engleski

A Neurodynamical Model of Visuo-spatial Selection

Huang and Pashler [2007, Psychological Review, 114(3), 599-631] showed that observers are able to simultaneously select all spatial locations occupied by a single feature value (e.g., red) per dimension (color). They suggested that visual system creates a Boolean map, that is, a spatial representation which partitions visual scene into two distinct and complementary regions (selected and not selected). The aim of the present work is to develop a recurrent neural network with the ability to select multiple visual objects simultaneously based on a shared feature value. The basic computational elements of the network are two types of inhibitory interneurons which mediates lateral and dendritic inhibition. Lateral inhibition implements competition between locations, while dendritic inhibition enables spatial grouping and feature-based selection. Interactions between lateral and dendritic inhibition result in a formation of a spatial map where maximal firing rate is assigned to selected feature value while neural activity at other locations is suppressed. Computer simulations showed that the proposed neural network is able to create a Boolean map and to elaborate it using logical operations of intersection and union. The proposed network provides a neural implementation of the Boolean theory of visual attention.

pažnja; odabir; neuronski model

nije evidentirano