data aggregation ayers of formal neurons, separability principles
Issue Date:
2008
Publisher:
Institute of Information Theories and Applications FOI ITHEA
Abstract:
Information extraction or knowledge discovery from large data sets should be linked to data
aggregation process. Data aggregation process can result in a new data representation with decreased number
of objects of a given set. A deterministic approach to separable data aggregation means a lesser number of
objects without mixing of objects from different categories. A statistical approach is less restrictive and allows for
almost separable data aggregation with a low level of mixing of objects from different categories. Layers of formal
neurons can be designed for the purpose of data aggregation both in the case of deterministic and statistical
approach. The proposed designing method is based on minimization of the of the convex and piecewise linear
(CPL) criterion functions.