The role of weight normalization in competitive learning

Authors:
Geoffrey J. Goodhill;Harry G. Barrow
Affiliations:
-;-
Venue:
Neural Computation
Year:
1994

Citing 0
Cited 11

Marr's theory of the neocortex as a self-organizing neural network

Neural Computation
Fixed-point attractor analysis for a class of neurodynamics

Neural Computation
Multiplicative synaptic normalization and a nonlinear Hebb rule underlie a neurotrophic model of competitive synaptic plasticity

Neural Computation
An analysis of synaptic normalization in a general class of Hebbian models

Neural Computation
Synaptic weight normalization effects for topographic mapping formation

Neural Networks - 2004 Special issue: New developments in self-organizing systems
Generalization in Interactive Networks: The Benefits of Inhibitory Competition and Hebbian Learning

Neural Computation
Effective Neuronal Learning with Ineffective Hebbian Learning Rules

Neural Computation
A nonlinear hebbian network that learns to detect disparity in random-dot stereograms

Neural Computation
Cross-talk induces bifurcations in nonlinear models of synaptic plasticity

Neural Computation
Spike-timing-dependent plasticity and short-term plasticity jointly control the excitation of hebbian plasticity without weight constraints in neural networks

Computational Intelligence and Neuroscience
Modeling binocular competition through Hebbian plasticity and constrained connectivity

Neurocomputing

Quantified Score

Hi-index	0.00

Visualization

Abstract

The effect of different kinds of weight normalization on theoutcome of a simple competitive learning rule is analyzed. It isshown that there are important differences in the representationformed depending on whether the constraint is enforced by dividingeach weight by the same amount ("divisive enforcement") orsubtracting a fixed amount from each weight ("subtractiveenforcement"). For the divisive cases weight vectors spread outover the space so as to evenly represent "typical" inputs, whereasfor the subtractive cases the weight vectors tend to the axes ofthe space, so as to represent "extreme" inputs. The consequences ofthese differences are examined.