Ultra-rapid Visual Categorisation Using One Spike per Neuron

1. Introduction In 1989, I argued that the response latency of face-selective visual responses in high order visual areas such as the primate inferotemporal cortex poses severe problems for almost all models of visual processing that rely on iterative processing (Thorpe & Imbert, 1989). Such neurones start firing 80-100 ms after stimulus onset, which leaves only about 10 ms for processing at each stage of the visual pathway. The original argument could be criticised on the grounds that face processing could be a special case and that in such neurophysiological studies, the stimuli are often highly predictable. However, more recent experiments have shown that very fast processing also occurs in much more challenging visual tasks where subjects must decide whether a briefly flashed natural image contains a target category such as an animal or a means of transport (The most serious implication of such experiments is that very sophisticated visual processing must be possible under conditions where individual neurones only have time to emit one spike. Clearly, this rules out most coding schemes that require the firing rates of individual neurones to be determined with precision. However, it leaves open a number of options in which the precise timing of spikes across a population of cells is used.