SAIM: A Model of Visual Attent ion and Neglect

Abstract. This paper examines the reason for a particular impairment of cognitive functioning in brain-damaged patients called visual neglect. To achieve this goal a Selective Attention Identification Model (SAIM) was developed which performs translation-invariant object recognition. SAIM uses a constraint satisfaction r utine based on a continuous Hop-field network to map an object into a focus of attention. The simulation results show that SAIM is a successful model of visual attention and visual neglect. 1 In t roduct ion There is a growing impact of neuropsychological findings on the understanding of the cognitive functioning of the brain. (e.g [1]). Neuropsychology is mainly concerned with cognitive deficits of brain-damaged patients. In this article we focus on examine the reason for a particular impairment called "visual neglect". The term "visual neglect " is used to refer to brain-damaged patients who fail to respond appropriately to stimuli presented on the side of space contralateral to their brain lesion. They fail to eat food on one side of an object, to cancel ines on one side of a sheet, to draw one half of an object or to read words on one side of a text. Classically, neglect is related to lesions of the right parietal obe [4]. In order to examine how visual neglect might emerge following damage to an object recognition system, we developed a model called SAIM (Selective Atten-tion Identification Model, Fig. 1), which aims at a translation-invariant object recognition [2] [5]. It does this by mapping from locations on a retina through to a smaller "attentional " window, the Focus of Attention (FOA), with activation within the FOA providing the input to an object recognition system. This ap-proach is similar to the model of [7], which focused on anatomical issues, whereas our work concentrates on psychological and neuropyschological modelling. 2 The Network The architecture of SAIM is illustrated in Fig. 1. It shows three different subnet-works: the contents network, the selection network and the knowledge network. These networks will be introduced in this section