-nn by replacing the kernel's default class distribution matrix with the matrix produced by the maximum-entropy learning algorithm. This hybrid algorithm is tested on fifteen machine learning benchmark tasks, comparing the hybrid to standard $k$-nn classification and maximum-entropy-based classification. Results show that the hybrid typically outperforms the lower-scoring of the two other algorithms, often significantly; in a majority of cases the hybrid yields the highest accuracy of the three algorithms. Error analysis indicates that the hybrid's errors overlap more with $k$-nn than with maximum entropy modeling

AB - We introduce an extension of the modified value-difference kernel of $k$-nn by replacing the kernel's default class distribution matrix with the matrix produced by the maximum-entropy learning algorithm. This hybrid algorithm is tested on fifteen machine learning benchmark tasks, comparing the hybrid to standard $k$-nn classification and maximum-entropy-based classification. Results show that the hybrid typically outperforms the lower-scoring of the two other algorithms, often significantly; in a majority of cases the hybrid yields the highest accuracy of the three algorithms. Error analysis indicates that the hybrid's errors overlap more with $k$-nn than with maximum entropy modeling

M3 - Conference contribution

VL - 16

SP - 19

EP - 26

BT - Proceedings of the 16th Belgian-Dutch Conference on Artificial Intelligence (BNAIC 2004), 21-22 october 2004, Groningen, The Netherlands

A2 - Verbruggen, R.

A2 - Taatgen, N.

A2 - Schomaker, L.

PB - [s.n.]

CY - [s.l]

ER -

Maximum-entropy parameter estimation for the k-NN modified value-difference kernel

Abstract

Publication series

Fingerprint

Projects

Algorithm development for memory models of language

Memory models of language

Cite this