Understanding how neurons in the primary visual cortex (V1) of primates respond to visual patterns has been a major focus of research in neuroscience for many decades. Numerous different experimental techniques have been used to provide data about how the spatiotemporal patterns of light projected from the visual environment onto the retina relate to the spatiotemporal patterns of neural activity evoked in the visual cortex, across disparate spatial and temporal scales. However, despite the variety of data sources available (or perhaps because of it), there is still no unified explanation for how the circuitry in the eye, the subcortical visual pathways, and the visual cortex responds to these patterns. This thesis outlines a research project to build computational models of V1 that incorporate observations and constraints from an unprecedented range of experimental data sources, reconciling each data source with the others into a consistent proposal for the underlying circuitry and computational mechanisms. The final mechanistic model is the first one shown to be compatible with measurements of: (1) temporal firing-rate patterns in single neurons over tens of milliseconds obtained using single-unit electrophysiology, (2) spatiotemporal patterns in membrane voltages in cortical tissues spanning several square millimeters over similar time scales, obtained using voltage-sensitive-dye imaging, and (3) spatial patterns in neural activity over several square millimeters of cortex, measured over the course of weeks of early development using optical imaging of intrinsic signals. Reconciling this data was not trivial, in part because single-unit studies suggested short, transient neural responses, while population measurements suggested gradual, sustained responses. The fundamental principles of the resulting models are (a) that the spatial and temporal patterns of neural responses are determined not only by the particular properties of a visual stimulus and the internal response properties of individual neurons, but by the collective dynamics of an entire network of interconnected neurons, (b) that these dynamics account both for the fast time course of neural responses to individual stimuli, and the gradual emergence of structure in this network via activity-dependent Hebbian modifications of synaptic connections over days, and (c) the differences between single-unit and population measurements are primarily due to extensive and wide-ranging forms of diversity in neural responses, which become crucial when trying to estimate population responses out of a series of individual measurements. The final model is the first to include all the types of diversity necessary to show how realistic single-unit responses can add up to the very different population-level evoked responses measured using voltage-sensitive-dye imaging over large cortical areas. Additional contributions from this thesis include (1) a comprehensive solution for doing exploratory yet reproducible computational research, implemented as a set of open-source tools, (2) a general-purpose metric for evaluating the biological realism of model orientation maps, and (3) a demonstration that the previous developmental model that formed the basis of the models in this thesis is the only developmental model so far that produces realistic orientation maps. These analytical results, computational models, and research tools together provide a systematic approach for understanding neural responses to visual stimuli across time scales from milliseconds to weeks and spatial scales from microns to centimeters.

Both visual area V1 and the medial temporal (MT) region of the human brain play a role in motion perception. V1 is thought to process "local motion," such as the movement of a single bird flying across a relatively small part of space, while MT is thought to process "global motion," such as the movement of an entire flock of birds flying across the sky. However, recent studies using fMRI to measure human brain activity have identified signals in V1 that appear to be global motion signals, although it is unclear whether these are related to global motion processing or stem from some other process. In two experiments, a series of stimulus manipulations were conducted to determine the extent to which these signals in V1 really reflect global motion. Although initial results have so far proven inconclusive, they highlight discrepancies between previous results, suggesting that these motion signals in V1 may be more interesting than researchers have assumed.

The capacity of animals to navigate through familiar or novel environments depends crucially on the integration of a disparate set of self motion cues. The study begins with one of the most simple, planar visual motion, and investigates the cortical organisation of motion sensitive areas. It finds evidence of columnar organisation in hMT+ and a large scale map in V1. Chapter 3 extends this by using stimuli designed to emulate visual and auditory forward motion. It finds that participants are able to determine their direction with a precision close to that predicted by Bayesian integration. Predictions were made regarding neural processing through a modified divisive normalisation model, which was also used to fit the behavioural adaptation results. The integration of different modalities requires visual and auditory streams to combine at some stage within the sensory processing hierarchy. Previous research suggests the ventral intraparietal region (VIP) may be the seat of such integration. Chapter 4 tests whether VIP does combine these cues and whether the correlation between VIP and the unimodal regions changes depending on the coherence of unimodal stimuli. The presence of such modulation is predicted by some models, such as the divisive normalisation model. The processing of such egocentric self motion cues leads to the updating of allocentric representations, these are believed to be encoded by head direction cells and place cells. The experiment in chapter 5 uses a virtual reality stimulus during fMRI scanning to give participants the sense of moving and navigating. Their location in the virtual environment was decoded above chance from voxels in the hippocampus. No head direction signal was classified above chance from any of the three cortical regions investigated. We tentatively conclude that head direction is considerably more difficult to classify from the BOLD signal, possibly due to the homogeneous organisation of head direction cells.

La thématique principale de nos travaux est l'étude de l'intégration du mouvement au niveau de la population du cortex visuel primaire du singe éveillé : de l'identification des circuits corticaux impliqués dans le traitement du mouvement,jusqu'à l'identification et l'émergence d'un signal de mouvement. Nous avons ainsi principalement utilisé deux protocoles (mouvement réel ou apparent).La réponse neuronale de population à l'entrée du système (V1) a été comparée à une réponse comportementale en sortie, la réponse de suivi oculaire réflexe (OFR).L'activité de population dans le cortex visuel primaire est enregistrée par imagerie optique de composés sensible au potentiel.Nous avons alors montré que la réponse au contraste dans V1 est contrôlée par un bassin de normalisation dynamique qui évolue lentement via un recrutement progressif et polysynaptique des circuits récurrents locaux. Ce bassin reçoit des afférents horizontaux liés au contraste qui suppriment graduellement le gain au contraste et à la réponse neuronale.Ensuite, en comparant l'activité de population de V1 avec la réponse de suivi oculaire réflexe avec un stimulus dont l'échelle intermédiaire active à la fois l'entrée et la sortie du système, nous avons identifié deux mécanismes distincts, impliqués dans les interactions contextuelles étudiées : un mécanisme précoce et rapide agissant sur les entrées fortes provenant majoritairement de MT et un mécanisme lent et soutenu plus visible sur les entrées faibles provenant majoritairement de V1.Finalement, en étudiant l'intégration et la représentation du mouvement apparent à la surface de V1, nous avons observé que la dynamique de l'activité corticale générée par des stimuli de mouvement apparent induit une suppression non-linéaire à la surface du cortex qui permet à la population de V1 de ne représenter qu'un seul stimulus à la fois, et ferait donc émerger un signal de mouvement non-ambigu.Pour conclure, nos expériences montrent que les interactions non-linéaires entre et parmi les aires corticales entraînent la normalisation, la modulation et l'émergence de différents signaux de mouvement. / Our goal is to study motion integration at population level in V1 in the awake behaving onkey. We compare V1population recorded with optical imaging of voltage sensitive dyes with ocular following response.We have shown that contrast response function in V1 is controlled by a dynamic normalization pool. Then we identified two distinct mechanisms involved in contextual modulations: a fast transient one originating from MT and a show and sustained one, originating from V1. Finally, we have observed that cortical activity dynamics in presponse to apparent motion can induce a suppression wave at acortical surface.

京都大学 / 0048 / 新制・課程博士 / 博士(医学) / 甲第21661号 / 医博第4467号 / 新制||医||1035(附属図書館) / 京都大学大学院医学研究科医学専攻 / (主査)教授 篠原 隆司, 教授 小川 修, 教授 近藤 玄 / 学位規則第4条第1項該当 / Doctor of Medical Science / Kyoto University / DFAM

The key to understanding vision is to acquire insight into the sensory coding of indi- vidual neurons. To this end, major advances were done over the past 50 years in fitting models to neural data to identify the mapping from sensory space to neural responses. Especially the advance of DNNs in neuroscience allowed for model fits with excellent predictive power. However, such advanced neural models are complex, and their poor in- terpretability has so far hindered deeper understanding of the principles of visual coding. To address this issue, a recent study proposed a method which identifies the stimulus that activates the neuron the most. However, the sensory coding of highly non-linear neurons, which are abundant already at the earliest stages of visual processing, is too complex for a single stimulus to sufficiently characterize it. A more robust way to char- acterize this coding is through identifying the input sub-space within which the neuron is activated identically - i.e. finding invariances of the neuron's sensory representation. In this thesis, a novel approach for finding such invariant stimuli is proposed. The proposed technique is based on a generator neural network, which maps Gaussian noise from latent space to a stimulus set which equally activates a given neuron. The method demonstrated the...

No description available.

Mémoire numérisé par la Direction des bibliothèques de l'Université de Montréal.

Despite its great complexity, a great deal is known about the organization and information-processing properties of the visual system. However, the neural correlates of visual awareness are not yet understood. By studying patients with blindsight, the primary visual cortex (V1) has attracted a lot of attention recently. Although this brain area appears to be important for visual awareness, its exact role is still a matter of debate. Interactive models propose a direct role for V1 in generating visual awareness through recurrent processing. Hierarchal models instead propose that awareness is generated in later visual areas and that the role of V1 is limited to transmitting the necessary information to these areas. Interactive and hierarchical models make different predictions and the aim of this thesis is to review the evidence from lesions, perceptual suppression, and transcranial magnetic stimulation (TMS), along with data from internally generated visual awareness in dreams, hallucinations and imagery, this in order to see whether current evidence favor one type of model over the other. A review of the evidence suggests that feedback projections to V1 appear to be important in most cases for visual awareness to arise but it can arise even when V1 is absent.

京都大学 / 0048 / 新制・課程博士 / 博士(医学) / 甲第20665号 / 医博第4275号 / 新制||医||1024(附属図書館) / 京都大学大学院医学研究科医学専攻 / (主査)教授 羽賀 博典, 教授 横出 正之, 教授 瀬原 淳子 / 学位規則第4条第1項該当 / Doctor of Medical Science / Kyoto University / DFAM