Fast dynamic organization without short-term synaptic plasticity: A new view on Hebb's dynamical assemblies
Introduction
Beyond the level of single spiking neurons, there still exists no serious computational theory about the manner in which the brain processes information. This big “how” of neuronal processing comprises two subquestions. First, what is the basic neural code for information processing? Second, what are the basic computational units that exploit this code? The latter question relates to the dynamics of the basic computational units and the particular properties that allow them to process the coded information. This means that representation is what matters, but only in the context of a defined machinery that makes use of it. An example: It has been claimed that “labeling” by synchronicity is a way of coding a solution of the superposition catastrophe, and, eventually, of the binding problem [7]. The claim does not tell us, however, how the solution comes about, an argument that has been used lately as a criticism to refute this kind of synchronicity as being able to solve the binding problem [5]. This issue remains controversial, but fact is, that the claim would only answer the first question of the preceeding paragraph. The problem of the machinery that is able to generate the solution remains unanswered. In this work we show how a particular code can be used in a simple network, and how such a network can generate the code leading to “dynamical assemblies” by internal organization.
Section snippets
Hebb's dynamical assemblies
We assume that, in some areas of the cortex, the firing of a single, isolated neuron is not a meaningful basic code for cortical information processing. Rather, we assume that neurons can be assigned to segregated, non-overlapping neuronal pools (each containing a number of neurons) and these in turn can be treated in a model as an identifiable processing unit using a macroscopic description (see e.g. [1]). The pools are static in the sense that they are defined by a network architecture with
Simulations
With the purpose to model Hebb's dynamical assemblies we use a simple minimal model for the pool activity dynamics and present a small network model composed of interconnected neuronal pools. The network is shown schematically in Fig. 1. The dashed and solid bars in the figure indicate the processing units (i.e., the neuronal pools) that are sensitive to local input consisting of bars of that particular orientation, i.e., they are simple orientation-selective feature detectors as those
References (7)
- et al.
Unifying framework for neuronal assembly dynamics
Phys. Rev. E
(2000) The Organization of Behavior
(1949)- et al.
Extracting oscillations: neuronal coincidence detection with noisy periodic spike input
Neural Comput.
(1998)
Cited by (2)
A Hypothesis About the Biological Basis of Expert Intuition
2011, Review of General PsychologyAn STDP training algorithm for a spiking neural network with dynamic threshold neurons
2010, International Journal of Neural Systems