What is it about?
The WiSARD n-tuple classifier is a simple and yet powerful learning model. In this work we determine an exact value for a learning capacity metric, as a means to better understand it. We also offer an exact value of this metric for an extension of WiSARD, and show how it can be related to other well known learning machines.
Featured Image
Why is it important?
The VC dimension is a learning capacity metric. It determines how well a given learning machine will learn given a set of examples. However, it is quite hard to achieve an exact value for this metric. Most learning machines have its VC dimension defined between bounds. Given WiSARD's nature, we could achieve an exact value for it, which may help in the better understanding of this learning model. Also, this publication is the first theoretical contribution made on the bleaching technique, which granted a big improvement in WiSARD's performance, but whose capability had only been given through empirical analyses.
Perspectives
I hope this article bring some light on weightless techniques, especially WiSARD n-tuple classifier, by showing that not only it has good empirical results, but also that there are theoretical studies that advocate its capabilities as a learning machine.
Hugo Cesar de Castro Carneiro
Universidade Federal do Rio de Janeiro
Read the Original
This page is a summary of: The Exact VC Dimension of the WiSARD n-Tuple Classifier, Neural Computation, January 2019, The MIT Press,
DOI: 10.1162/neco_a_01149.
You can read the full text:
Contributors
The following have contributed to this page
