Forecasting by density shaping using neural networks

Author

Baram, Yoram ; Roth, Ze´ev

Author_Institution

Dept. of Comput. Sci., Technion-Israel Inst. of Technol., Haifa, Israel

fYear

1995

fDate

9-11 Apr 1995

Firstpage

57

Lastpage

71

Abstract

An estimate of the probability density function of a random vector is obtained by maximizing the output entropy of a feedforward network of sigmoidal units with respect to the input weights. A normalized version of the sigmoidal transfer function simplifies the algorithm considerably and leads to a maximum entropy estimate of the input density under a certain model. Newton´s method, applied to the estimated density, yields a recursive estimator for a random variable or a random sequence. A constrained connectivity structure yields a linear estimator, which is particularly suitable for “real time” prediction

Keywords

Newton method; feedforward neural nets; mathematics computing; maximum entropy methods; probability; recursive estimation; transfer functions; Newton method; constrained connectivity structure; density shaping; estimated density; feedforward network; forecasting; input weights; maximum entropy estimate; neural networks; output entropy; probability density function; random sequence; random variable; random vector; real time prediction; recursive estimator; sigmoidal transfer function; sigmoidal units; Entropy; Feedforward systems; Function approximation; Neural networks; Probability density function; Random variables; Recursive estimation; Signal processing algorithms; Transfer functions; Yield estimation;

fLanguage

English

Publisher

ieee

Conference_Titel

Computational Intelligence for Financial Engineering, 1995.,Proceedings of the IEEE/IAFE 1995

Conference_Location

New York, NY

Print_ISBN

0-7803-2145-6

Type

conf

DOI

10.1109/CIFER.1995.495253

Filename

495253