Дисертації з теми "Apprentissage de similarités"
Оформте джерело за APA, MLA, Chicago, Harvard та іншими стилями
Ознайомтеся з топ-47 дисертацій для дослідження на тему "Apprentissage de similarités".
Біля кожної праці в переліку літератури доступна кнопка «Додати до бібліографії». Скористайтеся нею – і ми автоматично оформимо бібліографічне посилання на обрану працю в потрібному вам стилі цитування: APA, MLA, «Гарвард», «Чикаго», «Ванкувер» тощо.
Також ви можете завантажити повний текст наукової публікації у форматі «.pdf» та прочитати онлайн анотацію до роботи, якщо відповідні параметри наявні в метаданих.
Переглядайте дисертації для різних дисциплін та оформлюйте правильно вашу бібліографію.
Boyer, Laurent. "Apprentissage probabiliste de similarités d'édition." Phd thesis, Université Jean Monnet - Saint-Etienne, 2011. http://tel.archives-ouvertes.fr/tel-00718835.
Brezellec, Pierre. "Techniques d'apprentissage par explication et détections de similarités." Paris 13, 1992. http://www.theses.fr/1992PA132033.
Philippeau, Jérémy. "Apprentissage de similarités pour l'aide à l'organisation de contenus audiovisuels." Toulouse 3, 2009. http://thesesups.ups-tlse.fr/564/.
In the perspective of new usages in the field of the access to audiovisual archives, we have created a semi-automatic system that helps a user to organize audiovisual contents while performing tasks of classification, characterization, identification and ranking. To do so, we propose to use a new vocabulary, different from the one already available in INA documentary notices, to answer needs which can not be easily defined with words. We have conceived a graphical interface based on graph formalism designed to express an organisational task. The digital similarity is a good tool in respect with the handled elements which are informational objects shown on the computer screen and the automatically extracted audio and video low-level features. We have made the choice to estimate the similarity between those elements with a predictive process through a statistical model. Among the numerous existing models, the statistical prediction based on the univaried regression and on support vectors has been chosen. H)
Grimal, Clément. "Apprentissage de co-similarités pour la classification automatique de données monovues et multivues." Thesis, Grenoble, 2012. http://www.theses.fr/2012GRENM092/document.
Machine learning consists in conceiving computer programs capable of learning from their environment, or from data. Different kind of learning exist, depending on what the program is learning, or in which context it learns, which naturally forms different tasks. Similarity measures play a predominant role in most of these tasks, which is the reason why this thesis focus on their study. More specifically, we are focusing on data clustering, a so called non supervised learning task, in which the goal of the program is to organize a set of objects into several clusters, in such a way that similar objects are grouped together. In many applications, these objects (documents for instance) are described by their links to other types of objects (words for instance), that can be clustered as well. This case is referred to as co-clustering, and in this thesis we study and improve the co-similarity algorithm XSim. We demonstrate that these improvements enable the algorithm to outperform the state of the art methods. Additionally, it is frequent that these objects are linked to more than one other type of objects, the data that describe these multiple relations between these various types of objects are called multiview. Classical methods are generally not able to consider and use all the information contained in these data. For this reason, we present in this thesis a new multiview similarity algorithm called MVSim, that can be considered as a multiview extension of the XSim algorithm. We demonstrate that this method outperforms state of the art multiview methods, as well as classical approaches, thus validating the interest of the multiview aspect. Finally, we also describe how to use the MVSim algorithm to cluster large-scale single-view data, by first splitting it in multiple subsets. We demonstrate that this approach allows to significantly reduce the running time and the memory footprint of the method, while slightly lowering the quality of the obtained clustering compared to a straightforward approach with no splitting
Champesme, Marc. "Apprentissage par détection de similarités utilisant le formalisme des graphes conceptuels." Paris 13, 1993. http://www.theses.fr/1993PA132004.
Grimal, Clement. "Apprentissage de co-similarités pour la classification automatique de données monovues et multivues." Phd thesis, Université de Grenoble, 2012. http://tel.archives-ouvertes.fr/tel-00819840.
Akgül, Ceyhun Burak. "Descripteurs de forme basés sur la densité probabiliste et apprentissage des similarités pour la recherche d'objets 3D." Phd thesis, Télécom ParisTech, 2007. http://pastel.archives-ouvertes.fr/pastel-00003154.
Akgül, Ceyhun Burak. "Descripteurs de forme basés sur la densité de probabilité et apprentissage des similarités pour la recherche d'objets 3D." Paris, ENST, 2007. http://www.theses.fr/2007ENST0026.
Content-based retrieval research aims at developing search engines that would allow users to perform a query by similarity of content. This thesis deals with two fundamentals problems in content-based 3D object retrieval : (1) How to describe a 3D shape to obtain a reliable representative for the subsequent task of similarity search? (2) How to supervise the search process to learn inter-shape similarities for more effective and semantic retrieval? Concerning the first problem, we develop a novel 3D shape description scheme based on probability density of multivariate local surface features. We constructively obtain local characterizations of 3D points and then summarize the resulting local shape information into a global shape descriptor. For probability density estimation, we use the general purpose kernel density estimation methodology, coupled with a fast approximation algorithm: the fast Gauss transform. Experiments that we have conducted on several 3D object databases show that density-based descriptors are very fast to compute and very effective for 3D similarity search. Concerning the second problem, we propose a similarity learning scheme. Our approach relies on combining multiple similarity scores by optimizing a convex regularized version of the empirical ranking risk criterion. This score fusion approach to similarity learning is applicable to a variety of search engine problems. In this work, we demonstrate its effectiveness in 3D object retrieval
Morvant, Emilie. "Apprentissage de vote de majorité pour la classification supervisée et l'adaptation de domaine : approches PAC-Bayésiennes et combinaison de similarités." Phd thesis, Aix-Marseille Université, 2013. http://tel.archives-ouvertes.fr/tel-00879072.
Le, Boudic-Jamin Mathilde. "Similarités et divergences, globales et locales, entre structures protéiques." Thesis, Rennes 1, 2015. http://www.theses.fr/2015REN1S119/document.
This thesis focusses on local and global similarities and divergences inside protein structures. First, structures are scored, with criteria of similarity and distance in order to provide a supervised classification. This structural domain classification inside existing hierarchical databases is possible by using dominances and learning. These methods allow to assign new domains with accuracy and exactly. Second we focusses on local similarities and proposed a method of protein comparison modelisation inside graphs. Graph traversal allows to find protein similar substructures. This method is based on compatibility between elements and criterion of distances. We can use it and detect events such that circular permutations, hinges and structural motif repeats. Finally we propose a new approach of accurate protein structure analysis that focused on divergences between similar structures
Trouvilliez, Benoît. "Similarités de données textuelles pour l'apprentissage de textes courts d'opinions et la recherche de produits." Thesis, Artois, 2013. http://www.theses.fr/2013ARTO0403/document.
This Ph.D. thesis is about the establishment of textual data similarities in the client relation domain. Two subjects are mainly considered : - the automatic analysis of short messages in response of satisfaction surveys ; - the search of products given same criteria expressed in natural language by a human through a conversation with a program. The first subject concerns the statistical informations from the surveys answers. The ideas recognized in the answers are identified, organized according to a taxonomy and quantified. The second subject concerns the transcription of some criteria over products into queries to be interpreted by a database management system. The number of criteria under consideration is wide, from simplest criteria like material or brand, until most complex criteria like color or price. The two subjects meet on the problem of establishing textual data similarities thanks to NLP techniques. The main difficulties come from the fact that the texts to be processed, written in natural language, are short ones and with lots of spell checking errors and negations. Establishment of semantic similarities between words (synonymy, antonymy, ...) and syntactic relations between syntagms (conjunction, opposition, ...) are other issues considered in our work. We also study in this Ph. D. thesis automatic clustering and classification methods in order to analyse answers to satisfaction surveys
Vogel, Robin. "Similarity ranking for biometrics : theory and practice." Thesis, Institut polytechnique de Paris, 2020. http://www.theses.fr/2020IPPAT031.
The rapid growth in population, combined with the increased mobility of people has created a need for sophisticated identity management systems.For this purpose, biometrics refers to the identification of individuals using behavioral or biological characteristics. The most popular approaches, i.e. fingerprint, iris or face recognition, are all based on computer vision methods. The adoption of deep convolutional networks, enabled by general purpose computing on graphics processing units, made the recent advances incomputer vision possible. These advances have led to drastic improvements for conventional biometric methods, which boosted their adoption in practical settings, and stirred up public debate about these technologies. In this respect, biometric systems providers face many challenges when learning those networks.In this thesis, we consider those challenges from the angle of statistical learning theory, which leads us to propose or sketch practical solutions. First, we answer to the proliferation of papers on similarity learningfor deep neural networks that optimize objective functions that are disconnected with the natural ranking aim sought out in biometrics. Precisely, we introduce the notion of similarity ranking, by highlighting the relationship between bipartite ranking and the requirements for similarities that are well suited to biometric identification. We then extend the theory of bipartite ranking to this new problem, by adapting it to the specificities of pairwise learning, particularly those regarding its computational cost. Usual objective functions optimize for predictive performance, but recentwork has underlined the necessity to consider other aspects when training a biometric system, such as dataset bias, prediction robustness or notions of fairness. The thesis tackles all of those three examplesby proposing their careful statistical analysis, as well as practical methods that provide the necessary tools to biometric systems manufacturers to address those issues, without jeopardizing the performance of their algorithms
Labiadh, Mouna. "Méthodologie de construction de modèles adaptatifs pour la simulation énergétique des bâtiments." Thesis, Lyon, 2021. http://www.theses.fr/2021LYSE1158.
Predictive modeling of energy consumption in buildings is essential for intelligent control and efficient planning of energy networks. One way to perform predictive modeling is through machine learning approaches. Alongside their good performance, these approaches are time efficient and facilitates the integration of buildings into smart environments. However, accurate machine learning models rely heavily on collecting relevant building operational data in a sufficient amount, notably when deep learning is used. In the field of buildings energy, historical data are not available for training, such is the case in newly built or newly renovated buildings. Moreover, it is common to verify the energy efficiency of buildings before construction or renovation. For such cases, only a contextual description about the future building and its design is available. The goal of this dissertation is to address the predictive modeling tasks of building energy consumption when no historical data are available for the given target building. To that end, existing data collected from multiple different source buildings are leveraged. This is increasingly relevant with the growth of open data initiatives in various sectors, namely building energy. The main idea is to transfer knowledge across building models. There is little research at the intersection of building energy modeling and knowledge transfer. An important challenge arises when dealing with multi-source data, since large domain shift may exist between different sources and also between each source and the target. As a contribution, a two-fold query-adaptive methodology is developed for cross-building predictive modeling. The first process recommends relevant training data to a target building solely by using a minimal contextual description on it (metadata). Contextual descriptions are provided as user queries. To enable a task-specific recommendation, a deep similarity learning framework is used. The second process trains multiple predictive models based on recommended training data. These models are combined together using an ensemble learning framework to ensure a robust performance. The implementation of the proposed methodology is based on microservices. Logically independent workflows are modeled as microservices with single purposes and separate data sources. Building metadata and time series data collected from multiple sources are integrated into an unified ontology-based view. Experimental evaluation of the predictive model factory validates the effectiveness and the applicability for the use case of building energy modeling. Moreover, because of its generic design, the methodology for query-adaptive cross-domain predictive modeling can be re-used for a diverse range of use cases in different fields
Boutin, Luc. "Biomimétisme, génération de trajectoires pour la robotique humanoïde à partir de mouvements humains." Poitiers, 2009. http://theses.edel.univ-poitiers.fr/theses/2009/Boutin-Luc/2009-Boutin-Luc-These.pdf.
The true reproduction of human locomotion is a topical issue on humanoid robots. The goal of this work is to define a process to imitate the human motion with humanoid robots. In the first part, the motion capture techniques are presented. The measurement protocol adopted is exposed and the calculation of joint angles. An adaptation of three existing algorithms is proposed to detect the contact events during complex movements. The method is valided by measurements on thirty healthy subjects. The second part deals with the generation of humanoid trajectories imitating the human motion. Once the problem and the imitation process are defined, the balance criterion of walking robots is presented. Using data from human motion capture, the reference trajectories of the feet and ZMP are defined. These paths are modified to avoid collision between feet, particularly in the case of executing a slalom. Finally an inverse kinematics algorithm developed for this problem is used to determine the joint angles associated with the robot reference trajectories of the feet and ZMP. Several applications on robots HOAP-3 and HRP-2 are presented. The trajectories are validated according to the robot balance through dynamic simulations of the computed motion, and respecting the limits of actuators
Nicolae, Maria-Irina. "Learning similarities for linear classification : theoretical foundations and algorithms." Thesis, Lyon, 2016. http://www.theses.fr/2016LYSES062/document.
The notion of metric plays a key role in machine learning problems, such as classification, clustering and ranking. Learning metrics from training data in order to make them adapted to the task at hand has attracted a growing interest in the past years. This research field, known as metric learning, usually aims at finding the best parameters for a given metric under some constraints from the data. The learned metric is used in a machine learning algorithm in hopes of improving performance. Most of the metric learning algorithms focus on learning the parameters of Mahalanobis distances for feature vectors. Current state of the art methods scale well for datasets of significant size. On the other hand, the more complex topic of multivariate time series has received only limited attention, despite the omnipresence of this type of data in applications. An important part of the research on time series is based on the dynamic time warping (DTW) computing the optimal alignment between two time series. The current state of metric learning suffers from some significant limitations which we aim to address in this thesis. The most important one is probably the lack of theoretical guarantees for the learned metric and its performance for classification.The theory of (ℰ , ϓ, τ)-good similarity functions has been one of the first results relating the properties of a similarity to its classification performance. A second limitation in metric learning comes from the fact that most methods work with metrics that enforce distance properties, which are computationally expensive and often not justified. In this thesis, we address these limitations through two main contributions. The first one is a novel general framework for jointly learning a similarity function and a linear classifier. This formulation is inspired from the (ℰ , ϓ, τ)-good theory, providing a link between the similarity and the linear classifier. It is also convex for a broad range of similarity functions and regularizers. We derive two equivalent generalization bounds through the frameworks of algorithmic robustness and uniform convergence using the Rademacher complexity, proving the good theoretical properties of our framework. Our second contribution is a method for learning similarity functions based on DTW for multivariate time series classification. The formulation is convex and makes use of the(ℰ , ϓ, τ)-good framework for relating the performance of the metric to that of its associated linear classifier. Using uniform stability arguments, we prove the consistency of the learned similarity leading to the derivation of a generalization bound
Dhouib, Sofiane. "Contributions to unsupervised domain adaptation : Similarity functions, optimal transport and theoretical guarantees." Thesis, Lyon, 2020. http://www.theses.fr/2020LYSEI117.
The surge in the quantity of data produced nowadays made of Machine Learning, a subfield of Artificial Intelligence, a vital tool used to extract valuable patterns from them and allowed it to be integrated into almost every aspect of our everyday activities. Concretely, a machine learning algorithm learns such patterns after being trained on a dataset called the training set, and its performance is assessed on a different set called the testing set. Domain Adaptation is an active research area of machine learning, in which the training and testing sets are not assumed to stem from the same probability distribution, as opposed to Supervised Learning. In this case, the two distributions generating the training and testing data correspond respectively to the source and target domains. Our contributions focus on three theoretical aspects related to domain adaptation for classification tasks. The first one is learning with similarity functions, which deals with classification algorithms based on comparing an instance to other examples in order to decide its class. The second is large-margin classification, which concerns learning classifiers that maximize the separation between classes. The third is Optimal Transport that formalizes the principle of least effort for transporting probability masses between two distributions. At the beginning of the thesis, we were interested in learning with so-called (epsilon,gamma,tau)-good similarity functions in the domain adaptation framework, since these functions have been introduced in the literature in the classical framework of supervised learning. This is the subject of our first contribution in which we theoretically study the performance of a similarity function on a target distribution, given it is suitable for the source one. Then, we tackle the more general topic of large-margin classification in domain adaptation, with weaker assumptions than those adopted in the first contribution. In this context, we proposed a new theoretical study and a domain adaptation algorithm, which is our second contribution. We derive novel bounds taking the classification margin on the target domain into account, that we convexify by leveraging the appealing Optimal Transport theory, in order to derive a domain adaptation algorithm with an adversarial variation of the classic Kantorovich problem. Finally, after noticing that our adversarial formulation can be generalized to include several other cases of interest, we dedicate our last contribution to adversarial or minimax variations of the optimal transport problem, where we demonstrate the versatility of our approach
Aseervatham, Sujeevan. "Apprentissage à base de Noyaux Sémantiques pour le Traitement de Données Textuelles." Phd thesis, Université Paris-Nord - Paris XIII, 2007. http://tel.archives-ouvertes.fr/tel-00274627.
Dans le cadre de cette thèse, nous nous intéressons principalement à deux axes.
Le premier axe porte sur l'étude des problématiques liées au traitement de données textuelles structurées par des approches à base de noyaux. Nous présentons, dans ce contexte, un noyau sémantique pour les documents structurés en sections notamment sous le format XML. Le noyau tire ses informations sémantiques à partir d'une source de connaissances externe, à savoir un thésaurus. Notre noyau a été testé sur un corpus de documents médicaux avec le thésaurus médical UMLS. Il a été classé, lors d'un challenge international de catégorisation de documents médicaux, parmi les 10 méthodes les plus performantes sur 44.
Le second axe porte sur l'étude des concepts latents extraits par des méthodes statistiques telles que l'analyse sémantique latente (LSA). Nous présentons, dans une première partie, des noyaux exploitant des concepts linguistiques provenant d'une source externe et des concepts statistiques issus de la LSA. Nous montrons qu'un noyau intégrant les deux types de concepts permet d'améliorer les performances. Puis, dans un deuxième temps, nous présentons un noyau utilisant des LSA locaux afin d'extraire des concepts latents permettant d'obtenir une représentation plus fine des documents.
Gresse, Adrien. "L'Art de la Voix : Caractériser l'information vocale dans un choix artistique." Thesis, Avignon, 2020. http://www.theses.fr/2020AVIG0236.
To reach an international audience, audiovisual productions (films, TVshows, video games) must be translated into other languages. Generally, theoriginal voice is replaced by a new voice in the target language. This processis referred as dubbing. The voice casting process aimed at choosing avoice (an actor) in accordance with the original voice and the character, isperformed manually by an artistic director (AD). Today, ADs are looking fornew "talents" (less expensive and more available than experienced dubbers),but they cannot perform large-scale auditions. Automatic tools capable ofmeasuring the adequacy between a voice in a source language with a voicein a target language/culture and a given context is of great interest for audiovisualcompanies. In addition, beyond voice casting, this voice selectionproblematic echoes the major scientific questions of voice similarity andperception mechanism.In this work, we use the voices of professional actors selected by ADs indifferent languages from already dubbed works. First, we set up a protocolwith state-of-the-art methods in automatic speaker recognition to highlightthe existence of character/role specific information in our data. Wealso identify the influence of linguistic bias on the performance of the system.Then, we build methodological framework to evaluate the ability ofan automatic system to discriminate pairs of voices playing the same character.The system we created is based on Siamese Neural Networks. In thisevaluation protocol, we apply strong constraints to avoid possible biases(linguistic content, gender, etc.) and we learn a similarity measure that reflectsthe AD’s choices with a significant difference that is not attributed tochance. Finally, we train a new representational space highlighting the characterspecific information, called p-vector. Thanks to our methodologicalframework, we show that this representation allows to better discriminatethe voices of new characters, in comparison to a representation oriented onthe speaker information. In addition, we show that it is possible to benefitfrom the generalized knowledge of a model learned on a similar dataset using knowledge distillation in neural networks.This thesis gives a initial answer for assisted voice casting and providesautomatic tools capable of preselecting the relevant voices from a large setof voices in a target language. Despite the fact that the information characteristicof an artistic choice can be extracted from a large volume of data,even if this choice is difficult to formalize, we still have to highlight the explanatoryfactors of the decision of the system.We would like to explain, inaddition to the selection of voices, the reasons of this choice. Furthermore,understanding the decision process of the system would help us define the"voice palette". In future work, we would like to explore the influence of thetarget language and culture by extending our work to more languages. Inthe longer term, this work could help to understand how voice perceptionhas evolved since the beginning of dubbing
Qamar, Ali Mustafa. "Mesures de similarité et cosinus généralisé : une approche d'apprentissage supervisé fondée sur les k plus proches voisins." Phd thesis, Université de Grenoble, 2010. http://tel.archives-ouvertes.fr/tel-00591988.
Michel, Fabrice. "Multi-Modal Similarity Learning for 3D Deformable Registration of Medical Images." Phd thesis, Ecole Centrale Paris, 2013. http://tel.archives-ouvertes.fr/tel-01005141.
Ngo, Duy Hoa. "Enhancing Ontology Matching by Using Machine Learning, Graph Matching and Information Retrieval Techniques." Thesis, Montpellier 2, 2012. http://www.theses.fr/2012MON20096/document.
In recent years, ontologies have attracted a lot of attention in the Computer Science community, especially in the Semantic Web field. They serve as explicit conceptual knowledge models and provide the semantic vocabularies that make domain knowledge available for exchange and interpretation among information systems. However, due to the decentralized nature of the semantic web, ontologies are highlyheterogeneous. This heterogeneity mainly causes the problem of variation in meaning or ambiguity in entity interpretation and, consequently, it prevents domain knowledge sharing. Therefore, ontology matching, which discovers correspondences between semantically related entities of ontologies, becomes a crucial task in semantic web applications.Several challenges to the field of ontology matching have been outlined in recent research. Among them, selection of the appropriate similarity measures as well as configuration tuning of their combination are known as fundamental issues that the community should deal with. In addition, verifying the semantic coherent of the discovered alignment is also known as a crucial task. Furthermore, the difficulty of the problem grows with the size of the ontologies. To deal with these challenges, in this thesis, we propose a novel matching approach, which combines different techniques coming from the fields of machine learning, graph matching and information retrieval in order to enhance the ontology matching quality. Indeed, we make use of information retrieval techniques to design new effective similarity measures for comparing labels and context profiles of entities at element level. We also apply a graph matching method named similarity propagation at structure level that effectively discovers mappings by exploring structural information of entities in the input ontologies. In terms of combination similarity measures at element level, we transform the ontology matching task into a classification task in machine learning. Besides, we propose a dynamic weighted sum method to automatically combine the matching results obtained from the element and structure level matchers. In order to remove inconsistent mappings, we design a new fast semantic filtering method. Finally, to deal with large scale ontology matching task, we propose two candidate selection methods to reduce computational space.All these contributions have been implemented in a prototype named YAM++. To evaluate our approach, we adopt various tracks namely Benchmark, Conference, Multifarm, Anatomy, Library and Large BiomedicalOntologies from the OAEI campaign. The experimental results show that the proposed matching methods work effectively. Moreover, in comparison to other participants in OAEI campaigns, YAM++ showed to be highly competitive and gained a high ranking position
Alliod, Charlotte. "Conception et modélisation de nouvelles molécules hautement énergétiques en fonction des contraintes réglementaires et environnementales." Thesis, Lyon, 2018. http://www.theses.fr/2018LYSE1035.
For the last two decades, the military research has focused on the improvement of explosive performances, while taking into account their environmental and toxicological impacts. These issues are governed by strict regulations: REACh (Registration, Evaluation, Authorization and Restriction of Chemicals) to ensure a high level of health and environmental protection.Today, it's a major consideration to develop High Energetic Materials (HEM) or molecules who's hazard on human health and environment are reduced. Thus, in collaboration with Airbus Safran Lauchers (ASL), a research program was set up to obtain optimized tools for predicting the potential toxicity of HEM and to design new non-toxic and regulatory molecules.Different in silico methods have been used, including Quantitative Structure Activity Activity Relationships (QSARs) and Machine Learning.The search for structural similarity among molecules is an innovative tool on which we based our predictions in silico. This similarity is obtained thanks to an intelligent algorithm developed within the Pole Rhone Alpin de Bio-Informatique of Lyon which gave rise to a patent. This algorithm allows us to obtain more accurate predictions based on experimental data from European directives
Cuan, Bonan. "Deep similarity metric learning for multiple object tracking." Thesis, Lyon, 2019. http://www.theses.fr/2019LYSEI065.
Multiple object tracking, i.e. simultaneously tracking multiple objects in the scene, is an important but challenging visual task. Objects should be accurately detected and distinguished from each other to avoid erroneous trajectories. Since remarkable progress has been made in object detection field, “tracking-by-detection” approaches are widely adopted in multiple object tracking research. Objects are detected in advance and tracking reduces to an association problem: linking detections of the same object through frames into trajectories. Most tracking algorithms employ both motion and appearance models for data association. For multiple object tracking problems where exist many objects of the same category, a fine-grained discriminant appearance model is paramount and indispensable. Therefore, we propose an appearance-based re-identification model using deep similarity metric learning to deal with multiple object tracking in mono-camera videos. Two main contributions are reported in this dissertation: First, a deep Siamese network is employed to learn an end-to-end mapping from input images to a discriminant embedding space. Different metric learning configurations using various metrics, loss functions, deep network structures, etc., are investigated, in order to determine the best re-identification model for tracking. In addition, with an intuitive and simple classification design, the proposed model achieves satisfactory re-identification results, which are comparable to state-of-the-art approaches using triplet losses. Our approach is easy and fast to train and the learned embedding can be readily transferred onto the domain of tracking tasks. Second, we integrate our proposed re-identification model in multiple object tracking as appearance guidance for detection association. For each object to be tracked in a video, we establish an identity-related appearance model based on the learned embedding for re-identification. Similarities among detected object instances are exploited for identity classification. The collaboration and interference between appearance and motion models are also investigated. An online appearance-motion model coupling is proposed to further improve the tracking performance. Experiments on Multiple Object Tracking Challenge benchmark prove the effectiveness of our modifications, with a state-of-the-art tracking accuracy
Cerda, Reyes Patricio. "Apprentissage statistique à partir de variables catégorielles non-uniformisées Similarity encoding for learning with dirty categorical variables Encoding high-cardinality string categorical variables." Thesis, Université Paris-Saclay (ComUE), 2019. http://www.theses.fr/2019SACLS470.
Tabular data often contain columns with categorical variables, usually considered as non-numerical entries with a fixed and limited number of unique elements or categories. As many statistical learning algorithms require numerical representations of features, an encoding step is necessary to transform categorical entries into feature vectors, using for instance one-hot encoding. This and other similar strategies work well, in terms of prediction performance and interpretability, in standard statistical analysis when the number of categories is small. However, non-curated data give rise to string categorical variables with a very high cardinality and redundancy: the string entries share semantic and/or morphological information, and several entries can reflect the same entity. Without any data cleaning or feature engineering step, common encoding methods break down, as they tend to lose information in their vectorial representation. Also, they can create high-dimensional feature vectors, which prevent their usage in large scale settings. In this work, we study a series of categorical encodings that remove the need for preprocessing steps on high-cardinality string categorical variables. An ideal encoder should be: scalable to many categories; interpretable to end users; and capture the morphological information contained in the string entries. Experiments on real and simulated data show that the methods we propose improve supervised learning, are adapted to large-scale settings, and, in some cases, create feature vectors that are easily interpretable. Hence, they can be applied in Automated Machine Learning (AutoML) pipelines in the original string entries without any human intervention
Zhou, Zhyiong. "Recherche d'images par le contenu application à la proposition de mots clés." Thesis, Poitiers, 2018. http://www.theses.fr/2018POIT2254.
The search for information in masses of multimedia data and the indexing of these large databases by the content are very current problems. They are part of a type of data management called Digital Asset Management (or DAM) ; The DAM uses image segmentation and data classification techniques.Our main contributions in this thesis can be summarized in three points : - Analysis of the possible uses of different methods of extraction of local characteristics using the VLAD technique.- Proposed a new method for extracting dominant color information in an image.- Comparison of Support Vector Machines (SVM) to different classifiers for the proposed indexing keywords. These contributions have been tested and validated on summary data and on actual data. Our methods were then widely used in the DAM ePhoto system developed by the company EINDEN, which financed the CIFRE thesis in which this work was carried out. The results are encouraging and open new perspectives for research
Berrahou, Soumia Lilia. "Extraction d'arguments de relations n-aires dans les textes guidée par une RTO de domaine." Thesis, Montpellier, 2015. http://www.theses.fr/2015MONTS019/document.
Today, a huge amount of data is made available to the research community through several web-based libraries. Enhancing data collected from scientific documents is a major challenge in order to analyze and reuse efficiently domain knowledge. To be enhanced, data need to be extracted from documents and structured in a common representation using a controlled vocabulary as in ontologies. Our research deals with knowledge engineering issues of experimental data, extracted from scientific articles, in order to reuse them in decision support systems. Experimental data can be represented by n-ary relations which link a studied object (e.g. food packaging, transformation process) with its features (e.g. oxygen permeability in packaging, biomass grinding) and capitalized in an Ontological and Terminological Ressource (OTR). An OTR associates an ontology with a terminological and/or a linguistic part in order to establish a clear distinction between the term and the notion it denotes (the concept). Our work focuses on n-ary relation extraction from scientific documents in order to populate a domain OTR with new instances. Our contributions are based on Natural Language Processing (NLP) together with data mining approaches guided by the domain OTR. More precisely, firstly, we propose to focus on unit of measure extraction which are known to be difficult to identify because of their typographic variations. We propose to rely on automatic classification of texts, using supervised learning methods, to reduce the search space of variants of units, and then, we propose a new similarity measure that identifies them, taking into account their syntactic properties. Secondly, we propose to adapt and combine data mining methods (sequential patterns and rules mining) and syntactic analysis in order to overcome the challenging process of identifying and extracting n-ary relation instances drowned in unstructured texts
Zheng, Lilei. "Triangular similarity metric learning : A siamese architecture approach." Thesis, Lyon, 2016. http://www.theses.fr/2016LYSEI045/document.
In many machine learning and pattern recognition tasks, there is always a need for appropriate metric functions to measure pairwise distance or similarity between data, where a metric function is a function that defines a distance or similarity between each pair of elements of a set. In this thesis, we propose Triangular Similarity Metric Learning (TSML) for automatically specifying a metric from data. A TSML system is loaded in a siamese architecture which consists of two identical sub-systems sharing the same set of parameters. Each sub-system processes a single data sample and thus the whole system receives a pair of data as the input. The TSML system includes a cost function parameterizing the pairwise relationship between data and a mapping function allowing the system to learn high-level features from the training data. In terms of the cost function, we first propose the Triangular Similarity, a novel similarity metric which is equivalent to the well-known Cosine Similarity in measuring a data pair. Based on a simplified version of the Triangular Similarity, we further develop the triangular loss function in order to perform metric learning, i.e. to increase the similarity between two vectors in the same class and to decrease the similarity between two vectors of different classes. Compared with other distance or similarity metrics, the triangular loss and its gradient naturally offer us an intuitive and interesting geometrical interpretation of the metric learning objective. In terms of the mapping function, we introduce three different options: a linear mapping realized by a simple transformation matrix, a nonlinear mapping realized by Multi-layer Perceptrons (MLP) and a deep nonlinear mapping realized by Convolutional Neural Networks (CNN). With these mapping functions, we present three different TSML systems for various applications, namely, pairwise verification, object identification, dimensionality reduction and data visualization. For each application, we carry out extensive experiments on popular benchmarks and datasets to demonstrate the effectiveness of the proposed systems
Benhabiles, Halim. "3D-mesh segmentation : automatic evaluation and a new learning-based method." Phd thesis, Université des Sciences et Technologie de Lille - Lille I, 2011. http://tel.archives-ouvertes.fr/tel-00834344.
Kessler, Rémy. "Traitement automatique d’informations appliqué aux ressources humaines." Thesis, Avignon, 2009. http://www.theses.fr/2009AVIG0167/document.
Since the 90s, Internet is at the heart of the labor market. First mobilized on specific expertise, its use spreads as increase the number of Internet users in the population. Seeking employment through "electronic employment bursary" has become a banality and e-recruitment something current. This information explosion poses various problems in their treatment with the large amount of information difficult to manage quickly and effectively for companies. We present in this PhD thesis, the work we have developed under the E-Gen project, which aims to create tools to automate the flow of information during a recruitment process.We interested first to the problems posed by the routing of emails. The ability of a companie to manage efficiently and at lower cost this information flows becomes today a major issue for customer satisfaction. We propose the application of learning methods to perform automatic classification of emails to their routing, combining technical and probabilistic vector machines support. After, we present work that was conducted as part of the analysis and integration of a job ads via Internet. We present a solution capable of integrating a job ad from an automatic or assisted in order to broadcast it quickly. Based on a combination of classifiers systems driven by a Markov automate, the system gets very good results. Thereafter, we present several strategies based on vectorial and probabilistic models to solve the problem of profiling candidates according to a specific job offer to assist recruiters. We have evaluated a range of measures of similarity to rank candidatures by using ROC curves. Relevance feedback approach allows to surpass our previous results on this task, difficult, diverse and higly subjective
Michaud, Dorian. "Indexation bio-inspirée pour la recherche d'images par similarité." Thesis, Poitiers, 2018. http://www.theses.fr/2018POIT2288/document.
Image Retrieval is still a very active field of image processing as the number of available image datasets continuously increases.One of the principal objectives of Content-Based Image Retrieval (CBIR) is to return the most similar images to a given query with respect to their visual content.Our work fits in a very specific application context: indexing small expert image datasets, with no prior knowledge on the images. Because of the image complexity, one of our contributions is the choice of effective descriptors from literature placed in direct competition.Two strategies are used to combine features: a psycho-visual one and a statistical one.In this context, we propose an unsupervised and adaptive framework based on the well-known bags of visual words and phrases models that select relevant visual descriptors for each keypoint to construct a more discriminative image representation.Experiments show the interest of using this this type of methodologies during a time when convolutional neural networks are ubiquitous.We also propose a study about semi interactive retrieval to improve the accuracy of CBIR systems by using the knowledge of the expert users
Elgui, Kevin. "Contributions to RSSI-based geolocation." Thesis, Institut polytechnique de Paris, 2020. http://www.theses.fr/2020IPPAT047.
The Network-Based Geolocation has raised a great deal of attention in the context of the Internet of Things. In many situations, connected objects with low-consumption should be geolocated without the use of GPS or GSM. Geolocation techniques based on the Received Signal Strength Indicator (RSSI) stands out, because other location techniques may fail in the context of urban environments and/or narrow band signals. First, we propose some methods for the RSSI-based geolocation problem. The observation is a vector of RSSI received at the various base stations. In particular, we introduce a semi-parametric Nadaraya-Watson estimator of the likelihood, followed by a maximum a posteriori estimator of the object’s position. Experiments demonstrate the interest of the proposed method, both in terms of location estimation performance, and ability to build radio maps. An alternative approach is given by a k-nearest neighbors regressor which uses a suitable metric between RSSI vectors. Results also show that the quality of the prediction is highly related to the chosen metric. Therefore, we turn our attention to the metric learning problem. We introduce an original task-driven objective for learning a similarity between pairs of data points. The similarity is chosen as a sum of regression trees and is sequentially learned by means of a modified version of the so-called eXtreme Gradient Boosting algorithm (XGBoost). The last part of the thesis is devoted to the introduction of a Conditional Independence (CI) hypothesis test. The motivation is related to the fact that for many estimators, the components of the RSSI vectors are assumed independent given the position. The contribution is however provided in a general statistical framework. We introduce the weighted partial copula function for testing conditional independence. The proposed test procedure results from the following ingredients: (i) the test statistic is an explicit Cramér-von Mises transformation of the weighted partial copula, (ii) the regions of rejection are computed using a boot-strap procedure which mimics conditional independence by generating samples. Under the null hypothesis, the weak convergence of the weighted partial copula process is established and endorses the soundness of our approach
Kessler, Rémy. "Traitement automatique d'informations appliqué aux ressources humaines." Phd thesis, Université d'Avignon, 2009. http://tel.archives-ouvertes.fr/tel-00453642.
Schutz, Georges. "Adaptations et applications de modèles mixtes de réseaux de neurones à un processus industriel." Phd thesis, Université Henri Poincaré - Nancy I, 2006. http://tel.archives-ouvertes.fr/tel-00115770.
artificiels pour améliorer le contrôle de processus industriels
complexes, caractérisés en particulier par leur aspect temporel.
Les motivations principales pour traiter des séries temporelles
sont la réduction du volume de données, l'indexation pour la
recherche de similarités, la localisation de séquences,
l'extraction de connaissances (data mining) ou encore la
prédiction.
Le processus industriel choisi est un four à arc
électrique pour la production d'acier liquide au Luxembourg. Notre
approche est un concept de contrôle prédictif et se base sur des
méthodes d'apprentissage non-supervisé dans le but d'une
extraction de connaissances.
Notre méthode de codage se base sur
des formes primitives qui composent les signaux. Ces formes,
composant un alphabet de codage, sont extraites par une méthode
non-supervisée, les cartes auto-organisatrices de Kohonen (SOM).
Une méthode de validation des alphabets de codage accompagne
l'approche.
Un sujet important abordé durant ces recherches est
la similarité de séries temporelles. La méthode proposée est
non-supervisée et intègre la capacité de traiter des séquences de
tailles variées.
Hoffmann, Brice. "Développement d'approches de chémogénomique pour la prédiction des interactions protéine - ligand." Phd thesis, École Nationale Supérieure des Mines de Paris, 2011. http://pastel.archives-ouvertes.fr/pastel-00679718.
André, Barbara. "Atlas intelligent pour guider le diagnostic en endomicroscopie : une application clinique de la reconnaissance d'images par le contenu." Phd thesis, École Nationale Supérieure des Mines de Paris, 2011. http://pastel.archives-ouvertes.fr/pastel-00640899.
Makki, Sara. "An Efficient Classification Model for Analyzing Skewed Data to Detect Frauds in the Financial Sector." Thesis, Lyon, 2019. http://www.theses.fr/2019LYSE1339/document.
There are different types of risks in financial domain such as, terrorist financing, money laundering, credit card fraudulence and insurance fraudulence that may result in catastrophic consequences for entities such as banks or insurance companies. These financial risks are usually detected using classification algorithms. In classification problems, the skewed distribution of classes also known as class imbalance, is a very common challenge in financial fraud detection, where special data mining approaches are used along with the traditional classification algorithms to tackle this issue. Imbalance class problem occurs when one of the classes have more instances than another class. This problem is more vulnerable when we consider big data context. The datasets that are used to build and train the models contain an extremely small portion of minority group also known as positives in comparison to the majority class known as negatives. In most of the cases, it’s more delicate and crucial to correctly classify the minority group rather than the other group, like fraud detection, disease diagnosis, etc. In these examples, the fraud and the disease are the minority groups and it’s more delicate to detect a fraud record because of its dangerous consequences, than a normal one. These class data proportions make it very difficult to the machine learning classifier to learn the characteristics and patterns of the minority group. These classifiers will be biased towards the majority group because of their many examples in the dataset and will learn to classify them much faster than the other group. After conducting a thorough study to investigate the challenges faced in the class imbalance cases, we found that we still can’t reach an acceptable sensitivity (i.e. good classification of minority group) without a significant decrease of accuracy. This leads to another challenge which is the choice of performance measures used to evaluate models. In these cases, this choice is not straightforward, the accuracy or sensitivity alone are misleading. We use other measures like precision-recall curve or F1 - score to evaluate this trade-off between accuracy and sensitivity. Our objective is to build an imbalanced classification model that considers the extreme class imbalance and the false alarms, in a big data framework. We developed two approaches: A Cost-Sensitive Cosine Similarity K-Nearest Neighbor (CoSKNN) as a single classifier, and a K-modes Imbalance Classification Hybrid Approach (K-MICHA) as an ensemble learning methodology. In CoSKNN, our aim was to tackle the imbalance problem by using cosine similarity as a distance metric and by introducing a cost sensitive score for the classification using the KNN algorithm. We conducted a comparative validation experiment where we prove the effectiveness of CoSKNN in terms of accuracy and fraud detection. On the other hand, the aim of K-MICHA is to cluster similar data points in terms of the classifiers outputs. Then, calculating the fraud probabilities in the obtained clusters in order to use them for detecting frauds of new transactions. This approach can be used to the detection of any type of financial fraud, where labelled data are available. At the end, we applied K-MICHA to a credit card, mobile payment and auto insurance fraud data sets. In all three case studies, we compare K-MICHA with stacking using voting, weighted voting, logistic regression and CART. We also compared with Adaboost and random forest. We prove the efficiency of K-MICHA based on these experiments
Nader, Rafic. "A study concerning the positive semi-definite property for similarity matrices and for doubly stochastic matrices with some applications." Thesis, Normandie, 2019. http://www.theses.fr/2019NORMC210.
Matrix theory has shown its importance by its wide range of applications in different fields such as statistics, machine learning, economics and signal processing. This thesis concerns three main axis related to two fundamental objects of study in matrix theory and that arise naturally in many applications, that are positive semi-definite matrices and doubly stochastic matrices.One concept which stems naturally from machine learning area and is related to the positive semi-definite property, is the one of similarity matrices. In fact, similarity matrices that are positive semi-definite are of particular importance because of their ability to define metric distances. This thesis will explore the latter desirable structure for a list of similarity matrices found in the literature. Moreover, we present new results concerning the strictly positive definite and the three positive semi-definite properties of particular similarity matrices. A detailed discussion of the many applications of all these properties in various fields is also established.On the other hand, an interesting research field in matrix analysis involves the study of roots of stochastic matrices which is important in Markov chain models in finance and healthcare. We extend the analysis of this problem to positive semi-definite doubly stochastic matrices.Our contributions include some geometrical properties of the set of all positive semi-definite doubly stochastic matrices of order n with nonnegative pth roots for a given integer p. We also present methods for finding classes of positive semi-definite doubly stochastic matrices that have doubly stochastic pth roots for all p, by making use of the theory of M-Matrices and the symmetric doubly stochastic inverse eigenvalue problem (SDIEP), which is also of independent interest.In the context of the SDIEP, which is the problem of characterising those lists of real numbers which are realisable as the spectrum of some symmetric doubly stochastic matrix, we present some new results along this line. In particular, we propose to use a recursive method on constructing doubly stochastic matrices from smaller size matrices with known spectra to obtain new independent sufficient conditions for SDIEP. Finally, we focus our attention on the realizability by a symmetric doubly stochastic matrix of normalised Suleimanova spectra which is a normalized variant of the spectra introduced by Suleimanova. In particular, we prove that such spectra is not always realizable for odd orders and we construct three families of sufficient conditions that make a refinement for previously known sufficient conditions for SDIEP in the particular case of normalized Suleimanova spectra
Limnios, Stratis. "Graph Degeneracy Studies for Advanced Learning Methods on Graphs and Theoretical Results Edge degeneracy: Algorithmic and structural results Degeneracy Hierarchy Generator and Efficient Connectivity Degeneracy Algorithm A Degeneracy Framework for Graph Similarity Hcore-Init: Neural Network Initialization based on Graph Degeneracy." Thesis, Institut polytechnique de Paris, 2020. http://www.theses.fr/2020IPPAX038.
Extracting Meaningful substructures from graphs has always been a key part in graph studies. In machine learning frameworks, supervised or unsupervised, as well as in theoretical graph analysis, finding dense subgraphs and specific decompositions is primordial in many social and biological applications among many others.In this thesis we aim at studying graph degeneracy, starting from a theoretical point of view, and building upon our results to find the most suited decompositions for the tasks at hand.Hence the first part of the thesis we work on structural results in graphs with bounded edge admissibility, proving that such graphs can be reconstructed by aggregating graphs with almost-bounded-edge-degree. We also provide computational complexity guarantees for the different degeneracy decompositions, i.e. if they are NP-complete or polynomial, depending on the length of the paths on which the given degeneracy is defined.In the second part we unify the degeneracy and admissibility frameworks based on degree and connectivity. Within those frameworks we pick the most expressive, on the one hand, and computationally efficient on the other hand, namely the 1-edge-connectivity degeneracy, to experiment on standard degeneracy tasks, such as finding influential spreaders.Following the previous results that proved to perform poorly we go back to using the k-core but plugging it in a supervised framework, i.e. graph kernels. Thus providing a general framework named core-kernel, we use the k-core decomposition as a preprocessing step for the kernel and apply the latter on every subgraph obtained by the decomposition for comparison. We are able to achieve state-of-the-art performance on graph classification for a small computational cost trade-off.Finally we design a novel degree degeneracy framework for hypergraphs and simultaneously on bipartite graphs as they are hypergraphs incidence graph. This decomposition is then applied directly to pretrained neural network architectures as they induce bipartite graphs and use the coreness of the neurons to re-initialize the neural network weights. This framework not only outperforms state-of-the-art initialization techniques but is also applicable to any pair of layers convolutional and linear thus being applicable however needed to any type of architecture
El, Mahrsi Mohamed Khalil. "Analyse et fouille de données de trajectoires d'objets mobiles." Phd thesis, Telecom ParisTech, 2013. http://tel.archives-ouvertes.fr/tel-00943936.
Ngo, Duy Hoa. "Amélioration de l'alignement d'ontologies par les techniques d'apprentissage automatique, d'appariement de graphes et de recherche d'information." Phd thesis, Université Montpellier II - Sciences et Techniques du Languedoc, 2012. http://tel.archives-ouvertes.fr/tel-00767318.
Zhang, Yuyao. "Non-linear dimensionality reduction and sparse representation models for facial analysis." Thesis, Lyon, INSA, 2014. http://www.theses.fr/2014ISAL0019/document.
Face analysis techniques commonly require a proper representation of images by means of dimensionality reduction leading to embedded manifolds, which aims at capturing relevant characteristics of the signals. In this thesis, we first provide a comprehensive survey on the state of the art of embedded manifold models. Then, we introduce a novel non-linear embedding method, the Kernel Similarity Principal Component Analysis (KS-PCA), into Active Appearance Models, in order to model face appearances under variable illumination. The proposed algorithm successfully outperforms the traditional linear PCA transform to capture the salient features generated by different illuminations, and reconstruct the illuminated faces with high accuracy. We also consider the problem of automatically classifying human face poses from face views with varying illumination, as well as occlusion and noise. Based on the sparse representation methods, we propose two dictionary-learning frameworks for this pose classification problem. The first framework is the Adaptive Sparse Representation pose Classification (ASRC). It trains the dictionary via a linear model called Incremental Principal Component Analysis (Incremental PCA), tending to decrease the intra-class redundancy which may affect the classification performance, while keeping the extra-class redundancy which is critical for sparse representation. The other proposed work is the Dictionary-Learning Sparse Representation model (DLSR) that learns the dictionary with the aim of coinciding with the classification criterion. This training goal is achieved by the K-SVD algorithm. In a series of experiments, we show the performance of the two dictionary-learning methods which are respectively based on a linear transform and a sparse representation model. Besides, we propose a novel Dictionary Learning framework for Illumination Normalization (DL-IN). DL-IN based on sparse representation in terms of coupled dictionaries. The dictionary pairs are jointly optimized from normally illuminated and irregularly illuminated face image pairs. We further utilize a Gaussian Mixture Model (GMM) to enhance the framework's capability of modeling data under complex distribution. The GMM adapt each model to a part of the samples and then fuse them together. Experimental results demonstrate the effectiveness of the sparsity as a prior for patch-based illumination normalization for face images
Bertin-Mahieux, Thierry. "Apprentissage statistique pour l'étiquetage de musique et la recommandation." Thèse, 2009. http://hdl.handle.net/1866/7214.
Kannan, Sivakumar. "Molecular protein function prediction using sequence similarity-based and similarity-free approaches." Thèse, 2007. http://hdl.handle.net/1866/15681.
Thompson, Jessica A. F. "Characterizing and comparing acoustic representations in convolutional neural networks and the human auditory system." Thesis, 2020. http://hdl.handle.net/1866/24665.
Auditory processing in the human brain and in contemporary machine hearing systems consists of a cascade of representational transformations that extract and reorganize relevant information to enable task performance. This thesis is concerned with the nature of acoustic representations and the network design and learning principles that support their development. The primary scientific goals are to characterize and compare auditory representations in deep convolutional neural networks (CNNs) and the human auditory pathway. This work prompts several meta-scientific questions about the nature of scientific progress, which are also considered. The introduction reviews what is currently known about the mammalian auditory pathway and introduces the relevant concepts in deep learning.The first article argues that the most pressing philosophical questions at the intersection of artificial and biological intelligence are ultimately concerned with defining the phenomena to be explained and with what constitute valid explanations of such phenomena. I highlight relevant theories of scientific explanation which we hope will provide scaffolding for future discussion. Article 2 tests a popular model of auditory cortex based on frequency-specific spectrotemporal modulations. We find that a linear model trained only on BOLD responses to simple dynamic ripples (containing only one fundamental frequency, temporal modulation rate, and spectral scale) can generalize to predict responses to mixtures of two dynamic ripples. Both the third and fourth article investigate how CNN representations are affected by various aspects of training. The third article characterizes the language specificity of CNN layers and explores the effect of freeze training and random weights. We observed three distinct regions of transferability: (1) the first two layers were entirely transferable between languages, (2) layers 2--8 were also highly transferable but we found some evidence of language specificity, (3) the subsequent fully connected layers were more language specific but could be successfully finetuned to the target language. In Article 4, we use similarity analysis to find that the superior performance of freeze training achieved in Article 3 can be largely attributed to representational differences in the penultimate layer: the second fully connected layer. We also analyze the random networks from Article 3, from which we conclude that representational form is doubly constrained by architecture and the form of the input and target. To test whether acoustic CNNs learn a similar representational hierarchy as that of the human auditory system, the fifth article presents a similarity analysis to compare the activity of the freeze trained networks from Article 3 to 7T fMRI activity throughout the human auditory system. We find no evidence of a shared representational hierarchy and instead find that all of our auditory regions were most similar to the first fully connected layer. Finally, the discussion chapter reviews the merits and limitations of a deep learning approach to neuroscience in a model comparison framework. Together, these works contribute to the nascent enterprise of modeling the auditory system with neural networks and constitute a small step towards a unified science of intelligence that studies the phenomena that are exhibited in both biological and artificial intelligence.
Ghazi, Saidi Ladan. "Cross-Linguistic Transfer (CLT) in Bilingual Speakers : Neural Correlates of Language Learning." Thèse, 2012. http://hdl.handle.net/1866/8930.
The purpose of this thesis was to study the behavioral and neural correlates of Cross-linguistic Transfer effects (CLT) at the word level, in second language learning. Moreover, given that language distance has an impact on CLT, (Paradis, 1987, 2004, Odlin, 1989, 2004, 2005, Gollan, 2005, Ringbom, 2007), two distinct language pairs were examined: Close language pairs (Spanish-French) and distant language pairs (Persian-French). This thesis comprises three studies. In study I, Spanish speakers and in study II Persian speakers were trained for lexical learning until consolidation level. Cognates (phonologically and semantically similar words), Clangs (phonologically similar words with different meanings), and Non-cognate-non-clangs (semantically similar words), were presented in a picture naming task. Accuracy rates and response times as well as event-related fMRI BOLD responses to each word category were measured. Simple and direct contrasts with phonologically similar and phonologically distant words were performed. Thus, Study I reports the results of close languages (Spanish-French) and Study II, reports the results of distant languages (Persian-French). The neurocognitive processing of language learning was further investigated in terms of networks using functional connectivity analysis in distant languages (Persian-French) and the results are reported in Study III. The Results with the General Linear Model analysis show that with close language pairs (French-Spanish), the processing of phonologically similar words (cognates and clangs) relies upon a shared L1-L2 language specific neural areas, whereas processing of phonologically distant words (non-clang-non-cognates), activates L1 language processing areas, but also relies upon working memory, attentional, and processing structures. However, with distant language pairs (French-Persian), even phonologically similar words (cognates and clangs) activate areas known to be involved in attentional processing and cognitive control. Moreover, phonologically distant words (non-clang-non-cognates) also activate areas involved in working memory and executive function processing structures. Thus, the factor of L1-L2 cross-linguistic distance appears to modulate the executive load imposed to the system, on the basis of the degree of phonological overlap between L1-L2 items; thus in order to compensate for more effortful processing demands, the system recruits executive function supporting structures. The results of the connectivity analysis show that, in line with literature (Majerus, et al., 2008; Prat, et al., 2007; Veroude, et al., 2010; Dodel, et al., 2005; Coynel, et al., 2009), when the language proficiency is low, there is enhanced functional connectivity between and within language specific and other cognitive processing (working memory, attentional and cognitive control) networks. However, as proficiency increases, integration values (functional connectivity) decrease. This reflects that language tasks become less effortful and demand less cognitive resources. The results of this dissertation contribute to a better understanding of CLT effects on L2 learning, both in regards to different word types and L1-L2 language distance. These results have implications with regards to L2 learning and teaching strategies and approaches as well as with regards to the development of data-driven therapy approaches in the case of language break down in bilingual population.
Casagrande, Norman. "Automatic music classification using boosting algorithms and auditory features." Thèse, 2005. http://hdl.handle.net/1866/16694.
Maillet, François. "Algorithmes de recommandation musicale." Thèse, 2009. http://hdl.handle.net/1866/3875.
This thesis is composed of three papers which unite under the general theme of large-scale music recommendation. The first paper presents a recommendation technique that works by collecting text descriptions of items and using this textual aura to compute the similarity between them using techniques drawn from information retrieval. We show how this representation can be used to explain the similarities between items using terms from the textual aura and further how it can be used to steer the recommender. Because our system is content-based, it is not victim of the usual problems associated with collaborative filtering recommenders like the cold start problem. The second paper presents a machine learning model which automatically applies tags to music. The model uses features extracted from the audio files and was trained on a very large data set constructed with social data from the online community Last.fm. The third paper presents an approach to generating steerable playlists. We first demonstrate a method for learning song transition probabilities from audio features extracted from songs played in professional radio station playlists. We then show that by using this learnt similarity function as a prior, we are able to generate steerable playlists by choosing the next song to play not simply based on that prior, but on a tag cloud that the user is able to manipulate to express the high-level characteristics of the music he wishes to listen to.