Conversion of temporal correlations between stimuli to spatial correlations between attractors

Authors:
M. Griniasty;M. V. Tsodyks;Daniel J. Amit
Affiliations:
-;-;-
Venue:
Neural Computation
Year:
1993

Citing 0
Cited 12

Paradigmatic working memory (attractor) cell in IT cortex

Neural Computation
Predicting the distribution of synaptic strengths and cell firing correlations in a self-organizing, sequence prediction model

Neural Computation
Transform-invariant recognition by association in a recurrent network

Neural Computation
Encoding the Temporal Statistics of Markovian Sequences of Stimuli in Recurrent Neuronal Networks

ICANN '02 Proceedings of the International Conference on Artificial Neural Networks
Stable Neural Attractors Formation: Learning Rules and Network Dynamics

Neural Processing Letters
Two-level hierarchy with sparsely and temporally coded patterns and its possible functional role in information processing

Neural Networks
Mixed states on neural network with structural learning

Neural Networks
Part 3: brain science, information science and associative memory model

New Generation Computing
Hebbian learning of context in recurrent neural networks

Neural Computation
Symbolic Memory of Motion Patterns by an Associative Memory Dynamics with Self-organizing Nonmonotonicity

Neural Information Processing
Semantic priming in a cortical network model

Journal of Cognitive Neuroscience
Parallel retrieval of correlated patterns: From Hopfield networks to Boltzmann machines

Neural Networks

Quantified Score

Hi-index	0.01

Visualization

Abstract

It is shown that a simple modification of synaptic structures (ofthe Hopfield type) constructed to produce autoassociativeattractors, produces neural networks whose attractors arecorrelated with several (learned) patterns used in the constructionof the matrix. The modification stores in the matrix a fixedsequence of uncorrelated patterns. The network then has correlatedattractors, provoked by the uncorrelated stimuli. Thus, the networkconverts the temporal order (or temporal correlation) expressed bythe sequence of patterns, into spatial correlations expressed inthe distributions of neural activities in attractors. The modelcaptures phenomena observed in single electrode recordings inperforming monkeys by Miyashita et al. The correspondence is asclose as to reproduce the fact that given uncorrelated patterns assequentially learned stimuli, the attractors produced aresignificantly correlated up to a separation of 5 (five) in thesequence. This number 5 is universal in a range of parameters, andrequires essentially no tuning. We then discuss learning scenariosthat could lead to this synaptic structure as well as experimentalpredictions following from it. Finally, we speculate on thecognitive utility of such an arrangement.