Paper Info
Title
Automatic discriminative lossy binary conversion of redundant real training data inputs for simplifying an input data space and data representation
Abstract
Many times we come across the need to simplify or reduce an input data space in order to achieve a better model or better performance of an artificial intelligence solution. The well known PCA, ICA and rough sets can simplify and reduce input data space but they cannot transform real input data vectors into binary ones. Binary training vectors can simplify a training process of neural networks and let them to construct more compact topologies. This paper introduces a new algorithm that reduces input data space and simultaneously automatically lossy transforms real input training data vectors into binary vectors so that they do not lose their discrimination properties. The problem is how to effectively transform real input training data vectors into binary vectors so that an input data space could be simplified and the transformed binary vectors would be enough representative to be able to discriminate all training samples of all classes correctly? The described lossy conversion makes possible to achieve better generalization results for various soft-computing algorithms, can be widely used and avoids the curse of dimensionality problem. This paper introduces a new Automatic Discriminative Lossy Binary Conversion Algorithm (ADLBCA) that is able to solve all these tasks. Generally, no other method can simultaneously and so fast do all these tasks.
Year
DOI
Venue
2009
10.1007/978-3-642-04921-7_1
ICANNGA
Keywords
Field
DocType
binary vector,automatic discriminative lossy binary,input data space,real input training data,dimensionality problem,redundant real training data,training process,training sample,real input data vector,binary training vector,better performance,better model,data representation,curse of dimensionality,artificial intelligent,soft computing,rough set,neural network
External Data Representation,Lossy compression,Computer science,Curse of dimensionality,Rough set,Artificial intelligence,Iris flower data set,Artificial neural network,Discriminative model,Machine learning,Binary number
Conference
Volume
ISSN
ISBN
5495.0
0302-9743
3-642-04920-6
Citations 
PageRank 
References 
0
0.34
3
Authors
1
Name
Order
Citations
PageRank
Adrian Horzyk15312.76