5 ms·
> What matters most is not precisely when a particular neuron spikes but how often it does; the rate of firing is the main currency. I'm not too sure about tha
by ivan_ah 12y ago
> What matters most is not precisely when a particular neuron spikes but how often it does; the rate of firing is the main currency.
I'm not too sure about that.
My gut feeling is that at least half of the information in the brain is encoded in the exact timing of the spikes. In a model that takes time into account, certain neurons function as a "coincidence detector" sensitive to certain events.
e.g. your brain deduces which direction a sound is coming from by analyzing the delay difference for the sound input from each ear to arrive to the brain. Specifically, sound-orientation detector neurons read from neurons of different lengths to each ear. You know the sound is coming from front-and-left if the coincidence-detector neuron that has a short lead to the left ear and a long lead to the right ear fires.
- joeyo 12y agoThere are clear examples where spike timing matters a great deal and the Jeffres model of sound localization, as you point out, is a great example of this. However, spike timing codes seem to be more prevalent in the periphery; rate codes dominate in central structures, and in particular the cortex. The receptors of sensory afferents faithfully report the timing of individual events (of photons, pinpricks of the skin, etc) with single spikes. But after a few synapses the pattern of spikes looks much more like a (noisy) rate code. The origin of the noise, whether it even is noise, how the temporal code is transformed into a rate code, and many other questions along these lines are active research topics. All that aside, Koch is well aware that there is evidence for both rate and temporal codes. As he wrote, "spikes mean different things in different contexts."