Majority and Location_Based Fusers for Systems of PAC Concept Learners

Nageswara S.V. Rao; E. M. Oblow

doi:10.1109/21.293485

Majority and Location_Based Fusers for Systems of PAC Concept Learners

Nageswara S.V. Rao, E. M. Oblow

Autonomous and Complex Systems

Research output: Contribution to journal › Article › peer-review

11 Scopus citations

Abstract

A system of probably and approximately correct learners of Valiant type that infer concepts from a sample is considered. Each learner had been trained by a sample using the methods of minimizing the empirical error, and no examples are available to the fuser. A majority fuser is known to make the composite system better than the best of the learners in terms of normalized confidence (that corresponds to the same precision value). An analysis of general majority fusers is carried out to obtain bounds on actual and expected errors. Conditions under which the r_of_N fuser performs better, in terms of normalized confidence or precision, than best of the individual learners are obtained. For a special class of statistically independent learners, slightly weaker conditions are obtained. Two fusers that use the location information of a test point are proposed, and are shown to be better than a learner with least empirical error.

Original language	English
Pages (from-to)	713-727
Number of pages	15
Journal	IEEE Transactions on Systems, Man, and Cybernetics
Volume	24
Issue number	5
DOIs	https://doi.org/10.1109/21.293485
State	Published - May 1994

Funding

Manuscript received September 22, 1992; revised June 23, 1993. This paper was funded in part by the Basic Energy Sciences Program, Department of Energy and the Intelligent Systems Program, Office of Naval Research. The first author is partially funded by National Science Foundation under grant #IRI-9 1086 10. The authors gratefully acknowledge the continuing financial support of the learning research by Oscar Manley of the Basic Energy Sciences Program in the Department of Energy and Teresa McMullen in the Intelligent Systems Program of the Office of Naval Research in the Department of Defense. We thank the anonymous reviewer who pointed out errors in the original conditions of Theorem 3.3 and Theorem 3.7. We thank all the anonymous reviewers who contributed to the readability of this paper.

Funders	Funder number
Basic Energy Sciences Program
Department of Energy and Teresa McMullen
Office of Naval Research in the Department of Defense
National Science Foundation	#IRI-9 1086 10
Office of Naval Research
U.S. Department of Energy

Keywords

Computational learnability
General majority rule location_based fusers
N-learners problem

Access to Document

10.1109/21.293485

Cite this

@article{fff4ee9abe5c43dba1904d6067db3c2e,

title = "Majority and Location_Based Fusers for Systems of PAC Concept Learners",

abstract = "A system of probably and approximately correct learners of Valiant type that infer concepts from a sample is considered. Each learner had been trained by a sample using the methods of minimizing the empirical error, and no examples are available to the fuser. A majority fuser is known to make the composite system better than the best of the learners in terms of normalized confidence (that corresponds to the same precision value). An analysis of general majority fusers is carried out to obtain bounds on actual and expected errors. Conditions under which the r_of_N fuser performs better, in terms of normalized confidence or precision, than best of the individual learners are obtained. For a special class of statistically independent learners, slightly weaker conditions are obtained. Two fusers that use the location information of a test point are proposed, and are shown to be better than a learner with least empirical error.",

keywords = "Computational learnability, General majority rule location_based fusers, N-learners problem",

author = "Rao, {Nageswara S.V.} and Oblow, {E. M.}",

year = "1994",

month = may,

doi = "10.1109/21.293485",

language = "English",

volume = "24",

pages = "713--727",

journal = "IEEE Transactions on Systems, Man, and Cybernetics",

issn = "0018-9472",

number = "5",

}

TY - JOUR

T1 - Majority and Location_Based Fusers for Systems of PAC Concept Learners

AU - Rao, Nageswara S.V.

AU - Oblow, E. M.

PY - 1994/5

Y1 - 1994/5

N2 - A system of probably and approximately correct learners of Valiant type that infer concepts from a sample is considered. Each learner had been trained by a sample using the methods of minimizing the empirical error, and no examples are available to the fuser. A majority fuser is known to make the composite system better than the best of the learners in terms of normalized confidence (that corresponds to the same precision value). An analysis of general majority fusers is carried out to obtain bounds on actual and expected errors. Conditions under which the r_of_N fuser performs better, in terms of normalized confidence or precision, than best of the individual learners are obtained. For a special class of statistically independent learners, slightly weaker conditions are obtained. Two fusers that use the location information of a test point are proposed, and are shown to be better than a learner with least empirical error.

AB - A system of probably and approximately correct learners of Valiant type that infer concepts from a sample is considered. Each learner had been trained by a sample using the methods of minimizing the empirical error, and no examples are available to the fuser. A majority fuser is known to make the composite system better than the best of the learners in terms of normalized confidence (that corresponds to the same precision value). An analysis of general majority fusers is carried out to obtain bounds on actual and expected errors. Conditions under which the r_of_N fuser performs better, in terms of normalized confidence or precision, than best of the individual learners are obtained. For a special class of statistically independent learners, slightly weaker conditions are obtained. Two fusers that use the location information of a test point are proposed, and are shown to be better than a learner with least empirical error.

KW - Computational learnability

KW - General majority rule location_based fusers

KW - N-learners problem

UR - http://www.scopus.com/inward/record.url?scp=0028424223&partnerID=8YFLogxK

U2 - 10.1109/21.293485

DO - 10.1109/21.293485

M3 - Article

AN - SCOPUS:0028424223

SN - 0018-9472

VL - 24

SP - 713

EP - 727

JO - IEEE Transactions on Systems, Man, and Cybernetics

JF - IEEE Transactions on Systems, Man, and Cybernetics

IS - 5

ER -

Majority and Location_Based Fusers for Systems of PAC Concept Learners

Abstract

Funding

Keywords

Access to Document

Other files and links

Fingerprint

Cite this