Correlated Neuronal Activity: High- and Low-Level Views - Computational Neuroscience: A Comprehensive Approach - page 339

problem is the complexity of the total driving input, which is mediated by thousands
of synaptic contacts [14, 91]. What attributes of this input will the postsynaptic re-
sponse be most sensitive to? Correlations between synaptic inputs are crucial here
because they shape the total input and hence the postsynaptic response. Determining
what exactly is their role is a key requisite for understanding how neurons interact
dynamically and how the timing of their responses could be used for computational
purposes.
This chapter reviews work related to both perspectives on correlated activity: the
high-level approach at which function serves to guide the analysis, and the low-level
approach that is bound to the biophysics of single neurons. Eventually (and ideally),
the two should merge, but currently the gap between them is large. Nevertheless,
comparing results side by side provides an interesting panorama that may suggest
further clues as to how neurons and neural circuits perform their functions.

12.2 Functional roles for spike timing

There is little doubt that the correct timing of action potentials is critical for many
functions in the central nervous system. The detection of inter-aural time differences
in owls and the electrosensory capabilities of electric fish are two well-known exam-
ples [21]. In these cases it is not surprising that timing is important; it is ingrained
in the nature of the sensory signals being detected. The issue of timing also arises
naturally in the rodent somatosensory system [7]. To explore their surroundings, rats
move their whiskers periodically. To locate an object, whisker deflections need to be
interpreted relative to whisker position, which can be determined from the phase of
the motor signal. Thus, the latencies of stimulus-evoked responses relative to such
internal signal can be used to encode spatial information. This mechanism by which
sensory-triggered activity is interpreted relative to an internal, reference signal may
be applicable to other circuits and in a more general way [6, 57].

12.2.1 Stimulus representation

Spike timing, however, has been discussed in an even wider sense than implied by the
above examples. No doubt, this is partly because oscillations at various frequencies
and synchronous activity are so widespread [8-86]. One proposal that has received
considerable attention is that the coordinated timing of action potentials may be ex-
ploited for stimulus representation [71-44]. Specifically, neurons that have different
selectivities but fire synchronously may refer to the same object or concept, binding
its features. The following experiment [51] illustrates this point. The receptive fields
of two visual neurons were stimulated in two ways, by presenting a single object, and
by presenting two objects. Care was taken so that in the two conditions practically
the same firing rates were evoked. The synchrony between pairs of neurons varied