Alternating multiconlitron: A novel framework for piecewise linear classification

Abstract

Multiconlitron is a general framework for designing piecewise linear classifiers, but it may contain a relatively large number of conlitrons and linear functions. Based on the concept of maximal convexly separable subset (MCSS), we propose alternating multiconlitron as a novel framework for piecewise linear classification. Using the support alternating multiconlitron algorithm, an alternating multiconlitron can be constructed as a series of conlitrons alternately from a subset of one class to the MCSS of the other class. Experimental results show that in practice an alternating multiconlitron generally has a much simpler structure than a corresponding multiconlitron, performing very fast in testing phase with similar or better accuracies.