Differentiating noise and modulators in artificial neural networks

Docking, Philip John (1993). Differentiating noise and modulators in artificial neural networks. PhD thesis The Open University.

DOI: https://doi.org/10.21954/ou.ro.0000e004


Research in Computational Neural Networks is currently taking place at many different levels; from coarse-grain symbolic models to fine-grain representations of neurons and cell processes. One feature that the different approaches share, is that they are all in relative infancy. Thus, most research concentrates on gross aspects of neural communication and methods of computational simulation.

Recently, some clues have been found which point to more subtle mechanisms underlying the information processing capability of neural 'nodes'. These clues are the improvement in network operation by the injection of random noise; and the neurobiological finding that neuropeptides may exist as slower Signal transmission channels between neurons.

This study concerns the difference between random noise injection, and directed, low-level, activity injections which are postulated to be produced by neuromodulators such as neuropeptides. The findings of this study are that random noise does, indeed, enhance the operation of coarse-grain neural models; and that a 'neuropeptidergic' analogue also enhances operation; but to a different extent, and probably through a different mechanism. Further testing of a medium-grain computer model gives some indication of how a neuropeptidergic modulation might affect real neurons, by extending the time-course of the activation of the neuron. This appears to be a similar mechanism to that postulated for the coarse-grain 'neuropeptidergic' simulation model.

Given these findings, is it possible that signal transmission in real nervous systems assume these mechanisms? If so, it may be possible that a process of concurrent propagation, through different signal channels, also occurs in real nervous systems, making the nervous system much more complex than current models allow.

Viewing alternatives

Download history


Public Attention

Altmetrics from Altmetric

Number of Citations

Citations from Dimensions

Item Actions