Abstract
The ‘subunits’ that provide input to direction selective (DS) complex cells in macaque V1 have not been identified. Various experimental protocols have been used to characterise these subunits, and one enduring method has been that of examining the responses to pairs of stimuli that are offset in space and time. The resulting spatial interaction maps are often interpreted in the context of models based on linear-filters, although recent results have shown that the interaction maps can take asymmetric forms that are inconsistent with most common filter-based models (Livingstone and Conway, 2003, J Neurophysiol). Thus, connecting the physiological characteristics of these subunits with the sets of spiking inputs and circuitry present in V1 remains an unsolved problem. To bridge this gap, we have developed and compared two filter-based models of DS complex cells - a Reichardt detector and motion energy model - with several configurations of a physiologically realistic network model that incorporates populations of spiking units representing inhibitory and excitatory V1 simple cells and ON and OFF LGN inputs. One set of simulations we have performed to compare these models generates maps of 2D directional interactions. The interaction maps for the motion energy model had symmetrical elongated facilitory and suppressive subregions. In contrast, the Reichardt detector model generated round, symmetric suppressive and facilitory map subregions. Neither model generated curved, asymmetrical interactions that have been observed experimentally. However, the network model was able to generate DS units with both elongated and round maps as well as maps with curvature in which the suppressive subregion was reduced. All of these map shapes can be obtained in the model by simply varying the orientation distribution of inputs to the DS units. In continuing work on these models, we are using other protocols to further explore how DS complex cell receptive fields may be built.