Dissertations / Theses on the topic 'Réseaux de Graphes avec Attention'
Create a spot-on reference in APA, MLA, Chicago, Harvard, and other styles
Consult the top 26 dissertations / theses for your research on the topic 'Réseaux de Graphes avec Attention.'
Next to every source in the list of references, there is an 'Add to bibliography' button. Press on it, and we will generate automatically the bibliographic reference to the chosen work in the citation style you need: APA, MLA, Harvard, Chicago, Vancouver, etc.
You can also download the full text of the academic publication as pdf and read online its abstract whenever available in the metadata.
Browse dissertations / theses on a wide variety of disciplines and organise your bibliography correctly.
Amor, Yasmine. "Ιntelligent apprοach fοr trafic cοngestiοn predictiοn." Electronic Thesis or Diss., Normandie, 2024. http://www.theses.fr/2024NORMR129.
Full textTraffic congestion presents a critical challenge to urban areas, as the volume of vehicles continues to grow faster than the system’s overall capacity. This growth impacts economic activity, environmental sustainability, and overall quality of life. Although strategies for mitigating traffic congestion have seen improvements over the past few decades, many cities still struggle to manage it effectively. While various models have been developed to tackle this issue, existing approaches often fall short in providing real-time, localized predictions that can adapt to complex and dynamic traffic conditions. Most rely on fixed prediction horizons and lack the intelligent infrastructure needed for flexibility. This thesis addresses these gaps by proposing an intelligent, decentralized, infrastructure-based approach for traffic congestion estimation and prediction.We start by studying Traffic Estimation. We examine the possible congestion measures and data sources required for different contexts that may be studied. We establish a three-dimensional relationship between these axes. A rule-based system is developed to assist researchers and traffic operators in recommending the most appropriate congestion measures based on the specific context under study. We then proceed to Traffic Prediction, introducing our DECentralized COngestion esTimation and pRediction model using Intelligent Variable Message Signs (DECOTRIVMS). This infrastructure-based model employs intelligent Variable Message Signs (VMSs) to collect real-time traffic data and provide short-term congestion predictions with variable prediction horizons.We use Graph Attention Networks (GATs) due to their ability to capture complex relationships and handle graph-structured data. They are well-suited for modeling interactions between different road segments. In addition to GATs, we employ online learning methods, specifically, Stochastic Gradient Descent (SGD) and ADAptive GRAdient Descent (ADAGRAD). While these methods have been successfully used in various other domains, their application in traffic congestion prediction remains under-explored. In our thesis, we aim to bridge that gap by exploring their effectiveness within the context of real-time traffic congestion forecasting.Finally, we validate our model’s effectiveness through two case studies conducted in Muscat, Oman, and Rouen, France. A comprehensive comparative analysis is performed, evaluating various prediction techniques, including GATs, Graph Convolutional Networks (GCNs), SGD and ADAGRAD. The achieved results underscore the potential of DECOTRIVMS, demonstrating its potential for accurate and effective traffic congestion prediction across diverse urban contexts
Bertrand, Sébastien. "Optimisation de réseaux multiprotocoles avec encapsulation." Clermont-Ferrand 2, 2004. http://www.theses.fr/2004CLF21486.
Full textPoirier, Carl. "Assemblage d'ADN avec graphes de de Bruijn sur FPGA." Master's thesis, Université Laval, 2015. http://hdl.handle.net/20.500.11794/27132.
Full textJacob, Yann. "Classification dans les graphes hétérogènes et multi-relationnels avec contenu : Application aux réseaux sociaux." Paris 6, 2013. http://www.theses.fr/2013PA066494.
Full textThe emergence of the Web 2. 0 has seen the apparition of a large quantity of data that can easily be represented as complex graphs. There is many tasks of information analysis, prediction and retrieval on these data, while the state-of-the-art models are not adapted. In this thesis, we consider the task of node classification/labeling in complex partially labeled content networks. The applications for this task are for instance video/photo annotation in the Web 2. 0 websites, web spam detection or user labeling in social networks. The originality of our work is that we focus on two types of complex networks rarely considered in existing works: \textbf{multi-relationnal graphs} composed of multiple relation types and \textbf{heterogeneous networks} composed of multiple node types then of multiple joint labeling problems. First, we proposed two new algorithms for multi-relationnal graph labeling. These algorithms learn to weight the different relation types in the label propagation process according to their usefullness for the labeling task. They learn to combine the different relation types in an optimal manner for classification, while using the node content information. Then, we proposed an algorithm for heterogeneous graph labeling. Here, a specific problem is that each type of node has it own label set: for instance visual tags for a photo and groups for an user, then we must solve these different classification problems simultaneously using the graph structure. Our algorithm is based on the usage of a latent representation common to all node types allowing to process the different node types in an uniformized manner. Our experimental results show that this model is able to take in account the correlations between labels of different node types
Butelle, Franck. "Contribution à l'algorithmique distribuée de contrôle : arbres couvrants avec et sans contraintes." Paris 8, 1994. https://tel.archives-ouvertes.fr/tel-00082605.
Full textChopin, Morgan. "Problèmes d'optimisation avec propagation dans les graphes : complexité paramétrée et approximation." Phd thesis, Université Paris Dauphine - Paris IX, 2013. http://tel.archives-ouvertes.fr/tel-00933769.
Full textCarrillo, Hernan. "Colorisation d'images avec réseaux de neurones guidés par l'intéraction humaine." Electronic Thesis or Diss., Bordeaux, 2024. http://www.theses.fr/2024BORD0016.
Full textColorization is the process of adding colors to grayscale images. It is an important task in the image-editing and animation community. Although automatic colorization methods exist, they often produce unsatisfying results due to artifacts such as color bleeding, inconsistency, unnatural colors, and the ill-posed nature of the problem. Manual intervention is often necessary to achieve the desired outcome. Consequently, there is a growing interest in automating the colorization process while allowing artists to transfer their own style and vision to the process. In this thesis, we investigate various interaction formats by guiding colors of specific areas of an image or transferring them from a reference image or object. As part of this research, we introduce two semi-automatic colorization frameworks. First, we describe a deep learning architecture for exemplar-based image colorization that takes into account user’s reference images. Our second framework uses a diffusion model to colorize line art using user-provided color scribbles. This thesis first delves into a comprehensive overview of state-of-the-art image colorization methods, color spaces, evaluation metrics, and losses. While recent colorization methods based on deep-learning techniques are achieving the best results on this task, these methods are based on complex architectures and a high number of joint losses, which makes the reasoning behind each of these methods difficult. Here, we leverage a simple architecture in order to analyze the impact of different color spaces and several losses. Then, we propose a novel attention layer based on superpixel features to establish robust correspondences between high-resolution deep features from target and reference image pairs, called super-attention. This proposal deals with the quadratic complexity problem of the non-local calculation in the attention layer. Additionally, it helps to overcome color bleeding artifacts. We study its use in color transfer and exemplar-based colorization. We finally extend this model to specifically guide the colorization on segmented objects. Finally, we propose a diffusion probabilistic model based on implicit and explicit conditioning mechanism, to learn colorizing line art. Our approach enables the incorporation of user guidance through explicit color hints while leveraging on the prior knowledge from the trained diffusion model. We condition with an application-specific encoder that learns to extract meaningful information on user-provided scribbles. The method generates diverse and high-quality colorized images
Pirayre, Aurélie. "Reconstruction et classification par optimisation dans des graphes avec à priori pour les réseaux de gènes et les images." Thesis, Paris Est, 2017. http://www.theses.fr/2017PESC1170/document.
Full textThe discovery of novel gene regulatory processes improves the understanding of cell phenotypicresponses to external stimuli for many biological applications, such as medicine, environmentor biotechnologies. To this purpose, transcriptomic data are generated and analyzed from mi-croarrays or more recently RNAseq experiments. For each gene of a studied organism placed indifferent living conditions, they consist in a sequence of genetic expression levels. From thesedata, gene regulation mechanisms can be recovered by revealing topological links encoded ingeometric graphs. In regulatory graphs, nodes correspond to genes. A link between two nodesis identified if a regulation relationship exists between the two corresponding genes. Such net-works are called Gene Regulatory Networks (GRNs). Their construction as well as their analysisremain challenging despite the large number of available inference methods.In this thesis, we propose to address this network inference problem with recently developedtechniques pertaining to graph optimization. Given all the pairwise gene regulation informa-tion available, we propose to determine the presence of edges in the final GRN by adoptingan energy optimization formulation integrating additional constraints. Either biological (infor-mation about gene interactions) or structural (information about node connectivity) a priorihave been considered to reduce the space of possible solutions. Different priors lead to differentproperties of the global cost function, for which various optimization strategies can be applied.The post-processing network refinements we proposed led to a software suite named BRANE for“Biologically-Related A priori for Network Enhancement”. For each of the proposed methodsBRANE Cut, BRANE Relax and BRANE Clust, our contributions are threefold: a priori-based for-mulation, design of the optimization strategy and validation (numerical and/or biological) onbenchmark datasets.In a ramification of this thesis, we slide from graph inference to more generic data processingsuch as inverse problems. We notably invest in HOGMep, a Bayesian-based approach using aVariation Bayesian Approximation framework for its resolution. This approach allows to jointlyperform reconstruction and clustering/segmentation tasks on multi-component data (for instancesignals or images). Its performance in a color image deconvolution context demonstrates bothquality of reconstruction and segmentation. A preliminary study in a medical data classificationcontext linking genotype and phenotype yields promising results for forthcoming bioinformaticsadaptations
Hizem, Mohamed Mejdi. "Recherche de chemins dans un graphe à pondérationdynamique : application à l'optimisation d'itinéraires dans les réseaux routiers." Phd thesis, Ecole Centrale de Lille, 2008. http://tel.archives-ouvertes.fr/tel-00344958.
Full textRoux, Marine. "Inférence de graphes par une procédure de test multiple avec application en Neuroimagerie." Thesis, Université Grenoble Alpes (ComUE), 2018. http://www.theses.fr/2018GREAT058/document.
Full textThis thesis is motivated by the analysis of the functional magnetic resonance imaging (fMRI). The need for methods to build such structures from fMRI data gives rise to exciting new challenges for mathematics. In this regards, the brain connectivity networks are modelized by a graph and we study some procedures that allow us to infer this graph.More precisely, we investigate the problem of the inference of the structure of an undirected graphical model by a multiple testing procedure. The structure induced by both the correlation and the partial correlation are considered. The statistical tests based on the latter are known to be highly dependent and we assume that they have an asymptotic Gaussian distribution. Within this framework, we study some multiple testing procedures that allow a control of false edges included in the inferred graph.First, we theoretically examine the False Discovery Rate (FDR) control of Benjamini and Hochberg’s procedure in Gaussian setting for non necessary positive dependent statistical tests. Then, we explore both the FDR and the Family Wise Error Rate (FWER) control in asymptotic Gaussian setting. We present some multiple testing procedures, well-suited for correlation (resp. partial correlation) tests, which provide an asymptotic control of the FWER. Furthermore, some first theoretical results regarding asymptotic FDR control are established.Second, the properties of the multiple testing procedures that asymptotically control the FWER are illustrated on a simulation study, for statistical tests based on correlation. We finally conclude with the extraction of cerebral connectivity networks on real data set
Halftermeyer, Pierre. "Connexité dans les Réseaux et Schémas d’Étiquetage Compact d’Urgence." Thesis, Bordeaux, 2014. http://www.theses.fr/2014BORD0140/document.
Full textWe aim at assigning each vertex x of a n-vertices graph G a compact O(log n)-bit label L(x) in order to :1. construct, from the labels of the vertices of a forbidden set X C V (G), a datastructure S(X)2. decide, from S(X), L(u) and L(v), whether two vertices u and v are connected in G n X.We give a solution to this problem for the family of 3-connected graphs whith bounded genus.— We obtain O(g log n)-bit labels.— S(X) is computed in O(Sort([X]; n)) time.— Connection between vertices is decided in O(log log n) optimal time.We finally extend this result to H-minor-free graphs. This scheme requires O(polylog n)-bit labels
Coadou, Anthony. "Réseaux de processus flots de données avec routage pour la modélisation de systèmes embarqués." Phd thesis, Université de Nice Sophia-Antipolis, 2010. http://tel.archives-ouvertes.fr/tel-00545008.
Full textGONZáLEZ, GóMEZ Mauricio. "Jeux stochastiques sur des graphes avec des applications à l’optimisation des smart-grids." Thesis, Université Paris-Saclay (ComUE), 2019. http://www.theses.fr/2019SACLN064.
Full textWithin the research community, there is a great interest in exploring many applications of energy grids since these become more and more important in our modern world. To properly design and implement these networks, advanced and complex mathematical tools are necessary. Two key features for their design are correctness and optimality. While these last two properties are in the core of formal methods, their effective application to energy networks remains largely unexploited. This constitutes one strong motivation for the work developed in this thesis. A special emphasis is made on the generic problem of scheduling power consumption. This is a scenario in which the consumers have a certain energy demand and want to have this demand fulfilled before a set deadline (e.g., an Electric Vehicle (EV) has to be recharged within a given time window set by the EV owner). Therefore, each consumer has to choose at each time the consumption power (by a computerized system) so that the final accumulated energy reaches a desired level. The way in which the power levels are chosen is according to a ``strategy’’ mapping at any time the relevant information of a consumer (e.g., the current accumulated energy for EV-charging) to a suitable power consumption level. The design of such strategies may be either centralized (in which there is a single decision-maker controlling all strategies of consumers), or decentralized (in which there are several decision-makers, each of them representing a consumer). We analyze both scenarios by exploiting ideas originating from formal methods, game theory and optimization. More specifically, the power consumption scheduling problem can be modelled using Markov decision processes and stochastic games. For instance, probabilities provide a way to model the environment of the electrical system, namely: the noncontrollable part of the total consumption (e.g., the non-EV consumption). The controllable consumption can be adapted to the constraints of the distribution network (e.g., to the maximum shutdown temperature of the electrical transformer), and to their objectives (e.g., all EVs are recharged). At first glance, this can be seen as a stochastic system with multi-constraints objectives. Therefore, the contributions of this thesis also concern the area of multi-criteria objective models, which allows one to pursue several objectives at a time such as having strategy designs functionally correct and robust against changes of the environment
Butelle, Franck. "Contribution à l'algorithmique distribuée de contrôle : arbres couvrants avec et sans containtes." Phd thesis, Université Paris VIII Vincennes-Saint Denis, 1994. http://tel.archives-ouvertes.fr/tel-00082605.
Full textalgorithmes distribués asynchrones et déterministes de
contröle. Un système distribué consiste en un réseau
de sites (processeurs, ordinateurs ou réseaux locaux). Dans cette
thèse, nous ne considérons que des réseaux de sites
communicants n'ayant ni mémoire partagée ni horloge globale.
De nombreux problèmes de l'algorithmique distribuée sont
réductibles à la construction d'un Arbre Couvrant qui est la
structure de contrôle qui nous intéresse.
Nous étudions deux types d'algorithmes~: ceux utilisant
la notion de phase logique et les autres qui ne considèrent aucun
mécanisme de synchronisation. Ces derniers ont des comportements
imprévisibles améliorant la tolérance aux fautes. Nous
présentons un nouvel algorithme de ce type associé à une
élection qui n'est pas une recherche d'extremum contrairement
à l'usage. Cet algorithme est comparable au meilleur
algorithme connu qui utilise des jetons et des phases logiques
induisant un comportement plus "séquentiel".
D'autres algorithmes, construisant des AC contraints, sont
considérés. En particulier l'AC de Diamètre Minimum qui
est, à notre connaissance, un problème qui n'a jamais
été étudié dans ce domaine. Le diamètre d'un
graphe est la somme des poids des arêtes du plus long des plus
courts chemins. Si nous considérons la complexité temporelle,
cette contrainte est d'un intérêt &vident. Nous proposons
différents algorithmes suivant que la tolérance aux fautes est
nécessaire ou non.
Finalement, l'étude pratique des algorithmes distribués sur
des réseaux de grande taille nous a conduit à la construction
d'un simulateur. Il permet l'exécution d'un même code source
sur des machines séquentielles ou parallèles.
Minescu, Bogdan. "Construction et stratégie d'exploitation des réseaux de confusion en lien avec le contexte applicatif de la compréhension de la parole." Phd thesis, Université d'Avignon, 2008. http://tel.archives-ouvertes.fr/tel-00629195.
Full textBelilovsky, Eugene. "Apprentissage de graphes structuré et parcimonieux dans des données de haute dimension avec applications à l’imagerie cérébrale." Thesis, Université Paris-Saclay (ComUE), 2018. http://www.theses.fr/2018SACLC027.
Full textThis dissertation presents novel structured sparse learning methods on graphs that address commonly found problems in the analysis of neuroimaging data as well as other high dimensional data with few samples. The first part of the thesis proposes convex relaxations of discrete and combinatorial penalties involving sparsity and bounded total variation on a graph as well as bounded `2 norm. These are developed with the aim of learning an interpretable predictive linear model and we demonstrate their effectiveness on neuroimaging data as well as a sparse image recovery problem.The subsequent parts of the thesis considers structure discovery of undirected graphical models from few observational data. In particular we focus on invoking sparsity and other structured assumptions in Gaussian Graphical Models (GGMs). To this end we make two contributions. We show an approach to identify differences in Gaussian Graphical Models (GGMs) known to have similar structure. We derive the distribution of parameter differences under a joint penalty when parameters are known to be sparse in the difference. We then show how this approach can be used to obtain confidence intervals on edge differences in GGMs. We then introduce a novel learning based approach to the problem structure discovery of undirected graphical models from observational data. We demonstrate how neural networks can be used to learn effective estimators for this problem. This is empirically shown to be flexible and efficient alternatives to existing techniques
Gladkikh, Egor. "Optimisation de l'architecture des réseaux de distribution d'énergie électrique." Thesis, Université Grenoble Alpes (ComUE), 2015. http://www.theses.fr/2015GREAT055/document.
Full textTo cope with the changes in the energy landscape, electrical distribution networks are submitted to operational requirements in order to guarantee reliability indices. In the coming years, big investments are planned for the construction of flexible, consistent and effective electrical networks, based on the new architectures, innovative technical solutions and in response to the development of renewable energy. Taking into account the industrial needs of the development of future distribution networks, we propose in this thesis an approach based on the graph theory and combinatorial optimization for the design of new architectures for distribution networks. Our approach is to study the general problem of finding an optimal architecture which respects a set of topological (redundancy) and electrical (maximum current, voltage plan) constraints according to precise optimization criteria: minimization of operating cost (OPEX) and minimization of investment (CAPEX). Thus, the two families of combinatorial problems (and their relaxations) were explored to propose effective resolutions (exact or approximate) of the distribution network planning problem using an adapted formulation. We are particularly interested in 2-connected graphs and the arborescent flow problem with minimum quadratic losses. The comparative results of tests on the network instances (fictional and real) for the proposed methods were presented
Chen, Dexiong. "Modélisation de données structurées avec des machines profondes à noyaux et des applications en biologie computationnelle." Thesis, Université Grenoble Alpes, 2020. http://www.theses.fr/2020GRALM070.
Full textDeveloping efficient algorithms to learn appropriate representations of structured data, including sequences or graphs, is a major and central challenge in machine learning. To this end, deep learning has become popular in structured data modeling. Deep neural networks have drawn particular attention in various scientific fields such as computer vision, natural language understanding or biology. For instance, they provide computational tools for biologists to possibly understand and uncover biological properties or relationships among macromolecules within living organisms. However, most of the success of deep learning methods in these fields essentially relies on the guidance of empirical insights as well as huge amounts of annotated data. Exploiting more data-efficient models is necessary as labeled data is often scarce.Another line of research is kernel methods, which provide a systematic and principled approach for learning non-linear models from data of arbitrary structure. In addition to their simplicity, they exhibit a natural way to control regularization and thus to avoid overfitting.However, the data representations provided by traditional kernel methods are only defined by simply designed hand-crafted features, which makes them perform worse than neural networks when enough labeled data are available. More complex kernels inspired by prior knowledge used in neural networks have thus been developed to build richer representations and thus bridge this gap. Yet, they are less scalable. By contrast, neural networks are able to learn a compact representation for a specific learning task, which allows them to retain the expressivity of the representation while scaling to large sample size.Incorporating complementary views of kernel methods and deep neural networks to build new frameworks is therefore useful to benefit from both worlds.In this thesis, we build a general kernel-based framework for modeling structured data by leveraging prior knowledge from classical kernel methods and deep networks. Our framework provides efficient algorithmic tools for learning representations without annotations as well as for learning more compact representations in a task-driven way. Our framework can be used to efficiently model sequences and graphs with simple interpretation of predictions. It also offers new insights about designing more expressive kernels and neural networks for sequences and graphs
Chevallier, Sylvain. "Implémentation d'un système préattentionnel avec des neurones impulsionnels." Phd thesis, Université Paris Sud - Paris XI, 2009. http://tel.archives-ouvertes.fr/tel-00472849.
Full textNait-Sidi-Moh, Ahmed. "Contribution à la modélisation, l'analyse et la commande des systèmes à événements discrets par les réseaux de Petri et l'algèbre (max, plus) : Application aux systèmes de transport." Phd thesis, Université de Technologie de Belfort-Montbeliard, 2003. http://tel.archives-ouvertes.fr/tel-00467580.
Full textChen, Rui. "Dynamic optimal control for distress large financial networks and Mean field systems with jumps Optimal connectivity for a large financial network Mean Field BSDEs and Global Dynamic Risk Measures." Thesis, Paris Sciences et Lettres (ComUE), 2019. https://portail.bu.dauphine.fr/fileviewer/index.php?doc=2019PSLED042.
Full textThis thesis presents models and methodologies to understand the control of systemic risk in large systems. We propose two approaches. The first one is structural : a financial system is represented as a network of institutions. They have strategic interactions as well as direct interactions through linkages in a contagion process. The novelty of our approach is that these two types of interactions are intertwined themselves and we propose new notions of equilibria for such games and analyze the systemic risk emerging in equilibrium. The second approach is a reduced form.We model the dynamics of regulatory capital using a mean field operator : required capital depends on the standalone risk but also on the evolution of the capital of all other banks in the system. In this model, required capital is a dynamic risk measure and is represented as a the solution of a mean-field BDSE with jumps. We show a novel dual representation theorem. In the context of meanfield BSDEs the representation gives yield to a stochastic discount factor and a worst-case probability measure that encompasses the overall interactions in the system. We also solve the optimal stopping problem of dynamic risk measure by connecting it to the solution of reflected meanfield BSDE with jumps. Finally, We provide a comprehensive model for the order book dynamics and optimal Market making strategy appeared in liquidity risk problems
Belhadj, Djedjiga. "Multi-GAT semi-supervisé pour l’extraction d’informations et son adaptation au chiffrement homomorphe." Electronic Thesis or Diss., Université de Lorraine, 2024. http://www.theses.fr/2024LORR0023.
Full textThis thesis is being carried out as part of the BPI DeepTech project, in collaboration with the company Fair&Smart, primarily looking after the protection of personal data in accordance with the General Data Protection Regulation (RGPD). In this context, we have proposed a deep neural model for extracting information in semi-structured administrative documents (SSDs). Due to the lack of public training datasets, we have proposed an artificial generator of SSDs that can generate several classes of documents with a wide variation in content and layout. Documents are generated using random variables to manage content and layout, while respecting constraints aimed at ensuring their similarity to real documents. Metrics were introduced to evaluate the content and layout diversity of the generated SSDs. The results of the evaluation have shown that the generated datasets for three SSD types (payslips, receipts and invoices) present a high diversity level, thus avoiding overfitting when training the information extraction systems. Based on the specific format of SSDs, consisting specifically of word pairs (keywords-information) located in spatially close neighborhoods, the document is modeled as a graph where nodes represent words and edges, neighborhood connections. The graph is fed into a multi-layer graph attention network (Multi-GAT). The latter applies the multi-head attention mechanism to learn the importance of each word's neighbors in order to better classify it. A first version of this model was used in supervised mode and obtained an F1 score of 96% on two generated invoice and payslip datasets, and 89% on a real receipt dataset (SROIE). We then enriched the multi-GAT with multimodal embedding of word-level information (textual, visual and positional), and combined it with a variational graph auto-encoder (VGAE). This model operates in semi-supervised mode, being able to learn on both labeled and unlabeled data simultaneously. To further optimize the graph node classification, we have proposed a semi-VGAE whose encoder shares its first layers with the multi-GAT classifier. This is also reinforced by the proposal of a VGAE loss function managed by the classification loss. Using a small unlabeled dataset, we were able to improve the F1 score obtained on a generated invoice dataset by over 3%. Intended to operate in a protected environment, we have adapted the architecture of the model to suit its homomorphic encryption. We studied a method of dimensionality reduction of the Multi-GAT model. We then proposed a polynomial approximation approach for the non-linear functions in the model. To reduce the dimensionality of the model, we proposed a multimodal feature fusion method that requires few additional parameters and reduces the dimensions of the model while improving its performance. For the encryption adaptation, we studied low-degree polynomial approximations of nonlinear functions, using knowledge distillation and fine-tuning techniques to better adapt the model to the new approximations. We were able to minimize the approximation loss by around 3% on two invoice datasets as well as one payslip dataset and by 5% on SROIE
Riva, Mateus. "Spatial Relational Reasoning in Machine Learning : Deep Learning and Graph Clustering." Electronic Thesis or Diss., Institut polytechnique de Paris, 2022. http://www.theses.fr/2022IPPAT043.
Full textThis thesis studies the capabilities of machine learning methods for reasoning on spatial relationships, with a particular focus on directional relationships, and the use of prior relational information by these methods. There are many works in the field of applying knowledge on relationships to machine learning methods. However, this body of work still leaves several open questions. Throughout this thesis, we explore, investigate and attempt to explain different research questions linked to this field.We propose an improvement to the training of CNNs via a regularisation loss function based on relational information. To this end, we propose two novel loss functions which reward relationship satisfaction during CNN training, and design synthetic experiments to showcase their impact. While the proposed loss functions show improvements over an unmodified baseline in specific, strict synthetic scenarios, the impact on more ``generic'' training scenarios is less significant. This result is not easily explainable, as neural network training is a significantly opaque process, and as such, a deeper exploration is required to understand how a CNN learns (or fails to learn) to reason using relational information.To further understanding of how a CNN can learn to reason using relational information, we propose a wide array of distinct synthetic experiments. We explore the processes which enable, facilitate, or hinder ``standard'' CNN reasoning on relationships. We propose a fundamental experience to demonstrate that a basic, unmodified CNN is capable of relational reasoning in some scenarios. Next, we explore which relationships are learned by the CNN, by performing inference on scenes where the prior relationships are disturbed, by recording the difference in the results, and by training and testing CNNs on synthetic data with more or less relationships available. We then investigate the limits placed on relational reasoning by the network receptive field, as well as deepen our analysis on situations where the amount of training data is insufficient. Finally, we explore at which moment during training relationships are satisfied, as a proxy for understanding at which moment the relationships themselves are learned.Following a graph-clustering approach to the usage of relational information, we explore prior relationships in a different machine learning context, that of community discovery on graphs. We formulate graph clustering as an inexact matching problem between the graph to be clustered and a model graph which encodes prior knowledge on how the communities or clusters relate to each other. We compare this approach with traditional graph clustering approaches on a set of synthetic graphs, to showcase the advantages of a relational-aware approach, as well as on real graphs
El, Haj Abir. "Stochastics blockmodels, classifications and applications." Thesis, Poitiers, 2019. http://www.theses.fr/2019POIT2300.
Full textThis PhD thesis focuses on the analysis of weighted networks, where each edge is associated to a weight representing its strength. We introduce an extension of the binary stochastic block model (SBM), called binomial stochastic block model (bSBM). This question is motivated by the study of co-citation networks in a context of text mining where data is represented by a graph. Nodes are words and each edge joining two words is weighted by the number of documents included in the corpus simultaneously citing this pair of words. We develop an inference method based on a variational maximization algorithm (VEM) to estimate the parameters of the modelas well as to classify the words of the network. Then, we adopt a method based on maximizing an integrated classification likelihood (ICL) criterion to select the optimal model and the number of clusters. Otherwise, we develop a variational approach to analyze the given network. Then we compare the two approaches. Applications based on real data are adopted to show the effectiveness of the two methods as well as to compare them. Finally, we develop a SBM model with several attributes to deal with node-weighted networks. We motivate this approach by an application that aims at the development of a tool to help the specification of different cognitive treatments performed by the brain during the preparation of the writing
Legrand, Jonathan. "Toward a multi-scale understanding of flower development - from auxin networks to dynamic cellular patterns." Thesis, Lyon, École normale supérieure, 2014. http://www.theses.fr/2014ENSL0947/document.
Full textA striking aspect of flowering plants is that, although they seem to display a great diversity of size and shape, they are made of the same basics constituents, that is the cells. The major challenge is then to understand how multicellular tissues, originally undifferentiated, can give rise to such complex shapes. We first investigated the uncharacterised signalling network of auxin since it is a major phytohormone involved in flower organogenesis.We started by determining the potential binary network, then applied model-based graph clustering methods relying on connectivity profiles. We demonstrated that it could be summarise in three groups, closely related to putative biological groups. The characterisation of the network function was made using ordinary differential equation modelling, which was later confirmed by experimental observations.In a second time, we modelled the influence of the protein dimerisation sequences on the auxin interactome structure using mixture of linear models for random graphs. This model lead us to conclude that these groups behave differently, depending on their dimerisation sequence similarities, and that each dimerisation domains might play different roles.Finally, we changed scale to represent the observed early stages of A. thaliana flower development as a spatio-temporal property graph. Using recent improvements in imaging techniques, we could extract 3D+t cellular features, and demonstrated the possibility of identifying and characterising cellular identity on this basis. In that respect, hierarchical clustering methods and hidden Markov tree have proven successful in grouping cell depending on their feature similarities
Minescu, Bogdan. "Construction et stratégie d’exploitation des réseaux de confusion en lien avec le contexte applicatif de la compréhension de la parole." Thesis, 2008. http://www.theses.fr/2008AVIG0176/document.
Full textThe work presented in this PhD deals with the confusion networks as a compact and structured representation of multiple aligned recognition hypotheses produced by a speech recognition system and used by different applications. The confusion networks (CN) are constructed from word graphs and structure information as a sequence of classes containing several competing word hypothesis. In this work we focus on the problem of robust understanding from spontaneous speech input in a dialogue application, using CNs as structured representation of recognition hypotheses for the spoken language understanding module. We use France Telecom spoken dialogue system for customer care. Two issues inherent to this context are tackled. A dialogue system does not only have to recognize what a user says but also to understand the meaning of his request and to act upon it. From the user’s point of view, system performance is more accurately represented by the performance of the understanding process than by speech recognition performance only. Our work aims at improving the performance of the understanding process. Using a real application implies being able to process real heterogeneous data. An utterance can be more or less noisy, in the domain or out of the domain of the application, covered or not by the semantic model of the application, etc. A question raised by the variability of the data is whether applying the same processes to the entire data set, as done in classical approaches, is a suitable solution. This work follows a double perspective : to improve the CN construction algorithm with the intention of optimizing the understanding process and to propose an adequate strategy for the use of CN in a real application. Following a detailed analysis of two CN construction algorithms on a test set collected using the France Telecom customer care service, we decided to use the "pivot" algorithm for our work. We present a modified and adapted version of this algorithm. The new algorithm introduces different processing techniques for the words which are important for the understanding process. As for the variability of the real data the application has to process, we present a new multiple level decision strategy aiming at applying different processing techniques for different utterance categories. We show that it is preferable to process multiple recognition hypotheses only on utterances having a valid interpretation. This strategy optimises computation time and yields better global performance