文档介绍：ms for a Neural
ally I~~ib~~e~
Qiang Gan*#,Jun Yao' and . Subramanian*
* School of Electrical and Elecfronic Engineering # Department of Biomedical Engineering
Nanyang Technological University Southeast University
Nanyang Avenue, Singapore 639 798 Nanjing 2 I0096
E-mail: ******@tu. edusg
Abstract function shaped receptive fields bining local
smoothing and sharpening which provides the function of
This paper presents a work with its output lateral inhibition. We find that this kind of receptive fields
layer as a classijier and its hidden layer constrained by can not be formed automatically by learning without any
laterally inhibited receptive Pel& asfeature extractor, in constraints on the weights in a multilayer perceptron.
which the idea that wavelet transforms are very suitable Wavelet transform is a good model for the receptive
for modeling the primary visual information processing is fields in the human visual system [2]. Because a wavelet
reflected. Two learning algorithms for designing the function satisfies the admissibility condition [3], it must be
receptive fields are proposed. The problem associated oscillatory across its zero points. Hence, wavelet hctions
with local minima caused by the inherent oscillatory provide natural models for laterally inhibited receptive
property in laterally inhibited receptive fields