Neural Plasticity: Single Neuron Models for Discrimination and Generalization and an Experimental Ensemble Approach.

摘要

A special form for modification of neuronal response properties is described in which the change in the synaptic state vector is parallel to the vector of afferent activity. This process is termed 'parallel modification' and its theoretical and experimental implications are examined. A theoretical framework has been devised to describe the complementary functions of generalization and discrimination by single neurons. This constitutes a basis for three models each describing processes for the development of maximum selectivity (discrimination) and minimum selectivity (generalization) by neurons. Strengthening and weakening of synapses is expressed as a product of the presynaptic activity and a nonlinear modulatory function of two postsynaptic variables--namely a measure of the spatially integrated activity of the cell and a temporal integration (time-average) of that activity. Model neurons that achieve high selectivity mimic the development of cat visual cortex neurons in a wide variety of rearing conditions. A role for low-selectivity neurons is proposed in which they provide inhibitory input to neurons of the opposite type, thereby suppressing the common component of a pattern class and enhancing their selective properties. Ramifications of parallel models for certain experimental paradigms are analyzed.