Please use this identifier to cite or link to this item: https://hdl.handle.net/20.500.12202/381
Title: Adaptation in V1 as Inferences About Natural Movie Statistics
Authors: Snow, Michoel
Keywords: Neurosciences.
Applied mathematics.
Issue Date: 2016
Publisher: ProQuest Dissertations & Theses
Citation: Source: Dissertation Abstracts International, Volume: 77-11(E), Section: B.;Advisors: Odelia Schwartz.
Abstract: Temporal context has a profound influence on neural responses in the visual system. Stimuli seen in the past, over time scales of milliseconds, seconds, minutes, and even hours and days, can influence the neuronal responses to the present. Here I focus on primary visual cortex as a paradigmatic example. Temporal context in the past influences the responses of neurons in the present, but the strength of this influence is variable. The neuronal response can range from seemingly unaffected by context to nearly silenced, depending upon the stimuli, both in the past and present, as well as the neuronal properties. However, the parameters which govern these temporal contextual effects, known as adaptation, are poorly understood. In this thesis, I posit that neural adaptation reflects optimal processing of visual inputs in the natural environment. The connection between optimal processing of natural images and neuronal responses has been influential in the study of sensory processing, and it has been demonstrated that some aspects of sensory processing reflect optimal processing of the natural visual environment. However, the connection between adaptation and the dynamic natural inputs from the visual environment is poorly understood. The work presented in this dissertation explores the link between optimal processing of natural movies and the parameters which govern adaptation.;In the first part of this work, I discuss how I developed a temporal model of contextual effects of natural movies. This model was based on approaches developed in the spatial context domain for still images, but I generalized its application to temporal movie structure and to account for a wider variety of effects. First, I developed a principled set of constraints in which to learn the model parameters. Second, I learned temporal statistical regularities from an ensemble of natural movies, and linked these statistics to adaptation in the primary visual cortex via divisive normalization, a ubiquitous neural computation. The model divisively normalizes the present visual input by the past visual inputs only to the degree that these are inferred to be statistically dependent. I then formulated a set of complementary models from which to explore stimulus specific and neuronal specific adaptation effects. These amounted in the model to two different metrics of statistical similarity. Finally, to account for effects on multiple timescales I introduced an updating element to the model. Taken together, I created a common framework which is able to learn the statistics of natural movies and then utilize them to instruct its application of a diverse set of adaptation effects.;In the second part of this work, I tested the model's predictions against a series of classical and recent adaptation experimental findings. I was able to predict classical tuning curve suppression and repulsion effects using the stimulus specific model. With the updating neuron specific model, I could predict more recently quantified population equalization effects.;The modeling framework for temporal context I discuss here thus predicts a variety of adaptation effects for simple stimuli. Moreover, due to the nature of the model's foundation, it can make predictions about adaptation to more complex visual stimuli, including natural movies. The framework explored here has the potential to be a foundation for scene statistics derived models with significant predictive power and applicability.
URI: https://ezproxy.yu.edu/login?url=http://gateway.proquest.com/openurl?url_ver=Z39.88-2004&rft_val_fmt=info:ofi/fmt:kev:mtx:dissertation&res_dat=xri:pqm&rft_dat=xri:pqdiss:10140398
https://hdl.handle.net/20.500.12202/381
Appears in Collections:Albert Einstein College of Medicine: Doctoral Dissertations

Files in This Item:
There are no files associated with this item.


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.