Additional References: Re: [Comp-neuro] The Neuronal Replicator Hypothesis

padams at notes.cc.sunysb.edu padams at notes.cc.sunysb.edu
Sat Oct 23 14:44:28 CEST 2010


Comp-neuro readers of the recently announced Neural Computation paper on 
the Neuronal Replicator Hypothesis by Fernando, Goldstein and Szathmary 
might be also  interested in our related, but quite distinct, work on 
Synaptic Darwinism. References to key papers, including those available 
freely online, are given below. The papers are also available at our 
website syndar.org

The main difference from the Neuronal Replicator idea, and similar work by 
Edelman, Changeux etc, is that we propose that synaptic weights 
"replicate". This is merely new terminology for standard Hebbian learning 
("replication" = "strengthening"), and brings nothing interesting. 
However, we add an important novel ingredient, based on the experimental 
findings that ltp/ltd is not completely synapse-specific. Such "crosstalk" 
between adjustments at different connections introduces a "mutational" 
element into standard neural network learning procedures. In particular, 
for nonlinear learning rules (such as those used in ICA), qualitative, and 
undesirable, changes in learning occur, sometimes even at extremely low, 
probably unavoidable, crosstalk levels. We conclude that nontrivial 
unsupervised learning might require extraordinarily accurate weight 
changes, and speculate that the key feature which allows neocortex to do 
such learning is special "Hebbian proofreading" circuitry, analogous to 
the machinery underlying accurate DNA replication. This unusual circuitry 
might explain many of the more puzzling features of the neocortex. 

Paul Adams and Kingsley Cox
Stony Brook University
syndar.org

        •       Cox K and Adams PR (2009) Hebbian crosstalk prevents 
nonlinear unsupervised learning. Front. Comput. Neurosci. 3:1-20 2009
 
        •       Radulescu, A., Cox, K., and Adams, P. (2009). Hebbian 
errors in learning: an analysis using the Oja model. J. Theor. Biol. 258, 
489-501.
 
        •       Adams, P and Cox, K. (2006). A neurobiological perspective 
on building intelligent devices. The Neuromorphic Engineer 3, 2-10
 
        •       Adams, P. and Cox, K. (2002). A new interpretation of 
thalamocortical circuitry. Phil. Trans. R. Soc. Lond. B, 357, 1767-1779.
 
        •       Adams,P.R. and Cox, K.J.A. (2002). Synaptic Darwinism and 
neocortical function. Neurocomputing 42, 197-214 

        •       Cox, K.J.A. and Adams, P.R. (2000). Implications of 
synaptic digitisation and error for neocortical function. Neurocomputing. 
32-33, 673-678
 
        •       Adams P. (2000) To Sleep, Perchance to Dream. Science 
Spectra 23, 46-55

        •       Adams, P.R (1998). Hebb and Darwin. J. Theoret. Biol. 195
:419-438

-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://www.neuroinf.org/pipermail/comp-neuro/attachments/20101023/cb24b689/attachment.html


More information about the Comp-neuro mailing list