Canonical computations of cerebral cortex

Abstract

The idea that there is a fundamental cortical circuit that performs canonical computations remains compelling though far from proven. Here we review evidence for two canonical operations within sensory cortical areas: a feedforward computation of selectivity; and a recurrent computation of gain in which, given sufficiently strong external input, perhaps from multiple sources, intracortical input largely, but not completely, cancels this external input. This operation leads to many characteristic cortical nonlinearities in integrating multiple stimuli. The cortical computation must combine such local processing with hierarchical processing across areas. We point to important changes in moving from sensory cortex to motor and frontal cortex and the possibility of substantial differences between cortex in rodents vs. species with columnar organization of selectivity.

Figure 1

Modulation of gain in a simple model circuit. Reproduced from [103^••]. (a). 180 excitatory (E, red) and inhibitory (I, blue) units, with coordinates θ on a ring thought of as preferred orientations. Lines schematize connections. A stimulus grating evokes input ch(θ) equally to E and I units, with h(θ) a unit-height Gaussian centered at the stimulus orientation with standard deviation σ_FF = 30°. We consider gratings at 45°, 135° or both simultaneously. (b). The power-law input/output function: k = 0.04, n = 2.0. y represents a cell’s firing rate, x its net input, [x]₊ is x with negative values set to zero. (c–f) use a single-grating stimulus. (c,d). For E (c) and I (d) units at stimulus center: with increasing external input strength c (x-axis; dashed lines), network input (E, red; I, blue) transitions from weak to dominating (insets) and substantially cancels external input, so net input (green) grows slowly. Firing rates (black) are proportional to net input squared. (e,f). Summing input received by all E (red) or I (blue) units: with increasing c, input to network (sum of absolute values of E and I input) is increasingly network-driven (E; dashed, external input; solid, network input) and network input is increasingly inhibitory ((f); E_N/(E_N + I) where I, E_N are inhibitory and network excitatory input respectively). (g). Sublinear response summation for multiple stimuli. Top two rows: responses of E (left, red) and I (right, blue) units across network to 45° (top) and 135° (2nd row) stimulus, c = 50. 3rd row: responses to both stimuli presented simultaneously. 4th row: responses from 3rd row (black) vs. mean (orange) and linear sum (green) of responses to the two individual stimuli. Gray: rows 1–2, best fit of Gaussians to response; row 3, best fit of weight W (indicated) times sum of fits from rows 1–2.