Deep learning for computer vision: a comparison between convolutional neural networks and hierarchical temporal memories on object recognition tasks

In recent years, Deep Learning techniques have shown to perform well on a large variety of problems both in Computer Vision and Natural Language Processing, reaching and often surpassing the state of the art on many tasks. The rise of deep learning is also revolutionizing the entire field of Machine Learning and Pattern Recognition pushing forward the concepts of automatic feature extraction and unsupervised learning in general. However, despite the strong success both in science and business, deep learning has its own limitations. It is often questioned if such techniques are only some kind of brute-force statistical approaches and if they can only work in the context of High Performance Computing with tons of data. Another important question is whether they are really biologically inspired, as claimed in certain cases, and if they can scale well in terms of "intelligence". The dissertation is focused on trying to answer these key questions in the context of Computer Vision and, in particular, Object Recognition, a task that has been heavily revolutionized by recent advances in the field. Practically speaking, these answers are based on an exhaustive comparison between two, very different, deep learning techniques on the aforementioned task: Convolutional Neural Network (CNN) and Hierarchical Temporal memory (HTM). They stand for two different approaches and points of view within the big hat of deep learning and are the best choices to understand and point out strengths and weaknesses of each of them. CNN is considered one of the most classic and powerful supervised methods used today in machine learning and pattern recognition, especially in object recognition. CNNs are well received and accepted by the scientific community and are already deployed in large corporation like Google and Facebook for solving face recognition and image auto-tagging problems. HTM, on the other hand, is known as a new emerging paradigm and a new meanly-unsupervised method, that is more biologically inspired. It tries to gain more insights from the computational neuroscience community in order to incorporate concepts like time, context and attention during the learning process which are typical of the human brain. In the end, the thesis is supposed to prove that in certain cases, with a lower quantity of data, HTM can outperform CNN.

Elaborazione di video tramite un software web per la computer vision

La tesi tratta i temi di computer vision connessi alle problematiche di inserimento in una piattaforma Web. Nel testo sono spiegate alcune soluzioni per includere una libreria software per l'emotion recognition in un'applicazione web e tecnologie per la registrazione di un video, catturando le immagine da una webcam.

Differential forms on Carnot groups

The main goal of this thesis is to understand and link together some of the early works by Michel Rumin and Pierre Julg. The work is centered around the so-called Rumin complex, which is a construction in subRiemannian geometry. A Carnot manifold is a manifold endowed with a horizontal distribution. If further a metric is given, one gets a subRiemannian manifold. Such data arise in different contexts, such as: - formulation of the second principle of thermodynamics; - optimal control; - propagation of singularities for sums of squares of vector fields; - real hypersurfaces in complex manifolds; - ideal boundaries of rank one symmetric spaces; - asymptotic geometry of nilpotent groups; - modelization of human vision. Differential forms on a Carnot manifold have weights, which produces a filtered complex. In view of applications to nilpotent groups, Rumin has defined a substitute for the de Rham complex, adapted to this filtration. The presence of a filtered complex also suggests the use of the formal machinery of spectral sequences in the study of cohomology. The goal was indeed to understand the link between Rumin's operator and the differentials which appear in the various spectral sequences we have worked with: - the weight spectral sequence; - a special spectral sequence introduced by Julg and called by him Forman's spectral sequence; - Forman's spectral sequence (which turns out to be unrelated to the previous one). We will see that in general Rumin's operator depends on choices. However, in some special cases, it does not because it has an alternative interpretation as a differential in a natural spectral sequence. After defining Carnot groups and analysing their main properties, we will introduce the concept of weights of forms which will produce a splitting on the exterior differential operator d. We shall see how the Rumin complex arises from this splitting and proceed to carry out the complete computations in some key examples. From the third chapter onwards we will focus on Julg's paper, describing his new filtration and its relationship with the weight spectral sequence. We will study the connection between the spectral sequences and Rumin's complex in the n-dimensional Heisenberg group and the 7-dimensional quaternionic Heisenberg group and then generalize the result to Carnot groups using the weight filtration. Finally, we shall explain why Julg required the independence of choices in some special Rumin operators, introducing the Szego map and describing its main properties.

Rendre les stéréotypes dans le film "Qu'est-ce qu'on a fait au bon Dieu?": transposition ou traduction ?

Le rôle du traducteur est de permettre le dialogue entre deux « mondes autres », de connaître les éléments qui les différencient et de transmettre le message le plus fidèlement possible. Tout dans un texte peut résister à la traduction (vocabulaire, syntaxe, culture). Le but de la traduction est de recréer l’original, tout en sachant que la traduction a ses limites. Quelles sont les limites de la traduction ? Est-il suffisant de bien maîtriser une langue pour pouvoir la traduire ? Est-ce que tous nos mots ont une équivalence dans les autres langues ? Pouvons-nous faire rire l'autre avec les mêmes traits d’humour ? Une traduction, peut-elle transmettre au lecteur ou au spectateur cible toutes les nuances présentes dans le texte ou le film de départ ? Le film français Qu'est-ce qu'on a fait au bon Dieu ?3 sera l'objet de notre analyse. Tout comme le film de Dany Boon, il a eu beaucoup de succès en France et il a également été doublé en italien. Il s'agit ici aussi d'un film humoristique qui joue sur les stéréotypes que nous avons sur l'autre, d'où le choix de commenter le doublage et d'imaginer une piste de transposition italienne. Ce mémoire se propose de répondre aux questions posées ci-dessus, tout en proposant des commentaires de traduction et des pistes de transposition. Nous allons développer trois parties : tout d'abord, nous allons voir comment notre vision du monde influence notre langue et, vice versa, comment notre langue crée un ordre dans le monde. Ensuite, nous allons nous concentrer sur l'intraduisible verbal et non verbal, sur la traduction et l'adaptation de l'humour. Enfin, la troisième et dernière partie sera consacrée au thème de la transposition et, plus particulièrement, à la piste transposition du film Qu'est-ce qu'on a fait au bon Dieu ?.