Search Results

The ability to hold visual information in mind beyond the duration of the initial sensory stimulation critically underpins many higher-level cognitive functions. In particular, visual short-term memory (VSTM) provides the perceptual continuity that is necessary for visual information to guide behavior across short temporal delays. This chapter explores how the mechanisms of attention optimize VSTM. First, it considers how top-down attention biases VSTM encoding to favor information that is most likely to be relevant to behavior. Next, it looks at more recent evidence that top-down attention can also bias representations already stored within VSTM. Flexible allocation of attention within VSTM enables the visual system to prioritize and update stored representations to accommodate changing task demands.

This chapter presents an overview of the research on visual attention, which has been studied extensively in cognitive psychology and neuroscience. The studies discussed are limited to the major empirical findings on visual attention that have implications for the scientific understanding of consciousness. The chapter includes studies on feature-based attention, spatial attention, object-based attention, effortless attention, the mechanisms supporting the different forms of attention (e.g., neural structures and pathways), and the evolution of these mechanisms. This review is important for the book’s primary argument that consciousness and attention must be dissociated at some level, as there are functionally different forms of attention that seem to operate independently and to have evolved at different times from each other—an argument that is difficult to make for consciousness.

One mechanism of attention is bottom-up, and uses saliency to direct overt attentional mechanisms such as eye movements. A second mechanism of attention is top-down, in which the attentional effects can be covert, for example without eye or head movements. The architectures in Figs. 6.2 (showing a biased competition architecture for attention) and 6.21 (showing a biased activation architecture for attention) illustrate some key principles of operation of the cortex, used here for top-down attention. One principle is the function of cortical recurrent collateral connections to implement short-term memory attractor states to hold the object of attention active and on-line (Chapter 4). A second principle is the use of competitive networks to select winning populations of neurons that provide a useful output to the next stage of processing (Sections 7.4 and B.4). A third principle is allocation of cortical resources to respond to current processing needs efficiently, for example by using selective attention to select form vs location as currently relevant to make this processing faster and more efficient, and to minimize distraction by currently irrelevant information. This enables an individual with the benefit of short-term memory systems in the prefrontal cortex to focus on a task, to devote cortical resources efficiently to that task, and to not be too distractible by other stimuli. These operations are key to executive function, a key function of the prefrontal cortex which describes selecting a task, and focussing on that task without being continually distracted. In humans, this may include multi-step syntactic plans, as described in Chapter 22 and Section 5.6.10. A fourth principle is that whole cortical areas and streams may be biased by top-down attention. Examples of this include the biasing up of activity in the reward processing stream including the orbitofrontal cortex and anterior cingulate cortex when attention is to reward-value and pleasantness; and of other processing streams such as insular cortex regions for taste when attention is to the sensory attributes of stimuli, such as identity and intensity. A new biased activation theory of attention provides a model of this (Fig. 6.21). A fifth principle is that cognition even from the highest, word, level can have top-down effects that bias processing in areas such as the orbitofrontal cortex where reward value and subjective pleasantness are first reflected explicitly in the representation. (By made explicit in the representation is meant that the stimulus or event can be decoded easily from the firing rates using for example dot product decoding, as described in Appendix C.) A sixth principle is that there is in addition a system for bottom-up attention whereby salient sensory stimuli elicit orienting, for example eye movements and/or turning of the head. Part of this mechanism is implemented in the dorsal cortical visual stream, which operates in conjunction with collicular (midbrain) mechanisms. This Chapter on attention illustrates how modern computational models of interactions between cortical networks offer precise explanations of how the cerebral cortex operates to implement many complex computations that are very important in cognitive function.

Cognition does not work without attention. Attention enables us to focus on particular tasks and particular aspects in the environment. Psychological insights show that attention exhibits bottom-up and top-down components. Attention is attracted from the bottom-up towards unusual, exceptional, and unexpected sensory information. Top-down attention, on the other hand, filters information dependent on the current task-oriented expectations, which depend on the available generative models. This computational interpretation enables the explanation of conjunctive and disjunctive search. Different models of attention emphasize the importance of the unfolding interaction processes and a processing bottleneck can be detected. As a result, attention can be viewed as a dynamic control process that unfolds in redundant, neural fields, in which the selection of one interpretation and thus the processing bottleneck is strongest at the current focus of attention. The actual focus of attention itself is determined by the current behavioral and cognitive goals.

The prefrontal cortex receives perceptual information from the temporal and parietal cortices, and is in a position to perform ‘off-line’ processing, including holding items in a short-term memory when the items are no longer present in the input processing streams. This off-line capacity develops into a capability of manipulating and rearranging items in short-term memory, and this is called working memory, which is also implemented in the prefrontal cortex. This ability in humans develops into systems that can plan ahead, and then can control behaviour according to such plans, which is referred to as ‘executive function’. Attractor networks are fundamental to understanding the functions of the prefrontal cortex in short-term and working memory; and in providing the source of the top-down bias in top-down models of attention

Cortical backprojections do not transfer the signal computed in one cortical area back to the preceding cortical area, in that for example the responses of the neurons in the primary visual cortex are quite different from those in the inferior temporal visual cortex. Cortical backprojections are used for memory recall, for example in the backprojections from the hippocampus. To enable the number of memories recalled to be high in number, the backprojections must be high in number, as the number of memories recalled depends on the number of synapses from the backprojections onto each cortical neuron in the preceding cortical area (see Chapter 24). This remains the only quantitative argument for why there are as many backprojections as forward projections between two adjacent areas in a cortical hierarchy. To implement recall, the backprojections synapses must be associatively modifiable in at least some of the stages, in order to act as a pattern associator and achieve good recall capacity (see Chapter 24). Cortical backprojections can also be used for top-down attention, operating by biased competition or by biased activation.

In this chapter, Toto chases a cat, so the Wizard of Oz goes home alone; a woman is attacked by her own hand; and a mouse plunders a pantry. But mainly we consider memories that evolved in early mammals. These ancestors developed a new part of the brain called the neocortex, and it changed memory forever. The neocortex endowed early mammals with an ability to manage contradictory memories, each appropriate in certain contexts. Obsolete memories could be retained and used in case the ‘good old days’ returned, and new memories received a boost when it mattered most. The neocortex empowered early mammals to strike a balance among alternative ways of surviving: spending versus conserving energy; acting habitually versus making good choices; and acting with urgency versus patience.