Abstract
Attention has been increasingly paid to the question of how the train of action potentials, of a neuron, code the message accumulated in its dendrites. The resolution of this question is important for understanding the roles of particular visual neurons in the sensory pathway to and beyond striate cortex.
Two recent attempts to study the problem have posited unique roles for the noise or variability in the neural message. The first of these concerns the ‘iceberg’ phenomenon that assumes a high threshold so as to escape the deleterious effects of the noise. In some cases this is thought to sharpen the tuning characteristic of a neuron that codes spatial information.
The second type of study has examined the question of deciding between two mutually exclusive theories of coding: (i) that the fine timing structure of neural firing contains the relevant information, or (ii) that the information is to be found in small quantity in the overall spike count in each of a subpopulation of neurons.
We present calculations in a signal detection framework that show the deleterious nature of the iceberg effect upon detection and also upon identification for a model neural system. More is lost than is gained by this strategy. We also show retinal ganglion cell responses that manifest another plausible coding theory and which reject the two mentioned above. Neural responses, in the form of action potential count, have been observed to be so sensitive that all of the information must reside in that response, which means that, at least in the retina, (i) the timing of spikes is almost entirely irrelevant, and (ii) a single neuron possess and transmits all of the information available in its receptive field.