Abstract
Problem: Cells at early stages of the visual processing hierarchy, as in cortical area V1, have small receptive fields and thus respond to items within a restricted region of the visual field. Perception, however, is influenced by global spatial and temporal effects, such as context, figure-ground interpretation, or working memory (van Kerkoerle et al., Nature Comm, 2017; Poort et al., Cerebral Cortex, 2016). The neural mechanisms underlying these processes are not yet understood. Method: We propose a computational model of visual cortex which is represented as an array of columns with granular (input), superficial, and deep compartments. Each compartment comprises excitatory, inhibitory, and dis-inhibitory units with continuous dynamics specified by first-order ordinary differential equations. The granular compartment receives bottom-up input and performs initial filtering. This activation is fed to cells in the superficial compartment to utilize lateral long-range interactions. The resulting activation is further progressed in an intra-cortical loop, activating cells in the deep compartment which subsequently modulate the input cells. Superficial and deep compartments additionally receive top-down modulating signals from other areas. Model simulations and theoretical analyses are performed for different stimulus and contextual conditions. Results and Conclusion: The model has been probed for stimulus conditions of figure-ground modulation and attentional target object selection. In particular, model parametrizations were fit to an attentional version of a curve-tracing task (Roelfsema & Houtkamp, Atten Percept Psychophys, 2011). It is demonstrated that the model dynamics successfully predict the multiunit activity in a working-memory version of the task. Our results suggest that a single mechanism, involving canonical cortical operations such as input filtering, modulating feedback, and normalization, implemented in different V1 compartments, can explain the laminar profile of activity in visual cortex across a wide variety of tasks.