Central Processing of Visual Information

Weill Medical Coll Of Cornell Univ, New York NY

Investigators

Linked publications & trials

Abstract

? DESCRIPTION (provided by applicant): The overall goal of the research is to determine how early vision represents sensory information in a way that supports extraction of features and objects, and, ultimately, inferences about the visual world that guide action. The proposal will determine how complex, multidimensional perceptual spaces are represented at the algorithmic level and how these representations are used. Based on previous work, the perspective is that the representation of a perceptual space is tuned to operate efficiently, given the statistics of natural sensory inputs. The research will now determine how this is achieved within the constraints of neural hardware, in a manner than can be easily read out by later stages of processing. Among the many well-recognized perceptual spaces (for example, color, faces, and auditory textures), we choose to focus on simple form elements (e.g., lines, edges, corner), shading, shape, and motion signals (Fourier, non-Fourier, and others we recently discovered). Several reasons underlie these choices: (i) they are crucial to visual function, as extracting loca features, form, and motion is critical to scene analysis, object identification, and navigation (ii they are high-dimensional, and so are appropriate to probe mechanisms likely to underlie representations of other complex perceptual spaces, (iii) algorithms exist for construction of stimuli in all portions of the spaces, including - but not limited to-those that occur in the naturl environment, (iv) they can be investigated at the perceptual level in humans, yielding model predictions that can be tested at the neuronal level in non-human primates, and (v) the spaces are calibrated - that is, ideal observer performance can be calculated a priori, and this serves as a benchmark for the analysis of perceptual data. Aim 1 focuses on simple form elements. To constrain models at the algorithmic level, Aim 1A determines the global geometry of the perceptual space, and Aim 1B determines how the perceptual metric changes across the space. To test the hypothesis that the perceptual space serves as a common workspace for segmentation, discrimination, and visual working memory, Aim 1C will determine whether a single metric accounts for performance on all three kinds of tasks. Aim 2A (shading) and 2B (pattern and shape) seek to extend previous and anticipated findings to more complex aspects of form: do cues combine in a simple quadratic fashion? do efficient coding principles govern resource allocation? and does the perceptual geometry implied by threshold and suprathreshold judgments require a dual representation? To determine whether these principles further generalize to the temporal domain, Aim 3 extends the approach to analysis of local motion signals. Successful completion of this research will advance the understanding of the design principles and computations underlying sensory processing, and will thus support the rational design of advanced therapeutic modalities, such as neural prosthetics for loss of visual function.