Title :
Modified Bernstein polynomials and their connectionist interpretation
Author_Institution :
MCC, Austin, TX, USA
Abstract :
The Weierstrass polynomial approximation theorem plays a central role in showing approximation capability of polynomial-based higher-order feedforward networks. The Bernstein polynomials are a family of polynomials which satisfy the Weierstrass polynomial theorem. Baldi (1991) interpreted the Bernstein polynomials in connectionist framework and argued that they can be viewed as a model for biological bell-shaped receptive fields. However, due to their strict constraint of equally-spaced input points, their connectionist interpretation has some problems in the real-world setting. In this paper, we present the modified Bernstein polynomials which have lesser constraints on the position of input points. The modified Bernstein polynomials can directly utilize given input/output data with a minor constraint on the position of the input points. We present theorems that show the approximation capability of the modified Bernstein polynomials. The relationship between the modified Bernstein polynomials and other higher-order feedforward network approaches is also discussed
Keywords :
approximation theory; feedforward neural nets; polynomials; I/O data; Weierstrass polynomial approximation theorem; Weierstrass polynomial theorem; approximation capability; biological bell-shaped receptive fields; connectionist interpretation; equally-spaced input points; input/output data; modified Bernstein polynomials; polynomial-based high-order feedforward networks; Biological system modeling; Drives; Microelectronics; Polynomials;
Conference_Titel :
Neural Networks, 1994. IEEE World Congress on Computational Intelligence., 1994 IEEE International Conference on
Print_ISBN :
0-7803-1901-X
DOI :
10.1109/ICNN.1994.374496