Fig. 3: Schematic showing generalized linear model (GLM)-based Bayesian decoding.
From: Ignoring correlated activity causes a failure of retinal population codes
a One stimulus pixel, highlighted in red, is chosen over several frames (left), yielding a sequence of intensity values (right). b The corresponding response of a single retinal ganglion cell (RGC) (top) or population of RGCs (middle and bottom) are extracted. c The probability of each possible stimulus given the input response is computed under a GLM fit to that population. d Summing over the possible stimuli weighted by their probabilities gives the optimal Bayesian estimate of the stimulus. In general, each GLM provides a different estimate of the original stimulus.
