Thalamocortical Circuitry Regulating Synaptic Learning

Ph.D., London University
Postdoctoral, Max Planck Institute, Goettingen
Lab Page

I am interested in the principles underlying the rather stereotyped circuitry of the neocortex and its partner, the thalamus. Presumably this circuitry exists to allow the neocortex to do something that other brain regions cannot do, or do in other ways. One possibility is that the neocortex is specialised to learn relatively subtle relationships within ensembles of input patterns (provided to it by the thalamus), possibly combined with evaluations of the outcome of output patterns. It is likely that neocortical learning is achieved by local, activity-dependent, modifications of synaptic strengths. Two obvious problems that the neocortex must deal with in this "continuous learning" scenario are (1) the number of input, processing and output neurons is so large that connectivity is extremely sparse, greatly restricting the immediate scope of local synaptic learning (2) the heart of neocortical function, local synaptic learning, is not anatomically precise, and such imprecision is likely to be an important factor limiting the ability of neocortex to learn. These 2 problems, sparse connectivity and synaptic error, are intertwined, and the latter may mitigate the former.

My basic hypothesis is that the lower the effective error rate, the more successful learning is likely to be in the long run. I therefore interpret the neocortex as a machine that avoids the adverse consequences of synaptic error.

If activity-dependent learning culminates in formation of new synapses these synapses must be correctly placed (so as to connect co-active neurons), and rare errors would involve placing synapses at the neighbors of coactive neurons. The survival of these erroneous synapses (or "mutations") will depend on the activity across these newly-formed trial connections. Thus the propagation of error depends on the relative degree of co-activity across current connections and across incipient connections. If a special type of neocortical neuron (for example, layer 6 neurons) measures this coactivity ratio, and appropriately controls the plasticity of current connections on a cell by cell basis, the adverse consequences of locally imprecise learning could be avoided.

My work tests this idea using 3 approaches: computer simulations, mathematical calculation, and analysis of neocortical circuitry and physiology.

Selected Publications

• Zhou,Q., Godwin, D.W., O'Malley. D.M. & Adams, P.R. (1997). Visualisation of calcium influx through channels that shape the burst and tonic firing modes of thalamic relay cells. J. Neurophysiol. 77: 2816 - 2825.
  
  
• Adams,P.R. (1998) Hebb and Darwin. J. Theoretic. Biol. 195:419-438 O'Malley, D.M., Burbach, B.J. and Adams, P.R. 1999. Fluorescent calcium indicators: subcellular behavior and use in confocal imaging. In: "Methods in Molecular Biology, vol 122 : Confocal Microscopy Methods and Protocols. Ed. S. Paddock. Humana Press Totowa N.J.
  
  
• Cox,K.J.A. & Adams,P.R. (2000) Implications of synaptic digitization and error for neocortical function Neurocomputing 32 673-678 Adams,P.R. & Cox, K.J.A. (2001) Synaptic Darwinism and neocortical function. Neurocomputing. In Press.