Abstract
From LRDE
A
Common Lisp og packages fg provide a functionality similar to og namespaces fg present in C++. They allow to encapsulate symbols that can be either exported or private. The exported ones must be explicitly declared at the package definition. This list is tedious to maintain when developing large projects. In this report will study ways to automatically maintain this list. Several approaches are proposed and compared. +
Our goal is to segment ancient maps of France, being cut and glued on a canvas. The colors in the canvas and in the maps are very close, making the boundary difficult to distinguish. However, pixel-perfect parts of the boundary are obtained thanks to morphological operators. To complete the segmentation, the image is separated, thanks to a maskinto three zones (map, canvas, and a "thick" boundary)which are used by a classification method to obtain results close to the actual boundary. Finally, the first results are corrected thanks to such new knowledge. +
Social media have been extensively used in the attempt to influence users with biased information and rumors. Recently, a new set of accounts know as troll farms have emerged, lauching coordinated disinformation campains. In this regard, we aim to identify the creation of troll farms by studying how they differ from common users and how they attempt to influence them. By leveraging graph analysis techniques we will look into how trolls propagate information through Twitter. Our goal being to build a system capable of recognizing troll farm accounts on trending topics on Twitter. +
VAUCANSON est une plateforme de manipulation d'automates finis et de transducteurs. Après plusieurs années de developpement, il fût constaté que l'interface mise en place pour manipuler les automates était trop complexe. Des travaux furent donc entrepris pour résoudre ce problème, amenant ainsi à l'introduction des "label kinds". Deux versions de la plateforme sont donc en developpement aujourd'hui: VAUCANSON 1.4, qui vise a terminer et compléter le travail effectué avant l'introduction des kinds et VAUCANSON 2.0, dernière version de la plateforme, pour le moment incomplète. Ce rapport a pour but de présenter une nouvelle fonctionnalité de VAUCANSON 1.4, les semi-anneaux Z/nZainsi que le travail en cours sur VAUCANSON 2.0. +
B
Entrainer un réseau de neurones convolutionnel dépends de l'utilisation d'une fonction de coût, qui offre une évaluation de la performance du réseau et permet l'optimisation de celui-ci. Différentes fonctions de coût évaluent la performance de manières différentes, et affectent donc différemment l'entraînement du réseau. Ce rapport a pour but de partager notre progrès dans l'évaluation de la performance de plusieurs fonctions de coût dans l'entrainement de réseaux de neurones convolutionnels pour la segmentation de tumeurs cérébrales. +
Vaucanson is an extensive C++ library for the manipulation of finite state machines. Compared to its main competitorOpenFST, Vaucanson has major performance issues. In order to improve the performance of Vaucanson, a set of tools is required to analyze the library's behavior in terms of CPU time requirements and memory usage. Up to March 2009, no existing tool was fully adapted to Vaucanson and practical to use. CBS is a C++ Benchmarking Suite that measures the performance of C++ projects and provides tools to displayanalyze and compare results in a human-readable form. It is used for in-depth Vaucanson profiling, and it helps the development team rewrite algorithms. +
This report summarizes three methods implemented in the model checking tool Spot. The first method aims at improving the conversion of a program model to a Kripke structure with partial order reduction, a technique where the order of some actions of the program is not considered relevant. This may reduce the size of the Kripke structure. The second method is a modification of the first, which makes it possible to use for LTL properties checking, with any LTL formula. The third method goal is to verify that no livelock may occur in a model, without the need of LTL property. We give benchmarks of all these methods, focusing on testing their common goal: reducing the amount of memory needed, which is an important bottleneck in the model checking field. +
Binary Partition Tree is an efficient structure to store region information for image processing, segmentation and information retrieval. It is a hierarchical structure to simplify operations and recognition on an image. It can use different region models and distance function calculation to create itself. Usually, we construct this tree on a segmented image for efficiency and time saving. All this parameters can variate and change the BPT representation of your image. The resistance of our tree to noise can also be studied, to find out what level of the tree is influenced by the noise. +
Pylene is an image processing (IP) library aiming for high genericity and performance. To accomplish its goals, it was created using C++ templates, allowing for great flexibility and no loss of performance. Pylene is one of the few IP libraries that successfully uses templates, which means it could become a staple in the IP field, but it would first have to overcome a big issue resulting from the usage of C++ templates. That issue is the usability of the library. C++ is not known for its ease of use, and by extension, this makes Pylene hard to use in concrete IP problems. Here, we try to overcome this problem through a Python version of the library. Seeing as though recreating all the library in Python would be nigh impossible, it seems sensible to instead try to bind the high level API to Python. This in itself brings up another issue: the available techniques do not answer our problems well enough, especially in regards to C++ template bindings. Since the idea behind templates of "knowing the type at compile time" doesn't exist in Python, we are faced with a gap between the static world of C++ templates, and the dynamic world of Python. We are presenting in this paper our solution to cross this static-dynamic difference: removing the templates through type-erasure and using contextual information for the type retrieval in Python. +
Spot is a C++ library that relies on the automata theoretic approach to model checking. To represent properties we use LTL-formulae, which are translated into automata. In Spot these automata are Transition-based Generalized Büchi Automata (TGBA). A major issue for a model checker is to be fast. A good way to reach that is to make automata as small as possible. Scientific litterature proposes a lot of algorithm to achieve our goal. Bisimulation and simulation work on '"`UNIQ--math-00000003-QINU`"'-automata. We see in this report how to adapt these algorithms to make them work on TGBA. We will see the profit which is given by the bisimulation implementation. And we deduce the importance to implement the simulation to reduce TGBA. +
L'architecture du projet Vaucanson a été con,cue initialement autour du design pattern Element. Ce dernier a l'énorme avantage de distinguer à la fois les concepts et les implémentations. C'est à dire que pour un type d'automate comme les automates booléens, on peut théoriquement avoir plusieurs implémentations qui se côtoient dans un même programme. Malgré toutes ces précautions, aujourd'hui, ajouter une nouvelle structure s'avère très délicat et remet en cause de nombreux points au sein du projet. C'est pour cette raison que durant ce séminaire nous tenterons de répondre à ces problèmes. Les problèmes de performances qu'a pu rencontrer le projet sont également une bonne motivation pour s'attaquer à ce sujet : il est aujourd'hui indispensable de proposer des nouvelles structures plus efficaces, notamment implémentées avec la bibliothèque Boost. +
Vaucanson est une bibliothèque générique de manipulation d'automates. Le cur de sa généricité réside dans le support de types d'automates variés mais aussi sa capacité à s'appuyer sur différentes structures de données. Actuellementnous avons différentes manières de manipuler des transitions. Cependant, aucune d'entre elles n'est réellement indépendante de la structure de données utilisée. Afin de pallier cela, nous allons nous tourner vers le design pattern Iterator. Nous évaluerons l'impact de ce design pattern sur les performances et sur l'utilisation de la bibliothèque en termes d'écriture d'algorithmes. +
The work performed last year underlined the fact that the overall performance issues of Vaucanson could be widely improved by an internal use of hash tables and, more particularly by the Multi Index from the Boost C++ library. We tried to make good use of the new functionalities provided by Boost. It results in the implementation of a new graph structure. We present in this report the different issues implied by these modifications on the graph implementation and we try to answer to the new issues about the genericity of Vaucanson. +
Suite aux séminaires de l'année dernière, il en ressort que les performances globales de Vaucanson pouvaient largement être améliorées par l'usage de tables de hachage et plus particulièrement les Multi Index de la bibliothèque Boost. Pour ce séminaire, nous chercherons à tirer parti des nouvelles fonctionnalités offertes par Boost. Ceci impliquera l'apparition d'une nouvelle implémentation de graphe. Nous présenterons au cours de ce séminaire les enjeux induits par ces changements sur l'implémentation et tenterons de répondre au problèmatiques soulevées par la généricité de Vaucanson. +
Vaucanson is a generic finite state machine manipulation platform. We have based our genericity on the ability to not only support various types of automata, but also to use different data structures to represent them. In its current state, we have various techniques to iterate over sets of transitions, however, none of them is really independent of the data structures. To overcome this problem, we have integrated the design pattern Iterator. Our goal is to assess the improvements given by this method in terms of performance and code writing. +
L'architecture du projet Vaucanson a été con,cue initialement autour du design pattern Element. Ce dernier a l'énorme avantage de distinguer à la fois les concepts et les implémentations. C'est à dire que pour un type d'automate comme les automates booléens, on peut théoriquement avoir plusieurs implémentations qui se côtoient dans un même programme. Malgré toutes ces précautions, aujourd'hui, ajouter une nouvelle structure s'avère très délicat et remet en cause de nombreux points au sein du projet. C'est pour cette raison que durant ce séminaire nous tenterons de répondre à ces problèmes. Les problèmes de performances qu'a pu rencontrer le projet sont également une bonne motivation pour s'attaquer à ce sujet : il est aujourd'hui indispensable de proposer des nouvelles structures plus efficaces, notamment implémentées avec la bibliothèque Boost. +
Deep neural networks are increasingly used for their capacity to correlate concrete parameters to deduce abstract characteristics. The bottleneck neural network is a specific form of those. This work presents the principle of this kind of network and its use for the reprocessing of Mel Frequency Cepstral Coefficients in a speaker recognition system. Therefore, it is a matter of studying the convergence of such network but also the change in overall system performance. +
The detection of white matter hyperintensities in an efficient way is an important issue in the medical field. Indeed, these hyperintensities are complicated to detect with the naked eye, even for medical personnel. An efficient detection of these hyperintensities would allow a better diagnosis of certain neuro-degenerative diseases but also to avoid certain medical errors. This is the issue we have tried to address in this report. We propose a solution based on a convolutional neural network accompanied by a preprocessing performed on these inputs +
Brain development can be evaluated using brain Magnetic Resonance Imaging (MRI). It is useful in cases of preterm birth to ensure that no brain disease develops during the postnatal period. Such diseases can be visible on T2-weighted MR image as high signal intensity (DEHSI). To assess the presence of white matter hyperintensities, this work implements a new robust, semi-automated frameworkbased on mathematical morphology, specialized on neonate brain segmentation. We will go over the related work, the implementation of the different steps and the difficulties encountered. In the end, the version developped during this internship is not completely finished but it is in good shape for a later finalization. +