Convergent algorithm for sensory receptive field development

Authors:
Joseph J. Atick;A. Norman Redlich
Affiliations:
-;-
Venue:
Neural Computation
Year:
1993

Citing 0
Cited 12

Original Contribution: Efficient information transfer and anti-Hebbian neural networks

Neural Networks
Activation functions, computational goals, and learning rules for local processors with contextual guidance

Neural Computation
Spatial decorrelation in orientation-selective cortical cells

Neural Computation
Neural networks with local receptive fields and superlinear VC Dimension

Neural Computation
Spatiochromatic receptive field properties derived from information-theoretic analyses of cone mosaic responses to natural scenes

Neural Computation
Nonlinear Complex-Valued Extensions of Hebbian Learning: An Essay

Neural Computation
Using perceptual relation of regularity and anisotropy in the texture with independent component model for defect detection

Pattern Recognition
Supervised factorial learning

Neural Computation
Palmprint Recognition Using a Novel Sparse Coding Technique

ISNN '07 Proceedings of the 4th international symposium on Neural Networks: Part II--Advances in Neural Networks
A theoretical basis for emergent pattern discrimination in neural systems through slow feature extraction

Neural Computation
Learning image transformations without training examples

ISVC'11 Proceedings of the 7th international conference on Advances in visual computing - Volume Part II
Task-Oriented sparse coding model for pattern classification

ICNC'05 Proceedings of the First international conference on Advances in Natural Computation - Volume Part I

Quantified Score

Hi-index	0.00

Visualization

Abstract

An unsupervised developmental algorithm for linear maps is derivedwhich reduces the pixel-entropy (using the measure introduced inprevious work) at every update and thus removes pairwisecorrelations between pixels. Since the measure of pixel-entropy hasonly a global minimum the algorithm is guaranteed to converge tothe minimum entropy map. Such optimal maps have recently been shownto possess cognitively desirable properties and are likely to beused by the nervous system to organize sensory information. Thealgorithm derived here turns out to be one proposed by Goodall forpairwise decorrelation. It is biologically plausible since in aneural network implementation it requires only data availablelocally to a neuron. In training over ensembles of two-dimensionalinput signals with the same spatial power spectrum as naturalscenes, networks develop output neurons with center-surroundreceptive fields similar to those of ganglion cells in the retina.Some technical issues pertinent to developmental algorithms of thissort, such as symmetry fixing, are also discussed.