ADAPTIVE MODELLING OF LIKELIHOOD CLASSIFICATION-1.
Final rept., 13 Jan 65-13 Jan 66.
PHILCO CORP BLUE BELL PA COMMUNICATIONS AND ELECTRONICS DIV
Pagination or Media Count:
The purpose of this study is the development and evaluation of adaptive networks to represent broad classes of likelihood functions. Orthogonal expansions for multivariate distributions of discrete and continuous random variables are investigated for this application. To overcome the problem of high dimensionality, Markovian processing of discrete spatial patterns is introduced. In addition, adaptive threshold adjustment procedures and an optimal method for taking context into account are derived from Compound Decision Theory. Experimental results, on the classification of visual patterns by a two-layer, two threshold network, are also presented. Author
- Statistics and Probability