Perceptual learning, or the adjustment of perception to the stimulus aspects of the environment, is sometimes distinguished from cognitive learning, the latter term being reserved for the modification of problem-solving behavior (Walk, 1978). In early vision, learning occurs in processes such as adaptation, mentioned above. The famous experiment first made by Stratton (1897/1964) , in which a subject wearing inverting prisms gradually adapts to this condition, is a forceful reminder that the degree of plasticity at lower levels of the visual system should not be underestimated (see Rock, 1984, for a discussion of the inversion experiments).
Exactly what is learned and what is innate in vision (and in cognition in general) has been the subject of intense philosophical debate since Plato's time (see Dretske, 1990, for an overview). A century of research in visual psychophysics and neurobiology of vision shows that the basic perceptual abilities of the human visual system (such as the ability to perceive luminance contrast) are largely innate, while others (such as some varieties of object constancy) are acquired and depend on the visual experience (Spelke, 1990). Significantly, the mechanisms of perceptual organization used by infants in learning how to see seem to persist through adulthood. Thus, the study of the ontogeny of visual perception may help clarify the nature of the long-term memory representations of objects and scenes.
A similar angle on the problem of representation is provided by studies in which the subjects' perceptual performance in three-dimensional object recognition is modified merely as a result of practice or exposure to the stimuli, without any feedback from the experimenter (after all, infants acquire vision --- and language --- without being instructed). Normally, the subjects' response time in recognition depends monotonically on the misorientation of the stimulus with respect to some canonical attitude (Palmer et al., 1981; cf. the discussion of mental rotation in the section on Methods). Perceptual learning under such circumstances, inferred from the observed changes in the pattern of performance (specifically, from increasingly uniform response times for different aspects of the stimulus), can be attributed to a shift towards a more memory-intensive and less time-consuming recognition strategy (Tarr and Pinker, 1989; Edelman and Bülthoff, 1992). Indeed, such a strategy appears to be the most suitable one for a system in which memory is cheap, but time is expensive.