Figure 2.

Experimental predictions for efficient coding with a heterogeneous population of unimodal tuning curves. (a) Hypothetical example of a probability distribution over a sensory attribute, p(s). (b) Five tuning curves of a neural population arranged to maximize the amount of information transmitted about stimuli drawn from this distribution. (c–e) Predicted shapes of experimentally accessible attributes of the neural population, derived from the prior distribution using Eq. (14). (c) Histogram of the observed preferred stimuli (stimuli associated with the peaks of the tuning curves) provides an estimate of local cell density, d(s), which should be proportional to the prior distribution (black line). (d) Tuning widths of the neurons (measured as the full width at half maximum of the tuning curves) should be inversely proportional to the prior (points correspond to example neurons from (b)). (e) The gain, g(s), measured as the maximum average firing rate of each of the neurons, should be constant (points correspond to example neurons from (b)). (f) Minimum achievable discrimination thresholds of a perceptual system that relies on this efficient population are inversely proportional to the prior distribution (Eq. (16)).