Neural Computations In Visual Cortex

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): We aim to advance our understanding of the nature of the operations performed by primary visual cortex, and how these operations are carried out. The research plan is organized around two basic questions. A. Are the spatial computations performed by neurons in V1 adequately described by oriented receptive fields, simple nonlinearities, and gain controls, or, conversely, do more complex spatial operations play an important role? B. Can firing rate of a local population be considered the main carrier of visual information, or, conversely, is the fine structure of neural activity (across time or across a local population) also important? Question A is motivated by growing experimental evidence that the operations performed by V 1 neurons are indeed complex. The experimental strategy described here uses new stimuli, precisely balanced in space and spatial frequency - in contrast to traditional stimuli, that are highly localized in space but broad in spatial frequency (e.g., spots and bars), or extended in space but highly localized in spatial frequency (e.g., gratings). Pilot studies using the new stimuli demonstrate response patterns not anticipated from standard models - including shifts in the balance of linear and nonlinear influences, and shifts in orientation tuning. The stimulus set is a hierarchy of patterns that are progressively less restricted in their combined space bandwidth aperture. The patterns that are the most restricted in space-bandwidth aperture are Gaussians and profiles that resemble edges and lines. As the space-bandwidth aperture is broadened, intrinsically two-dimensional patterns and "non-Cartesian" patterns emerge, thus suggesting that our approach will also be useful beyond V1. Question B builds on our identification of the statistical features of the fine structure of neural activity that are reliably correlated with visual stimuli. We now propose to determine whether these features indeed influence the activity of downstream neurons. Our approach relies on a combination of multiple-neuron recordings via tetrodes, and experimental results and analytic tools that emerged during the previous funding period. Together, these investigations will advance the understanding of neural computations that lead to extraction of features and objects from the visual image.