Multi-class multi-scale stacked sequential learning

Abstract

One assumption in supervised learning is that data is independent and identically distributed. However, this assumption does not hold true in many real cases. Sequential learning is that discipline of machine learning that deals with dependent data. In this paper, we revise the Multi-Scale Sequential Learning approach (MSSL) for applying it in the multi-class case (MMSSL). We have introduced the ECOC framework in the MSSL base classifiers and a formulation for calculating confidence maps from the margins of the base classifiers. Another important contribution of this papers is the MMSSL compression approach for reducing the number of features in the extended data set. The proposed methods are tested on 5-class and 9-class image databases. © 2011 Springer-Verlag.