Item

Abstract

Common estimates of information processing capabilities such as the dynamic range are defined in terms of expected values
that do not incorporate effects from finite processing time. However, in real-world tasks, animals and humans have only a
limited amount of time to process input and make a decision. Here, using a stochastic neural network with integrate-and-fire
units as an example, we propose new definitions for dynamic range and perceptual resolution that explicitly include
uncertainty arising from the finite processing time. We evaluate these measures both analytically and using simulations and
find that the network's ability to discriminate inputs within a finite time is maximized in a subcritical regime. Interestingly, with
a longer processing time, the optimal regime moves closer to criticality (Fig.1d-e). This result highlights the importance of
incorporating the constraints of finite processing time when studying the information processing capacity of any system in
noisy real-world situations. (for a decent-quality figure visit
https://www.dropbox.com/s/d7fazthbgoa1hr5/FENS_highQuality.png?dl=0)