Abstract
Visual systems appear like homogenous structures, where identical functional units repeat themselves across the eye. This architecture is thought to ensure a uniform sampling of the surrounding environment. Furthermore, anatomically and functionally identical properties of single units belonging to the same cell type, yet located across retinotopical positions are thought to ensure translational invariance. At the same time, regional differences and stochastic variations in microcircuit architecture have been linked to the processing of specific visual features. Recent access to connectomic datasets has revealed heterogeneity in visual circuitry that is at odds with these criteria: Cells considered to belong to the same type are variable in number and identity of connected partners, as well as in the relative number of synapses. This variable connectivity suggests that heterogeneous computations, even within defined cell types, is the rule, rather than the exception. It is therefore an exciting question whether these network properties increase functional variability, or even functional robustness, of visual processing.