TOP-DOWN MECHANISMS

Perceptual processes that are concerned solely with sensory input are often called ‘bottom-up’. But perception also depends on ‘top-down’ processes, which reflect our personal goals and past experience. ‘Bottom-up’ processes are governed only by information from the retinal image. ‘Top-down’ is a vaguer concept, since it is not clear where the ‘top’ of the visual pathway is or what it does. But ‘top-down’ certainly involves the voluntary components of perception, such as moving the eyes. For example, when we discussed conjunction search, earlier in the chapter, remember the observer moving their attention around a display to search for a target (such as a tilted red line or a blue Ford car). This kind of deployment of attention to locate a target is generated internally rather than externally, and is therefore considered to be ‘topdown’ (in contrast to, say, the sudden appearance of an object in peripheral vision, which will capture the observer’s attention and gaze automatically). Additional support for this dea comes from studies showing that selective adaptation phenomena can be affected by changes in attention. As we saw earlier in this chapter, adaptation can occur at relatively early stages of visual processing, perhaps including V1. A major finding of the anatomical studies we discussed earlier, in the section Serial versus parallel theories of perception, is that almost all the connections between the visual areas of cortex (e.g. figure 8.12) are reciprocal. In other words, information passes not only serially up the system but also backwards, from ‘higher’ regions, down towards (but not reaching) the sense organs. For example, just as area V1 projects to V2, so area V2 also sends messages to V1. How might these reverse connections mediate the perceptual functions that involve top-down influences? The idea that attention to different aspects of the world is mediated by top-down connections is supported by several recent brain scanning studies indicating that relevant regions of the visual cortices alter heir activity levels when the person is attending (Kastner & Ungerleider, 2000; Martínez et al., 2001). The idea is that ‘higher’ parts of the brain decide what to concentrate on, causing messages to be sent back down to prime the relevant parts of the visual cortex. This facilitates cell responses to expected stimuli and improves cell selectivity (tuning), so there are now increased differences in the output of a cell when it is tested with its preferred and some non-preferred stimuli (Dosher & Lu, 2000; Lee et al., 1999; Olson et al., 2001). It has been further noted that even the LGN can be affected when attention changes (O’Connor et al., 2002). Another idea is that perceptual learning, recognition and recall depend upon these top-down connections. The hippocampus is important in laying down new long-term memories. Feedback connections from the hippocampus to the cortex, and within the cortex, appear to be responsible for building these new memories into the fabric of the cortex (Rols, 1990; Mishkin, 1993; Squire & Zola, 1996). Physiological studies of cells in area V1 of the monkey support theories (Gregory, 1970; Rock, 1983) that memory for objects interacts with the early, bottom-up stages of sensory processing. So the selectivities of the cells in V1 change in the first few hundred milliseconds after a stimulus is presented (Lamme & Roelfsema, 2000; Lee et al., 1998). As activity reaches the ‘higher’ visual centres, it activates neural feedback, which reaches V1 after a delay. The latency of this feedback is caused by the limited conduction velocity of the messages along the nerve axons and by the time taken to process the information in the ‘higher’ cortical areas. Recent studies of practice on perceptual tasks indicate that the learning triggered by these feedback projections is so specific for the relevant stimuli that it can only be taking place in the ‘early’ processing areas of the visual cortex (Ahissar & Hochstein, 2000; Fahle, 1994; Lee et al., 2002; Sowden et al., 2002). Moreover, scans taken of observers’ brains when they are recalling or imagining a visual scene show activation of the same early areas of visual cortex that are activated during stimulus presentation itself (Kosslyn et al., 1993; Le Bihan et al., 1993). As indicated by this discussion, the old division between sensory and cognitive processing by early and higher neural centres has recently been replaced by a new dynamic model. Incoming sensory information interacts with task-relevant knowledge, acquired during the development of the individual concerned, and has been built into the neural network structures in several different cortical areas. Acting together, these influences create an integrated and dynamic representation of the relevant aspects of the environment (e.g. Friston & Price, 2001; Hochstein & Ahissar, 2002; Lamme & Roelfsema, 2000; Schroeder et al., 2001).