Agreement between relatives of Parkinson's patients and clinical observer in home diary assessments.

Generalization is at the heart of visually guided behavior. We perform visual generalization every time we ignore the vast amount of irrelevant information bombarding our visual system to identify an object, feature, or location of interest. We perform behavioral generalization every time we use different ways to communicate what we see, whether by reaching for an object with one hand or another, looking at a location of interest, or writing or telling a friend what we saw. The neurophysiological basis of visual and behavioral generalization has arguably been best studied in the context of visual attention. The two theories guiding most studies (including our own) were that attention performed visual generalization by enhancing the way that task-relevant stimuli and features are encoded in visual cortex and behavioral generalization by flexibly changing the mapping from visual neurons to those involved in decision-making and motor planning. When we recorded simultaneously from a visual and a decision area while monkeys performed a task that varied spatial attention, we found that neither of those hypotheses accounted for attention-related changes in behavior. Instead, we found that the well-known effects of attention on visual cortical neurons modulated the representation so that a relatively fixed mapping between visual neurons and behavior more efficiently guides the activity of decision neurons and ultimately behavior. However, spatial attention may be a special case. Neurons that encode the same location are near each other both in the visual cortex and in pre-oculomotor neurons that have been hypothesized to be the source of spatial attention. This topographic organization may allow simple mechanisms, such as feedback from oculomotor neurons to a spatially localized group of visual neurons, to transform visual representations in the desired ways. It is less obvious how the brain accomplishes visual and behavioral generalization when the visual information of interest is not spatially organized. For example, performing when estimating a visual feature, like the curvature of stimuli that vary in other irrelevant features, involves neurons throughout a visual cortical area with different receptive fields and tuning for many relevant and irrelevant visual features. And performing when mapping one inference to many movements involves modulation by premotor or similar processes that might not be organized in the visual cortex. Although the existence of these premotor signals in sensory areas (and vice versa) is now widely appreciated, understanding how they are mixed with sensory signals and the format of the resultant activity of populations of neurons holds the key to constraining mechanistic hypotheses about how sensory activity in a single area guides behavior. Specifically for behavioral generalization, the ability to flexibly map a single perceptual estimate onto different actions depending on the behavioral context will involve analyzing the differences between visual representations resulting from different task- or context-dependent inputs to visual cortices.