Do the 
dynamics, discussed earlier in the context of spontaneous
activity, play an active role in neural computations? We should note
that 
fluctuations appear when the network system is in a
transitional regime: the perceptual system is sensitive to small
fluctuations and thus is flexible; at the same time, the system
preserves a memory of the past stimuli in the long time correlation
tails.
In order to test this hypothesis, we measured the model's response
characteristics to localized input modulations, and to spatial
frequency (sinusoidal gratings) modulations [56],
which were applied transiently to the network. Despite the fact that
the amplitude of the modulation was small relative to the background
we find that the network has a strong amplifying response with a
relatively fast onset and a slow decay. The former is due to the
rearranging of the activity pattern in relation to the stimuli, while
the slow decay is due to the slow diffusion. In particular we find
that the 
regime accomplishes a trade-off between signal
amplification and memory; very fluid activity patterns (low
inhibition) have fast and strong amplification response but no memory,
and the opposite occurs at very rigid activity patterns (high
inhibition). Moreover, when stimulating with spatial gratings, we
find that the stronger and longer lasting response occurs at a
specific spatial frequency (determined by the intra-cluster
distance). We also find that a more complex stimulation (at two
localized spots), has a similar effect: a stronger and longer-lasting
effect when the inter-stimulus distance is compatible with the
preferred characteristic frequency of the network. Thus the network
response depends not only on the stimuli but also on spatial
relationships between stimuli (Fig. 15).
Figure 15:
The stimulus consists of raising the external input rate by 12 % of over a
circular region of radius 3. The response shown above is the
mean spike count per msec within a region of radius 5, averaged over trials.
The three traces represent different strengths of inhibition
(
, respectively). The inset displays the response to two successive stimuli at the same location.
Finally, we find that a succession of two stimuli separated by 100
msec at the same location results in a `priming effect' due to the
memory inherent in the system: the second stimulus elicits a faster
and stronger response.
Using 
signals for neural computation
is an intriguing and promising possibility deserving further research
[22].