abstract: The performance of most of the visual computing tasks depends on the quality of the features extracted from the raw data. Insightful feature representation increases the performance of many learning algorithms by exposing the underlying explanatory factors of the output for the unobserved input. A good representation should also handle anomalies in the data such as missing samples and noisy input caused by the undesired, external factors of variation. It should also reduce the data redundancy. Over the years, many feature extraction processes have been invented to produce good representations of raw images and videos. The feature extraction processes can be categorized into three groups. The first group contains processes that are hand-crafted for a specific task. Hand-engineering features requires the knowledge of domain experts and manual labor. However, the feature extraction process is interpretable and explainable. Next group contains the latent-feature extraction processes. While the original feature lies in a high-dimensional space, the relevant factors for a task often lie on a lower dimensional manifold. The latent-feature extraction employs hidden variables to expose the underlying data properties that cannot be directly measured from the input. Latent features seek a specific structure such as sparsity or low-rank into the derived representation through sophisticated optimization techniques. The last category is that of deep features. These are obtained by passing raw input data with minimal pre-processing through a deep network. Its parameters are computed by iteratively minimizing a task-based loss. In this dissertation, I present four pieces of work where I create and learn suitable data representations. The first task employs hand-crafted features to perform clinically-relevant retrieval of diabetic retinopathy images. The second task uses latent features to perform content-adaptive image enhancement. The third task ranks a pair of images based on their aestheticism. The goal of the last task is to capture localized image artifacts in small datasets with patch-level labels. For both these tasks, I propose novel deep architectures and show significant improvement over the previous state-of-art approaches. A suitable combination of feature representations augmented with an appropriate learning approach can increase performance for most visual computing tasks. / Dissertation/Thesis / Doctoral Dissertation Computer Science 2017

L'acquisition autonome de representations et de fonctionnalites en robotique pose de nombreux problemes theoriques. Aujourd'hui, les systemes robotiques autonomes sont concus autour d'un ensemble de fonctionnalites. Leurs representations du monde sont issues de l'analyse d'un probleme et d'une modelisation prealablement donnees par les concepteurs. Cette approche limite les capacites d'apprentissage. Nous proposons dans cette these un systeme ouvert de representations et de fonctionnalites. Ce systeme apprend en experimentant son environnement et est guide par l'augmentation d'une fonction de valeur. L'objectif du systeme consiste a agir sur son environnement pour reactiver les representations dont il avait appris une connotation positive. Une analyse de la capacite a generaliser la production d'actions appropriees pour ces reactivations conduit a definir un ensemble de proprietes necessaires pour un tel systeme. Le systeme de representation est constitue d'un reseau d'unites de traitement semblables et utilise un codage par position. Le sens de l'etat d'une unite depend de sa position dans le reseau. Ce systeme de representation possede des similitudes avec le principe de numeration par position. Une representation correspond a l'activation d'un ensemble d'unites. Ce systeme a ete implemente dans une suite logicielle appelee NeuSter qui permet de simuler des reseaux de plusieurs millions d'unites et milliard de connexions sur des grappes heterogenes de machines POSIX. Les premiers resultats permettent de valider les contraintes deduites de l'analyse. Un tel systeme permet d'apprendre dans un meme reseau, de facon hierarchique et non supervisee, des detecteurs de bords et de traits, de coins, de terminaisons de traits, de visages, de directions de mouvement, de rotations, d'expansions, et de phonemes. NeuSter apprend en ligne en utilisant uniquement les donnees de ses capteurs. Il a ete teste sur des robots mobiles pour l'apprentissage et le suivi d'objets.