“We must consider what we mean when we say that the spiking activity of a neuron 'encodes' information. We normally think of a code as something that conveys information from a sender to a recipient, and this requires that the recipient 'understands' the code. But the spiking activity of every neuron seems to encode information in a slightly different way, a way that depends on that neuron's intrinsic properties. So what sense can a recipient make of the combined input from many neurons that all use different codes? It seems that what matters must be the 'population code' - not the code that is used by single cells, but the average or aggregate signal from a population of neurons.
In a now classic paper, Shadlen and Newsome considered how information is communicated among neurons of the cortex - neurons that typically receive between 3,000 and 10,000 synaptic inputs.They argued that, although some neural structures in the brain may convey information in the timing of successive spikes, when many inputs converge on a neuron the information present in the precise timing of spikes is irretrievably lost, and only the information present in the average input rate can be used. They concluded that 'the search for information in temporal patterns, synchrony and specially labeled spikes is unlikely to succeed' and that 'the fundamental signaling units of cortext may be pools on the order of 100 neurons in size.' The phasic firing of vasopressin cells is an extreme demonstration of the implausibility of spike patterning as a way of encoding usable information, but the key message - that the only behaviorally relevant information is that which is collectively encoded by the aggregate activity of a population - may be generally true.”