Integrating the Kernel Method to Autonomous Learning Multi-Model Systems for Online Data

摘要

We present a novel and simple approach to incorporate the kernel function method to a recently proposed autonomous learning system (ALS). An ALS can learn online from data streams without any need to offline batch training and it is both memory and computational power efficient. We have codenamed our approach: KALMMo for Kernelized Autonomous Learning Multi Model system. Using the Radial Basis Function (RBF) kernel, we have tested the performance of KALMMo using four well-known and challenging datasets and compared the results to other well established algorithms. Our results have shown that KALMMo performed at least as good as other competitive approaches or even better. Its training time is linearly proportional to the number of instances of a dataset and that the training time is better by an order of magnitude to the nearest competitor. KALMMo shows interesting feature to several applications including big data and machine learning classification. The performance of the proposed systems should be tested with other types of kernel functions.