The Hebrew University
Usually, it is assumed that the spike trains represent stochastic point processes, for which the main parameter are the (possibly variable) firing rates. However, when evaluating the relations among the firing times of several neurons, it is common to find very precise timing. Activity of 8-15 single units was recorded in parallel from the frontal cortex of monkeys performing a variety of localization and problem solving tasks.
Patterns of firing in which several (3 or more) spikes appear repeatedly with the same intervals are called here spatio-temporal firing patterns. The present study refer to spatio-temporal firing patterns lasting for up to 450 ms, and in which the firing time is accurate to within ±1 ms.
In many cases the spatio-temporal firing patterns were associated with the monkey's behavior. In some cases, the entire response of a single unit was composed of such accurate firing patterns. The internal time-locking among spikes of any given spatio-temporal firing pattern were tighter then the time- locking of the pattern to the external stimulus (±1 ms versus ±50 ms). Often, the spikes which were involved in spatio-temporal firing patterns conveyed more specific information about the stimulus or the response as compared to all the spikes of a single unit. Some patterns may appear when a stimulus is delivered in one behavioral context, while others may appear when the same stimulus is delivered in another context.
In many cases, the pair-wise cross-correlation between spikes that took part in spatio-temporal firing patterns was qualitatively different then the cross-correlation between spikes of the same neurons that did not take part in spatio-temporal firing patterns. This difference may be due to the tendency of different spatio-temporal firing patterns to appear in "bursts" lasting 0.5-2 seconds.
A simple neural network that can generate such patterns is the syn-fire chain. Insight of the properties of syn-fire chains was gained by simulating networks of neurons in which chains of feed-forward, diverging converging connections were laid. The simulations revealed that such chains convert diffuse (asynchronous) excitation into synchronized volleys of spikes that can propagate over many links with very little jitter in time. According to this view the relevant physiological parameter is the sequence of synchronously firing pools of neurons. The recorded spatio-temporal pattern merely indicates to the experimenter that a certain syn-fire chain became active. In the experimentally-recorded spatio-temporal firing patterns, we found many cases in which two or more spikes from the same single unit participated in one pattern. This suggested that a given neuron may take part in more then one link of the syn-fire chain.
The experimental data and the simulations suggest that cortical activity is generated and maintained by reverberations in syn-fire chains, and that binding between such syn-fire activities is the mechanism for generating a hierarchy of representations.