Dans le domaine des neurosciences computationnelles, l'hypothèse a été émise que le système visuel, depuis la rétine et jusqu'au cortex visuel primaire au moins, ajuste continuellement un modèle probabiliste avec des variables latentes, à son flux de perceptions. Ni le modèle exact, ni la méthode exacte utilisée pour l'ajustement ne sont connus, mais les algorithmes existants qui permettent l'ajustement de tels modèles ont besoin de faire une estimation conditionnelle des variables latentes. Cela nous peut nous aider à comprendre pourquoi le système visuel pourrait ajuster un tel modèle; si le modèle est approprié, ces estimé conditionnels peuvent aussi former une excellente représentation, qui permettent d'analyser le contenu sémantique des images perçues. Le travail présenté ici utilise la performance en classification d'images (discrimination entre des types d'objets communs) comme base pour comparer des modèles du système visuel, et des algorithmes pour ajuster ces modèles (vus comme des densités de probabilité) à des images. Cette thèse (a) montre que des modèles basés sur les cellules complexes de l'aire visuelle V1 généralisent mieux à partir d'exemples d'entraînement étiquetés que les réseaux de neurones conventionnels, dont les unités cachées sont plus semblables aux cellules simples de V1; (b) présente une nouvelle interprétation des modèles du système visuels basés sur des cellules complexes, comme distributions de probabilités, ainsi que de nouveaux algorithmes pour les ajuster à des données; et (c) montre que ces modèles forment des représentations qui sont meilleures pour la classification d'images, après avoir été entraînés comme des modèles de probabilités. Deux innovations techniques additionnelles, qui ont rendu ce travail possible, sont également décrites : un algorithme de recherche aléatoire pour sélectionner des hyper-paramètres, et un compilateur pour des expressions mathématiques matricielles, qui peut optimiser ces expressions pour processeur central (CPU) et graphique (GPU). / Computational neuroscientists have hypothesized that the visual system from the retina to at least primary visual cortex is continuously fitting a latent variable probability model to its stream of perceptions. It is not known exactly which probability model, nor exactly how the fitting takes place, but known algorithms for fitting such models require conditional estimates of the latent variables. This gives us a strong hint as to why the visual system might be fitting such a model; in the right kind of model those conditional estimates can also serve as excellent features for analyzing the semantic content of images perceived. The work presented here uses image classification performance (accurate discrimination between common classes of objects) as a basis for comparing visual system models, and algorithms for fitting those models as probability densities to images. This dissertation (a) finds that models based on visual area V1's complex cells generalize better from labeled training examples than conventional neural networks whose hidden units are more like V1's simple cells, (b) presents novel interpretations for complex-cell-based visual system models as probability distributions and novel algorithms for fitting them to data, and (c) demonstrates that these models form better features for image classification after they are first trained as probability models. Visual system models based on complex cells achieve some of the best results to date on the CIFAR-10 image classification benchmark, and samples from their probability distributions indicate that they have learnt to capture important aspects of natural images. Two auxiliary technical innovations that made this work possible are also described: a random search algorithm for selecting hyper-parameters, and an optimizing compiler for matrix-valued mathematical expressions which can target both CPU and GPU devices.

No description available.

<div>The visual system is responsible for processing visual input, inferring its environmental causes, and assessing its behavioral significance that eventually relates to visual perception and guides animal behavior. There is emerging evidence that visual perception does not simply mirror the outside world but is heavily influenced by contextual information. Specifically, context might refer to the sensory, cognitive, and/or behavioral cues that help to assess the behavioral relevance of image features. One of the most famous examples of such behavior is visual or optical illusions. These illusions contain sensory cues that induce a subjective percept that is not aligned with the physical nature of the stimulation, which, in turn, suggests that a visual system is not a passive filter of the outside world but rather an active inference machine.</div><div>Such robust behavior of the visual system is achieved through intricate neural computations spanning several brain regions that allow dynamic visual processing. Despite the numerous attempts to gain insight into those computations, it has been challenging to decipher the circuit-level implementation of contextual processing due to technological limitations. These questions are of great importance not only for basic research purposes but also for gaining deeper insight into neurodevelopmental disorders that are characterized by altered sensory experiences. Recent advances in genetic engineering and neurotechnology made the mouse an attractive model to study the visual system and enabled other researchers and us to gain unprecedented cellular and circuit-level insights into neural mechanisms underlying contextual processing.</div><div>We first investigated how familiarity modifies the neural representation of stimuli in the mouse primary visual cortex (V1). Using silicon probe recordings and pupillometry, we probed neural activity in naive mice and after animals were exposed to the same stimulus over the course of several days. We have discovered that familiar stimuli evoke low-frequency oscillations in V1. Importantly, those oscillations were specific to the spatial frequency content of the familiar stimulus. To further validate our findings, we investigated how this novel form of visual learning is represented in serotonin-transporter (SERT) deficient mice. These transgenic animals have been previously found to have various neurophysiological alterations. We found that SERT-deficient animals showed longer oscillatory spiking activity and impaired cortical tuning after visual learning. Taken together, we discovered a novel phenomenon of familiarity-evoked oscillations in V1 and utilized it to reveal altered perceptual learning in SERT-deficient mice.</div><div>16</div><div>Next, we investigated how spatial context influences sensory processing. Visual illusions provide a great opportunity to investigate spatial contextual modulation in early visual areas. Leveraging behavioral training, high-density silicon probe recordings, and optogenetics, we provided evidence for an interplay of feedforward and feedback pathways during illusory processing in V1. We first designed an operant behavioral task to investigate illusory perception in mice. Kanizsa illusory contours paradigm was then adapted from primate studies to mouse V1 to elucidate neural correlates of illusory responses in V1. These experiments provided behavioral and neurophysiological evidence for illusory perception in mice. Using optogenetics, we then showed that suppression of the lateromedial area inhibits illusory responses in mouse V1. Taken together, we demonstrated illusory responses in mice and their dependence on the top-down feedback from higher-order visual areas.</div><div>Finally, we investigated how temporal context modulates neural responses by combining silicon probe recordings and a novel visual oddball paradigm that utilizes spatial frequency filtered stimuli. Our work extended prior oddball studies by investigating how adaptation and novelty processing depends on the tuning properties of neurons and their laminar position. Furthermore, given that reduced adaptation and sensory hypersensitivity are one of the hallmarks of altered sensory experiences in autism, we investigated the effects of temporal context on visual processing in V1 of a mouse model of fragile X syndrome (FX), a leading monogenetic cause of autism. We first showed that adaptation was modulated by tuning properties of neurons in both genotypes, however, it was more confined to neurons preferring the adapted feature in FX mice. Oddball responses, on the other hand, were modulated by the laminar position of the neurons in WT with the strongest novelty responses in superficial layers, however, they were uniformly distributed across the cortical column in FX animals. Lastly, we observed differential processing of omission responses in FX vs. WT mice. Overall, our findings suggest that reduced adaptation and increased oddball processing might contribute to altered perceptual experiences in FX and autism.</div>