Supervised learning in the brain.

Experience shapes the functional organization of the brain, optimizing and customizing its properties for the individual and his or her environment. One way that experience shapes the constituent networks ofthe brain is through supervised learning. In supervised learning, information from one network of neurons acts as an instructive signal to influence the pattern of connectivity in another network. As a result, the instructed network learns to process information so that a particular goal or transformation specified by the instructive signal is achieved. In so doing, supervised learning establishes patterns of connectivity efficiently and with a precision that does not need to be and, often, cannot be encoded in the genome. Supervised learning contributes to the development and maintenance of a variety of brain functions. For example, sensorimotor networks that control goal-directed movements are calibrated by sensory feedback indicating the accuracy with which the movements are made. In a specific example that will be discussed at some length, a visual instructive signal, indicating the slip of images across the retinae, is used to calibrate the transformation of vestibular sensory information (indicating rotation of the head) into precise, compensatory movements of the eyes that stabilize the images on the retinae (Miles and Eighmy, 1980). Supervised learning can also control the representation of information in sensory networks. For example, in the development of binocular neurons in the optic tectum of the frog Xenopus, visually driven activity from the contralateral eye specifies the topography of the visual map originating from the ipsilateral eye (Gaze et al., 1970; Udin, 1985). In this example, which also will be discussed in detail, the activity from the contralateral eye provides an instructive signal that assures the mutual alignment of leftand right-eye receptive fields. It is likely that supervised learning also contributes to the establishment of networks that support certain cognitive skills, such as pattern recognition and language acquisition, although there is, as yet, no experimental confirmation of this proposition. This article discusses supervised learning as it might be implemented in the brain. Different kinds of instructive signals,