Autonomous learning machine for online big data analytics

Abstract

Incremental learning is a method of machine learning in which input data is continuously used to extend the existing model's knowledge to further train the model. However, as real world data is not constant, incremental learning often suffers from concept drift issues. Another challenge faced by incremental learning is it’s fully supervised nature which requires all input data to have true class labels. However, this might not be possible in all cases as real world industrial data often comes with an uncertainty level and might contain unlabelled data. As a result, despite the rapid advancements in incremental learning, it’s applications in specific real world scenarios like factory and manufacturing surveillance are often limited. In order to solve these problems, a self evolving structure, Parsimonious network (ParsNet) is proposed. It is developed from a closed-loop configuration of the self-evolving generative and discriminative training processes exploiting shared parameters in which its structure flexibly grows and shrinks to overcome the issue of concept drift with/without labels. This paper explains the working of ParsNet and proposes its application in real world industrial and manufacturing scenarios by evaluating the performance of ParsNet in two distinct real world applications namely RFID localization and injection moulding in comparison to other similar algorithms like Devdan and Pensemble plus.