Pattern Analysis and Applications

, Volume 18, Issue 2, pp 247–261

Generalized multi-scale stacked sequential learning for multi-class classification

Theoretical Advances

DOI: 10.1007/s10044-013-0333-y

Cite this article as:
Puertas, E., Escalera, S. & Pujol, O. Pattern Anal Applic (2015) 18: 247. doi:10.1007/s10044-013-0333-y

Abstract

In many classification problems, neighbor data labels have inherent sequential relationships. Sequential learning algorithms take benefit of these relationships in order to improve generalization. In this paper, we revise the multi-scale sequential learning approach (MSSL) for applying it in the multi-class case (MMSSL). We introduce the error-correcting output codesframework in the MSSL classifiers and propose a formulation for calculating confidence maps from the margins of the base classifiers. In addition, we propose a MMSSL compression approach which reduces the number of features in the extended data set without a loss in performance. The proposed methods are tested on several databases, showing significant performance improvement compared to classical approaches.

Keywords

Stacked sequential learning Multi-scale Error-correct output codes (ECOC) Contextual classification 

Abbreviation List

X

Set of samples

Y

Set of labels

x

A sample

y

A label

h(x)

A classifier

\(y^{\prime}\)

A prediction from a classifier

\(y^{\prime\prime}\)

A final prediction from a chain of classifiers

xext

Extended set

J

Neighborhood relationship function

z

Neighborhood model features

ρ

Neighborhood

θ

Neighborhood parameterization

w

Number of elements in the neighborhood window

s

Number of scales

c

Set of different classes in a multi-class problem

\(\hat{F}(\mathbf{x}, c)\)

A prediction confidence map

N

Number of classes in a multi-class problem

n

Number of dichotomizers

σ

Parameter of a Gaussian filter

Set of scales defined by σ parameters

b

A dichotomizer

M

ECOC coding matrix

\({\mathcal{Y}}\)

A class codeword in ECOC framework

\({\mathcal{X}}\)

A sample prediction codeword in ECOC framework

mx

Margin for a prediction of sample x

β

Constant which governs transition in a sigmoidean function

t

Number of iterations in an ADABoost classifier

δ

A soft distance

α

Normalization parameter for soft distance δ

gσ

A multidimensional isotropic gaussian filter with zero mean and σ standard deviation

\({\mathcal{P}}\)

A set of partitions of classes

P

A partition of groups of classes

γ

A symbol in a partition codeword

\(\Upgamma\)

A partition codeword

R

The mean ranking for each system configurations

E

The total number of experiments

k

The total number of system configuration

\(\chi_{2}^{F}\)

Friedman statistic value

Copyright information

© Springer-Verlag London 2013

Authors and Affiliations

  1. 1.Dept. Matemàtica Aplicada i AnàlisiUniversitat de BarcelonaBarcelonaSpain
  2. 2.Computer Vision CenterBellaterraSpain