Adaptive Least-Squares Support Vector Machine and its Combined Learning-Selflearning in Image Recognition Task

Abstract

In the paper, we introduce an image recognition system that is based on least squares support vector machines and has matrix inputs. It is designed to solve a vast class of tasks within general problems of Data Stream Mining and Big Data Mining, in particular, an image recognition task when observations are fed sequentially in online mode. Its distinctive features include not just the ability to process images in their initial matrix form without vectorization, but also that centers of activation functions are formed with the observations from the training set. The tuning procedure of the system is characterized by the combination of the supervised learning paradigm, “lazy” learning using the “neurons at data points” concept, T. Kohonen’s self-learning, and learning vector quantization.