Improvements to Nearest Neighbor Classifier: Pattern Synthesis, Compact Data Representation and other Schemes - Couverture souple

Pulabaigari, Viswanath; Hitendra Sarma, T

 
9783845416496: Improvements to Nearest Neighbor Classifier: Pattern Synthesis, Compact Data Representation and other Schemes
Couverture souple
ISBN 10 : 3845416491 ISBN 13 : 9783845416496
Editeur : LAP LAMBERT Academic Publishing, 2011

Synopsis

Nearest neighbor classifier (NNC), a non-parametric pattern classification technique is not only simple to use, but often shows good performance. It is used in several domains, like data mining, machine learning, image/video/audio data analysis and retrieval, etc. It has some shortcomings or limitations, like it can be biased due to curse of dimensionality effect, huge computational requirements, etc. With large number of training instances it can achieve a better classification accuracy. But, this can worse the computational burden. The monograph presents a series of techniques whereby number of training instances can be artificially increased and can be stored in compact data representation schemes. This means, training set size can be virtually increased, but computational burden is not. The monograph presents, mainly, pattern synthesis techniques called partition based pattern synthesis and overlap based pattern synthesis, and their respective compact data structures. It is shown, both theoretically and experimentally that the proposed methods are effective. The monograph also presents some other improvements to NNC and presents a constant time NNC also.

Les informations fournies dans la section « Synopsis » peuvent faire référence à une autre édition de ce titre.

Présentation de l'éditeur

Nearest neighbor classifier (NNC), a non-parametric pattern classification technique is not only simple to use, but often shows good performance. It is used in several domains, like data mining, machine learning, image/video/audio data analysis and retrieval, etc. It has some shortcomings or limitations, like it can be biased due to curse of dimensionality effect, huge computational requirements, etc. With large number of training instances it can achieve a better classification accuracy. But, this can worse the computational burden. The monograph presents a series of techniques whereby number of training instances can be artificially increased and can be stored in compact data representation schemes. This means, training set size can be virtually increased, but computational burden is not. The monograph presents, mainly, pattern synthesis techniques called partition based pattern synthesis and overlap based pattern synthesis, and their respective compact data structures. It is shown, both theoretically and experimentally that the proposed methods are effective. The monograph also presents some other improvements to NNC and presents a constant time NNC also.

Biographie de l'auteur

P. Viswanth received his M.Tech from IIT-Madras, and Ph.D.(best thesis award given) from IISc-Bangalore. He worked at IIT-Guwahati from 2005 to 2008. Since then he is working as a Professor and Dean, R&D (Electrical Sciences)at Rajeev Gandhi Memorial College of Eng. & Tech., Nandyal,A.P, India. URL:https://sites.google.com/site/viswanathpulabaigari

Les informations fournies dans la section « A propos du livre » peuvent faire référence à une autre édition de ce titre.

Éditeur
LAP LAMBERT Academic Publishing
Date d'édition
2011
Langue
anglais
ISBN 10 
3845416491
ISBN 13 
9783845416496
Reliure
Broché
Nombre de pages
184
Coordonnées du fabricant
non disponible
Personne responsable
non disponible