Search Results

This chapter presents the theory of visual attention (TVA). It begins with a review of choice models for visual recognition (categorization) and for visual search (partial report). Roughly speaking, these models are non-process models; they provide descriptive equations with strong empirical constraints, but make little or no attempt to specify the temporal course of the information processing underlying performance. It then considers a race model for selection from multi-element displays. The race model is a process model; it specifies temporal characteristics of processing. Further, the race model is perfectly consistent with the descriptive-choice model for visual search, which can simply be derived from it mathematically. Finally, the unified theory of visual recognition and attentional selection (TVA) is developed by integrating choice models for recognition into the race model framework. In a mathematical sense, TVA includes the previous models as special cases, and hence inherits their success in accounting for empirical findings. TVA is not only a process model, but is also computational; it specifies the computations by which selection is supposed to be done.

The neural theory of visual attention (NTVA) interprets attentional selection at the level of individual neurons. The interpretation was suggested by studies of single-cell activity in monkeys, and this chapter shows how NTVA accounts for a very large part of the findings in this field. Since the original presentation of the model by Bundesen et al. (2005) important new studies have been published, and these are considered along with the older studies. The empirical review in this chapter includes more than 20 central studies and covers all major effects reported in the single-cell attention literature. To illustrate the precision of NTVA, detailed mathematical fits to a few studies are presented.

This chapter discusses four theories of attention and how studies of visual processing and attentional selection in primates have largely shaped these theories. The first theory is the feature-integration theory of Treisman and colleagues. It then describes its offspring (the guided-search model and the ambiguity-resolution theory). Finally, it discusses the biased-competition account of selection and the premotor theory of attention.

This chapter presents a model of working memory and change detection in visual scenes that addresses the binding problem: When memorizing a group of objects, each with a location and multiple surface features, how are the properties of each individual object bound together and kept separate from the properties of other objects? In the proposed architecture a stack of feature maps forms the neural substrate for scene working memory. Binding makes use of a spatial dimension shared by all feature maps. The operations of attentional selection and space-feature integration act jointly to memorize multiple objects in a bound fashion and detect changes in visual scenes. The need for attentional selection in some task implies sequential processing of individual items, whereas in other tasks, items can be processed in parallel. This aspect of the model’s behavior provides qualitative predictions about human performance.

This chapter expands the concepts of dynamic field theory (DFT) to activation distributions over multidimensional spaces. Multidimensional fields can be used either to represent inherently multidimensional features or for bringing different feature dimensions together in a single, integrated representation. Such integrated representations are costly in terms of neural resources, but greatly increase the behavioral repertoire of dynamic field (DF) models by enabling interactions between different feature spaces. Such interactions are demonstrated here in a model of attentional selection in early visual representations, which combines separate one-dimensional and integrated multidimensional DFs. The DF architecture accounts for visual search and touches on the binding problem in visual cognition. The chapter concludes with expansions of the basic architecture that deal with the effects of visual working memory on attentional processing and an explanation of the origin of illusory conjunctions.

This chapter discusses the changes that occur in the brain when a motor task is learned and then practiced until it has become automatic. This chapter describes two studies conducted to understand this mechanism, firstly by Jenkins et al and secondly by Jueptner et al using a more sensitive scanner. They tested the subjects using a dual task paradigm by checking whether the task can be performed with minimal interference at the same time as another task. It proposes that the more routine a task, the less the prefrontal and anterior cingulate cortex will be activated. The chapter also explores the interference that occurs when subjects must attend to the selection of more than one action at the same time, and conclusively states that that attentional selection occurs ‘late’ rather than ‘early’.