We study the problem of multiclass classification within the framework of error correcting output codes (ECOC) using margin-based binary classifiers. Specifically, we address two important open problems in this context: decoding and model selection. The decoding problem concerns how to map the outputs of the classifiers into class codewords. In this paper we introduce a new decoding function that combines the margins through an estimate of their class conditional probabilities. Concerning model selection, we present new theoretical results bounding the leave-one-out (1,00) error of ECOC of kernel machines, which can be used to tune kernel hyperparameters. We report experiments using support vector machines as the base binary classifiers, showing the advantage of the proposed decoding function over other functions of the margin commonly used in practice. Moreover, our empirical evaluations on model selection indicate that the bound leads to good estimates of kernel parameters.

New results on error correcting output codes of kernel machines / A. Passerini;M. Pontil;P. Frasconi. - In: IEEE TRANSACTIONS ON NEURAL NETWORKS. - ISSN 1045-9227. - STAMPA. - 15:(2004), pp. 45-54. [10.1109/TNN.2003.820841]

New results on error correcting output codes of kernel machines

FRASCONI, PAOLO
2004

Abstract

We study the problem of multiclass classification within the framework of error correcting output codes (ECOC) using margin-based binary classifiers. Specifically, we address two important open problems in this context: decoding and model selection. The decoding problem concerns how to map the outputs of the classifiers into class codewords. In this paper we introduce a new decoding function that combines the margins through an estimate of their class conditional probabilities. Concerning model selection, we present new theoretical results bounding the leave-one-out (1,00) error of ECOC of kernel machines, which can be used to tune kernel hyperparameters. We report experiments using support vector machines as the base binary classifiers, showing the advantage of the proposed decoding function over other functions of the margin commonly used in practice. Moreover, our empirical evaluations on model selection indicate that the bound leads to good estimates of kernel parameters.
2004
15
45
54
A. Passerini;M. Pontil;P. Frasconi
File in questo prodotto:
File Dimensione Formato  
01263577.pdf

Accesso chiuso

Tipologia: Versione finale referata (Postprint, Accepted manuscript)
Licenza: Tutti i diritti riservati
Dimensione 398.2 kB
Formato Adobe PDF
398.2 kB Adobe PDF   Richiedi una copia

I documenti in FLORE sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.

Utilizza questo identificatore per citare o creare un link a questa risorsa: https://hdl.handle.net/2158/386595
Citazioni
  • ???jsp.display-item.citation.pmc??? ND
  • Scopus 121
  • ???jsp.display-item.citation.isi??? 105
social impact