A polychotomizer which assigns the input to one of K3 classes is constructed using a set of dichotomizers which assign the input to one of two classes. Defining classes in terms of the dichotomizers is the binary decomposition matrix of size K×L where each of the K3 classes is written as error-correcting output codes (ECOC), i.e., an array of the responses of binary decisions made by L dichotomizers. We use linear dichotomizers and by combining them suitably, we build nonlinear polychotomizers, thereby reducing complex decisions into a group of simpler decisions. We propose a method to learn the error-correcting codes from data based on soft weight sharing which forces parameters to take one of a set (here two: -1/+1) values. Simulation results on eight datasets indicate that compared with a linear one-per-class polychotomizer and ECOC proper, these methods generate more accurate classifiers, using less dichotomizers than pairwise classifiers.
[1]
Thomas G. Dietterich.
Approximate Statistical Tests for Comparing Supervised Classification Learning Algorithms
,
1998,
Neural Computation.
[2]
Heekuck Oh,et al.
Neural Networks for Pattern Recognition
,
1993,
Adv. Comput..
[3]
Thomas G. Dietterich,et al.
Solving Multiclass Learning Problems via Error-Correcting Output Codes
,
1994,
J. Artif. Intell. Res..
[4]
Geoffrey E. Hinton,et al.
Simplifying Neural Networks by Soft Weight-Sharing
,
1992,
Neural Computation.
[5]
Eddy Mayoraz,et al.
Improved Pairwise Coupling Classification with Correcting Classifiers
,
1998,
ECML.
[6]
Y. Chien,et al.
Pattern classification and scene analysis
,
1974
.
[7]
Geoffrey E. Hinton,et al.
Adaptive Mixtures of Local Experts
,
1991,
Neural Computation.
[8]
Eddy Mayoraz,et al.
On the Decomposition of Polychotomies into Dichotomies
,
1997,
ICML.