Task-Driven Learning of Contour Integration Responses in a V1 Model

Salman Khan, Alexander Wong, Bryan P. Tripp

Keywords: Contour integration, V1 cortex, lateral connections

TL;DR: A biologically realistic model of contour integration embedded inside a task driven artificial neural network is presented. It is compared with observed neurophysiological and behavioral data

Abstract: Under difficult viewing conditions, the brain's visual system uses a variety of modulatory techniques to augment its core feed-forward signals. Incorporating these into artificial neural networks can potentially improve their robustness. However, before such mechanisms can be recommended, they need to be fully understood. Here, we present a biologically plausible model of one such mechanism, contour integration, embedded in a task-driven artificial neural network. The model is neuroanatomically realistic and all its connections can be mapped onto existing connections in the V1 cortex. We find that the model learns to integrate contours from high-level tasks including those involving natural images. Trained models exhibited several observed neurophysiological and behavioral properties. In contrast, a parameter matched feed-forward control achieved comparable task-level performance but was largely inconsistent with neurophysiological data.