Semi-supervised topo-Bayesian ARTMAP for noisy data
Document Type
Article
Publication Date
1-1-2018
Abstract
This paper presents a novel semi-supervised ART network that inherits the ability of noise insensitivity, topology learning, and incremental learning from the Bayesian ARTMAP. It is combined with a label prediction strategy based on a clustering technique to determine the neighboring neurons. The procedure of updating Bayesian ARTMAP is modified to allow the network in altering the learning rate. This results in a classifier that works online and lifts several limitations of the original Bayesian ARTMAP. It processes arbitrarily scaled values even when their range is not entirely known in advance. The classifier has the capability to be employed in online learning applications, in which no prior-knowledge about the structure and distribution of data is available. Experimental results indicate good results, even with noisy data.
Keywords
Bayes decision theory, Category proliferation, Incremental learning, Neural network
Divisions
fsktm
Funders
University of Malaya Grand Challenge Grant GC003A-14HTM
Publication Title
Applied Soft Computing
Volume
62
Publisher
Elsevier