Dissertations / Theses on the topic 'Apprentissage continu en ligne'
Create a spot-on reference in APA, MLA, Chicago, Harvard, and other styles
Consult the top 50 dissertations / theses for your research on the topic 'Apprentissage continu en ligne.'
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.
Wagner, Baptiste. "Apprentissage continu en ligne pour la classification d'images et la détection d'objets." Electronic Thesis or Diss., Université Grenoble Alpes, 2024. http://www.theses.fr/2024GRALT111.
Full textIn this thesis, we focus on the problem of online continual learning in artificial neural networks, which involves learning continuously from a data stream. The main challenge is that integrating new information from the stream tends to overwrite previously acquired knowledge, a phenomenon known as catastrophic forgetting.In the field of online continual learning, our research focus on two important applications in computer vision: image classification and object detection. In these cases, the data stream consists of a sequence of images.In image classification, the neural network must progressively learn to classify images from new classes without forgetting the previous ones. The most common method to address this problem is experience replay, which involves retraining the model with images from previously seen classes stored in external memory. However, this method is less suitable when both storage capacity and computational resources are limited. We propose a new method based on a one-vs-all classifier training scheme to overcome this limitation. Our method, called ILOVA (Incremental Learning of One-Vs-All classifiers), offers a better trade-off between accuracy, forgetting, computational time, and memory footprint compared to state-of-the-art methods and proves particularly effective with very limited memory, down to a single image per class.In object detection, many test scenarios are constructed from real video sequences in which objects can reappear multiple times at different moments in the data stream. However, this phenomenon of reappearance, which we call natural replay, is poorly documented, and its impact on performance and forgetting remains poorly understood. We propose a new metric, called NRS (Natural Replay Score), which quantifies the degree of natural replay in a scenario, and show that it is impossible to properly evaluate model forgetting in its presence. The next part of our study focuses on analyzing forgetting in the Faster R-CNN architecture when used for online object detection. On the one hand, our results show that periodic recalls reduce forgetting. On the other hand, we propose a new protocol, called Module Probing, which allows us to measure forgetting locally within the architecture. We show that forgetting is concentrated in the classification layer of Faster R-CNN. Finally, these analyses lead us to propose a method called Configurable Recall, based on experience replay. Our method optimizes the frequency and duration of the recalls and uses a modified loss function to limit forgetting in the classification layer. By combining these two elements, we significantly reduce forgetting in the Faster R-CNN architecture
Oulhadj, Hamouche. "Des primitives aux lettres : une méthode structurelle de reconnaissance en ligne de mots d'écriture cursive manuscrite avec un apprentissage continu." Paris 12, 1990. http://www.theses.fr/1990PA120045.
Full textYang, Rui. "Online continual learning for 3D detection of road participants in autonomous driving." Electronic Thesis or Diss., Bourgogne Franche-Comté, 2023. http://www.theses.fr/2023UBFCA021.
Full textAutonomous driving has witnessed remarkable progress over the past decades, and machine perception stands as a critical foundational issue, encompassing the detection and tracking of road participants such as vehicles, pedestrians, and cyclists. While vision-based object detection has achieved significant progress thanks to deep learning techniques, challenges still exist in 3D detection.Firstly, non-visual sensors, such as 3D LiDAR, demonstrate unparalleled advantages in achieving precise detection and adaptability to varying lighting conditions. However, the complexity of handling points cloud data, which can be challenging to interpret, coupled with the high cost of manual annotation, pose primary challenges in the use of 3D LiDAR.Secondly, concerns arise from the lack of interpretability in deep learning models, coupled with their heavy reliance on extensive training data, which often necessitates costly retraining for acceptable generalization performance when adapting to new scenes or environments.This dissertation addresses these challenges from three main perspectives: Generation of Samples, Preservation of Knowledge, and Avoidance of Catastrophic Forgetting. We introduce the concept of Online Continual Learning (OCL) and propose a general framework that encompasses detection, tracking, learning, and control. This framework enables models to update in real-time, preserving knowledge rather than raw data, and effectively mitigating the performance degradation caused by catastrophic forgetting.The main work of this dissertation includes: 1) Generation of Samples: To address sparse point clouds generated by 3D LiDAR and the labor-intensive manual annotation, we leverage the advantages of multi-sensor data and employ an efficient online transfer learning framework. This framework effectively transfers mature image-based detection capabilities to 3D LiDAR-based detectors. An innovative aspect is the "learn-by-use" process, achieved through closed-loop detection, facilitating continuous self-supervised learning. A novel information fusion strategy is proposed to combine spatio-temporal correlations, enhancing the effectiveness of knowledge transfer. 2) Preservation of Knowledge: Online Learning (OL) is introduced to address knowledge preservation without retaining training data. An improved Online Random Forest (ORF) model is incorporated, enabling rapid model training with limited computational resources and immediate deployment. The ORF model's parameters are dynamically shared throughout the training process to address the unknown data distribution. The exploration of ORF tree structures ensures independence in training processes, enhancing the model's ability to capture complex patterns and variations. Implementing octrees improves storage efficiency and model access. 3) Avoidance of Catastrophic Forgetting: To tackle the inevitable forgetting problem in online learning frameworks during long-term deployment, we propose the Long Short-Term Online Learning (LSTOL) framework. LSTOL combines multiple short-term learners based on ensemble learning with a long-term controller featuring a probabilistic decision mechanism. This framework ensures effective knowledge maintenance and adapts to changes during long-term deployment, without making assumptions about model types and data continuity. Cross-dataset evaluations on tasks such as 3D detection of road participants demonstrate the effectiveness of LSTOL in avoiding forgetting
Hocquet, Guillaume. "Class Incremental Continual Learning in Deep Neural Networks." Thesis, université Paris-Saclay, 2021. http://www.theses.fr/2021UPAST070.
Full textWe are interested in the problem of continual learning of artificial neural networks in the case where the data are available for only one class at a time. To address the problem of catastrophic forgetting that restrain the learning performances in these conditions, we propose an approach based on the representation of the data of a class by a normal distribution. The transformations associated with these representations are performed using invertible neural networks, which can be trained with the data of a single class. Each class is assigned a network that will model its features. In this setting, predicting the class of a sample corresponds to identifying the network that best fit the sample. The advantage of such an approach is that once a network is trained, it is no longer necessary to update it later, as each network is independent of the others. It is this particularly advantageous property that sets our method apart from previous work in this area. We support our demonstration with experiments performed on various datasets and show that our approach performs favorably compared to the state of the art. Subsequently, we propose to optimize our approach by reducing its impact on memory by factoring the network parameters. It is then possible to significantly reduce the storage cost of these networks with a limited performance loss. Finally, we also study strategies to produce efficient feature extractor models for continual learning and we show their relevance compared to the networks traditionally used for continual learning
Désoyer, Adèle. "Appariement de contenus textuels dans le domaine de la presse en ligne : développement et adaptation d'un système de recherche d'information." Thesis, Paris 10, 2017. http://www.theses.fr/2017PA100119/document.
Full textThe goal of this thesis, conducted within an industrial framework, is to pair textual media content. Specifically, the aim is to pair on-line news articles to relevant videos for which we have a textual description. The main issue is then a matter of textual analysis, no image or spoken language analysis was undertaken in the present study. The question that arises is how to compare these particular objects, the texts, and also what criteria to use in order to estimate their degree of similarity. We consider that one of these criteria is the topic similarity of their content, in other words, the fact that two documents have to deal with the same topic to form a relevant pair. This problem fall within the field of information retrieval (ir) which is the main strategy called upon in this research. Furthermore, when dealing with news content, the time dimension is of prime importance. To address this aspect, the field of topic detection and tracking (tdt) will also be explored.The pairing system developed in this thesis distinguishes different steps which complement one another. In the first step, the system uses natural language processing (nlp) methods to index both articles and videos, in order to overcome the traditionnal bag-of-words representation of texts. In the second step, two scores are calculated for an article-video pair: the first one reflects their topical similarity and is based on a vector space model; the second one expresses their proximity in time, based on an empirical function. At the end of the algorithm, a classification model learned from manually annotated document pairs is used to rank the results.Evaluation of the system's performances raised some further questions in this doctoral research. The constraints imposed both by the data and the specific need of the partner company led us to adapt the evaluation protocol traditionnal used in ir, namely the cranfield paradigm. We therefore propose an alternative solution for evaluating the system that takes all our constraints into account
Boudjema, Cédric. "La fonction éducative des musées dans la société numérique : analyse comparative de l'offre pédagogique en ligne de huit musées nationaux dans quatre pays (France, Angleterre, Australie, Etats-Unis)." Thesis, Lille 3, 2016. http://www.theses.fr/2016LIL30013/document.
Full textThis research studies museum internet sites and in particular the pedagogy of eight national institutions in four different countries and suggests that online museums are educational content players.The interest is to investigate the educational content of the internet sites using a content analysis and implementing a comparison between the four countries and the types of internet sites to be able to understand the practices – and especially what Jean Davallon calls « the anticipation by the “sender” » that the visitor will engage in (the sender aiming for example to keep the attention of the latter or to provide guidance in the contents), the typology of content and the teaching strategies put in place by the online museum institutions. The online educational offer is defined here as a permanent activity as a source of building knowledge, consultation, criticism, and entertainment, from the museum resources. This offer is also constructed according to the consistent rules of Web design.We have chosen to study the online pedagogy according to a constructivist approach that drives us to privilege certain key concepts : individual learning ways, learning processes, cognitive strategies, meta-cognitive strategies, {learning styles}, taxonomy. From a methodological point of view, this thesis relies on a qualitative approach and privileges a content analysis from an analysis grid with eleven categories : the corpus is composed of eight internet sites and of two types of national museums : the art museums and the science museums with an educational section. The thesis is composed of two tomes. The tome 2 contains the complete analysis of the sites and the tome 1 includes three parts. In the first part, the research discusses the educational role of museums with its specificities and complexities. This part defines the historical context of the educational function of museums that very early on developed an educational strategy for the public. It also shows the specificity of museums in informal education as a place of learning concepts and development that develop two types of mediation. The museum favours the formulation of questions; it orientates reflexion and raises questions. It then shows the museum as an important partner and complementary to school. Finally, this part precises the historical context of online museums of the four countries from our analysis and the progressive development of the cultural policies of the present and the progressive actions put into place by the museums.Secondly, the research focuses on the thematic analysis of the internet sites and on their educational sections and attempts to show the successive steps of the content analysis via the analysis grid constructed for this research. Firstly, it is about showing the ergonomics of the sites to progressively arrive upon the general treatment of the educational sections of the sites, that is to say to identify the mechanisms of underlying internet sites and of their educational sections and secondly to identify the differences between the types of museums and their countries. Finally, the third part of the research attaches importance to the typology of the online educational content and focuses on the strategies put into place in the sites as well as the pedagogy deployed. The internet sites are thus viewed as interconnected elements, intended for a target audience and reinforcing the social role of the museum. The schools and the teaching body are a privileged population; a prominent place for them is underlined
Munos, Rémi. "Apprentissage par renforcement, étude du cas continu." Paris, EHESS, 1997. http://www.theses.fr/1997EHESA021.
Full textSors, Arnaud. "Apprentissage profond pour l'analyse de l'EEG continu." Thesis, Université Grenoble Alpes (ComUE), 2018. http://www.theses.fr/2018GREAS006/document.
Full textThe objective of this research is to explore and develop machine learning methods for the analysis of continuous electroencephalogram (EEG). Continuous EEG is an interesting modality for functional evaluation of cerebral state in the intensive care unit and beyond. Today its clinical use remains more limited that it could be because interpretation is still mostly performed visually by trained experts. In this work we develop automated analysis tools based on deep neural models.The subparts of this work hinge around post-anoxic coma prognostication, chosen as pilot application. A small number of long-duration records were performed and available existing data was gathered from CHU Grenoble. Different components of a semi-supervised architecture that addresses the application are imagined, developed, and validated on surrogate tasks.First, we validate the effectiveness of deep neural networks for EEG analysis from raw samples. For this we choose the supervised task of sleep stage classification from single-channel EEG. We use a convolutional neural network adapted for EEG and we train and evaluate the system on the SHHS (Sleep Heart Health Study) dataset. This constitutes the first neural sleep scoring system at this scale (5000 patients). Classification performance reaches or surpasses the state of the art.In real use for most clinical applications, the main challenge is the lack of (and difficulty of establishing) suitable annotations on patterns or short EEG segments. Available annotations are high-level (for example, clinical outcome) and therefore they are few. We search how to learn compact EEG representations in an unsupervised/semi-supervised manner. The field of unsupervised learning using deep neural networks is still young. To compare to existing work we start with image data and investigate the use of generative adversarial networks (GANs) for unsupervised adversarial representation learning. The quality and stability of different variants are evaluated. We then apply Gradient-penalized Wasserstein GANs on EEG sequences generation. The system is trained on single channel sequences from post-anoxic coma patients and is able to generate realistic synthetic sequences. We also explore and discuss original ideas for learning representations through matching distributions in the output space of representative networks.Finally, multichannel EEG signals have specificities that should be accounted for in characterization architectures. Each EEG sample is an instantaneous mixture of the activities of a number of sources. Based on this statement we propose an analysis system made of a spatial analysis subsystem followed by a temporal analysis subsystem. The spatial analysis subsystem is an extension of source separation methods built with a neural architecture with adaptive recombination weights, i.e. weights that are not learned but depend on features of the input. We show that this architecture learns to perform Independent Component Analysis if it is trained on a measure of non-gaussianity. For temporal analysis, standard (shared) convolutional neural networks applied on separate recomposed channels can be used
Salperwyck, Christophe. "Apprentissage incrémental en ligne sur flux de données." Phd thesis, Université Charles de Gaulle - Lille III, 2012. http://tel.archives-ouvertes.fr/tel-00845655.
Full textOrseau, Laurent. "Imitation algorithmique : Apprentissage Incrémental En-ligne de Séquences." Rennes, INSA, 2007. http://www.theses.fr/2007ISAR0014.
Full textIn continual learning, an agent is continually interacting with its environment. At each time step, it receives inputs, uses a small amount of computations (online) and gives outputs. There is no real definition of a goal to learn, the agent must acquire more and more knowledge, incrementally, and re-use it in more complex tasks. In this framework, we are interested in learning complex sequences, involving recurrence, variables and conditions. But the agent cannot use a large number of trials and error, because of its interaction with the environment. How then can learning be possible from a small number of examples?Traditional methods that are able to solve such complex tasks do not fit in the continual learning framework, because difficulties become harder. To simplify the task, an imitation protocol is used, allowing the agent to learn by seeing a teacher doing, but this respects the continual learning constraints and keeps a high autonomy. Imitation is usually used in a robotic framework, so we extend it to learn more complex sequences~: this is Algorithmic Imitation. A learning system, CSAAL, is then developed and tested on experiments showing that it is indeed able to learn complex sequences within few examples. An extension of this system, H-CSAAL, allows to re-use hierarchically recurrent functions, increasing both the autonomy of the agent and its generalization capacities
Ammarcha, Chawki. "Mélange des poudres en continu : modèles dynamiques et caractérisation des mélanges en ligne." Thesis, Toulouse, INPT, 2010. http://www.theses.fr/2010INPT0138/document.
Full textThe implementation of a continuous mixer in the industry requires detailed studies for a better understanding of this process, with essential aim the development of a control process strategy. The present work reports experimental and modelling results concerning the dynamics of a continuous powder mixer in steady and unsteady states. In particular, we will focus on the transitory phases that are likely to occur : starting, emptying, feeder's feeding, accidental perturbation, etc. We investigate the effect of operating variables, as rotational speed of the stirrer and the inflow rate, on the distribution of particles mass in the mixer and the intermediates flow rates, as well as that of the homogeneity of binary mixtures at the outlet of continuous mixer. A specific experimental protocol, based on image analysis, has been developed for determining mixture quality. The scale of scrutiny can be adjusted and mixture homogeneity can be calculated for this scale in real time. AMarkov chain model is proposed to describe the phenomena observed at both macro-and meso-scales. The model allows to describe the composition of the mixture in different zones of the mixer as well as in the outlet of the vessel, during steady and unsteady regimes and especially at high speed perturbations, whose interest is discussed
Ferreira, Emmanuel. "Apprentissage automatique en ligne pour un dialogue homme-machine situé." Thesis, Avignon, 2015. http://www.theses.fr/2015AVIG0206/document.
Full textA dialogue system should give the machine the ability to interactnaturally and efficiently with humans. In this thesis, we focus on theissue of the development of stochastic dialogue systems. Thus, we especiallyconsider the Partially Observable Markov Decision Process (POMDP)framework which yields state-of-the-art performance on goal-oriented dialoguemanagement tasks. This model enables the system to cope with thecommunication ambiguities due to noisy channel and also to optimize itsdialogue management strategy directly from data with Reinforcement Learning (RL)methods.Considering statistical approaches often requires the availability of alarge amount of training data to reach good performance. However, corpora of interest are seldom readily available and collectingsuch data is both time consuming and expensive. For instance, it mayrequire a working prototype to initiate preliminary experiments with thesupport of expert users or to consider other alternatives such as usersimulation techniques.Very few studies to date have considered learning a dialogue strategyfrom scratch by interacting with real users, yet this solution is ofgreat interest. Indeed, considering the learning process as part of thelife cycle of a system offers a principle framework to dynamically adaptthe system to new conditions in an online and seamless fashion.In this thesis, we endeavour to provide solutions to make possible thisdialogue system cold start (nearly from scratch) but also to improve its ability to adapt to new conditions in operation (domain extension, new user profile, etc.).First, we investigate the conditions under which initial expertknowledge (such as expert rules) can be used to accelerate the policyoptimization of a learning agent. Similarly, we study how polarized userappraisals gathered throughout the course of the interaction can beintegrated into a reinforcement learning-based dialogue manager. Morespecifically, we discuss how this information can be cast intosocially-inspired rewards to speed up the policy optimisation for bothefficient task completion and user adaptation in an online learning setting.The results obtained on a reference task demonstrate that a(quasi-)optimal policy can be learnt in just a few hundred dialogues,but also that the considered additional information is able tosignificantly accelerate the learning as well as improving the noise tolerance.Second, we focus on reducing the development cost of the spoken language understanding module. For this, we exploit recent word embedding models(projection of words in a continuous vector space representing syntacticand semantic properties) to generalize from a limited initial knowledgeabout the dialogue task to enable the machine to instantly understandthe user utterances. We also propose to dynamically enrich thisknowledge with both active learning techniques and state-of-the-artstatistical methods. Our experimental results show that state-of-the-artperformance can be obtained with a very limited amount of in-domain andin-context data. We also show that we are able to refine the proposedmodel by exploiting user returns about the system outputs as well as tooptimize our adaptive learning with an adversarial bandit algorithm tosuccessfully balance the trade-off between user effort and moduleperformance.Finally, we study how the physical embodiment of a dialogue system in a humanoid robot can help the interaction in a dedicated Human-Robotapplication where dialogue system learning and testing are carried outwith real users. Indeed, in this thesis we propose an extension of thepreviously considered decision-making techniques to be able to take intoaccount the robot's awareness of the users' belief (perspective taking)in a RL-based situated dialogue management optimisation procedure
Sidana, Sumit. "Systèmes de recommandation pour la publicité en ligne." Thesis, Université Grenoble Alpes (ComUE), 2018. http://www.theses.fr/2018GREAM061/document.
Full textThis thesis is dedicated to the study of Recommendation Systems for implicit feedback (clicks) mostly using Learning-to-rank and neural network based approaches. In this line, we derive a novel Neural-Network model that jointly learns a new representation of users and items in an embedded space as well as the preference relation of users over the pairs of items and give theoretical analysis. In addition we contribute to the creation of two novel, publicly available, collections for recommendations that record the behavior of customers of European Leaders in eCommerce advertising, Kelkoofootnote{url{https://www.kelkoo.com/}} and Purchfootnote{label{purch}url{http://www.purch.com/}}. Both datasets gather implicit feedback, in form of clicks, of users, along with a rich set of contextual features regarding both customers and offers. Purch's dataset, is affected by popularity bias. Therefore, we propose a simple yet effective strategy on how to overcome the popularity bias introduced while designing an efficient and scalable recommendation algorithm by introducing diversity based on an appropriate representation of items. Further, this collection contains contextual information about offers in form of text. We make use of this textual information in novel time-aware topic models and show the use of topics as contextual information in Factorization Machines that improves performance. In this vein and in conjunction with a detailed description of the datasets, we show the performance of six state-of-the-art recommender models.Keywords. Recommendation Systems, Data Sets, Learning-to-Rank, Neural Network, Popularity Bias, Diverse Recommendations, Contextual information, Topic Model
BOUDOKHANE, CHEDLY. "Automatisation d'une colonne de rectification en continu analyse en ligne par chromatographie en phase gazeuse." Paris 6, 1990. http://www.theses.fr/1990PA066421.
Full textBassagou, Dikagma. "EXOLINE : Dispositif instrumenté pour analyser les interactions en apprentissage collaboratif en ligne." Thesis, Lille 1, 2019. http://www.theses.fr/2019LIL1I074.
Full textThis study concerns a mechanism that could make it possible to analyse students' motivation and autonomy in the absence of human regulation in distance education. Indeed, in the literature, motivation and autonomy have an implication in the drop-out of learners in e-learning situations. To analyze these factors, some researchers use questionnaires on an ad hoc basis and others use digital traces produced by TEL (Technology Enhanced Learning). In most of these studies, the intervention of a tutor to regulate learning activity is important. The objective of this research is to identify students who can follow a teacher from a distance with the best chances of success. To do this, we have designed a system that combines the use of questionnaires and digital learning traces. This system contains a platform called "Exoline" that supports a pedagogical scenario that alternates individual work and collaborative activities in seven steps. Collaborative activities are regulated by a voting system (I like/dislike) of the platform and individual work corresponds to phases of editorial contribution by each learner member of a group. To collect data on collaboration and learning dynamics, we conducted an experiment with 794 students from Kara University with a participation rate of 40.55%. We then applied different statistical methods to the data (questionnaires and traces) from the experiment to identify and study the relationships between learning, motivation and autonomy. Our study highlighted, beyond the initial motivation, the role of the dynamics of maintaining motivation throughout the learning process. The level of progression (number of steps performed) by the learner in our Exoline device has proven to be an interesting indicator of learning performance. In other words, the study shows that most students who drop out do so at the beginning of the course. On a more contextual level, our study also shows how the socioeconomic environment influences the educational path of students, particularly in Togo
Bovo, Angela. "Apprentissage automatique pour l'assistance au suivi d'étudiants en ligne : approches classique et bio-inspirée." Thesis, Toulouse 1, 2014. http://www.theses.fr/2014TOU10035/document.
Full textThis Ph.D. took the shape of a partnership between the VORTEX team in the computer science research laboratory IRIT and the company Andil, which specializes in software for e-learning. This partnership was concluded around a CIFRE Ph.D. This plan is subsidized by the French state through the ANRT. The Ph.D. student, Angela Bovo, worked in Université Toulouse 1 Capitole. Another partnership was built with the training institute Juriscampus, which gave us access to data from real trainings for our experiments. Our main goal for this project was to improve the possibilities for monitoring students in an e-learning training to keep them from falling behind or giving up. We proposed ways to do such monitoring with classical machine learning methods, with the logs from students' activity as data. We also proposed, using the same data, indicators of students' behaviour. With Andil, we designed and produced a web application called GIGA, already marketed and sold, and well appreciated by training managers, which implements our proposals and served as a basis for first clustering experiments which seem to identify well students who are failing or about to give up. Another goal of this project was to study the capacities of the human brain inspired machine learning algorithm Hierarchical Temporal Memory (HTM), in its Cortical Learning Algorithm (CLA) version, because its base hypotheses are well adapted to our problem. We proposed ways to adapt HTM-CLA to classical machine learning functionalities (clustering, classification, regression, prediction), in order to compare its results to those of more classical algorithms; but also to use it as a basis for a behaviour generation engine, which could be used to create an intelligent tutoring system tasked with advising students in real time. However, our implementations did not get to the point of conclusive results
Bouillon, Manuel. "Apprentissage actif en-ligne d'un classifieur évolutif, application à la reconnaissance de commandes gestuelles." Thesis, Rennes, INSA, 2016. http://www.theses.fr/2016ISAR0019/document.
Full textUsing gesture commands is a new way of interacting with touch sensitive interfaces. In order to facilitate user memorization of several commands, it is essential to let the user customize the gestures. This applicative context gives rise to a crosslearning situation, where the user has to memorize the set of commands and the system has to learn and recognize the different gestures. This situation implies several requirements, from the recognizer and from the system that supervizes its learning process. For instance, the recognizer has to be able to learn from few data samples, to keep learning during its use and to follow indefinitely any change of the data now. The supervisor has to optimize the cooperation between the recognizer and the system to minimize user interactions while maximizing recognizer learning. This thesis presents on the one hand the evolving recognition system Evolve oo, that is capable of fast teaming from few data samples, and that follows concept drifts. On the other hand, this thesis also presents the on line active supervisor lntuiSup, that optimizes user-system cooperation when the user is in the training loop, as during customized gesture command use for instance. The evolving classifier Evolve oo is a fuzzy inference system that is fast learning thanks to the generative capacity of rule premises, and at the same time giving high precision thanks to the discriminative capacity of first order rule conclusion. The use of forgetting in the learning process allows to maintain the learning gain indefinitely, enabling class adding at any stage of system learning, and guaranteeing lifelong evolving capacity. The on line active supervisor IntuiSup optimizes user interactions to train a classifier when the user is in the training loop. The proportion of data that is labeled by the user evolves to adapt to problem difficulty and to follow environment evolution (concept drift s). The use of a boosting method optimizes the timing of user interactions to maximize their impact on classifier learning process
Portet, François. "Pilotages d'algorithmes pour la reconnaissance en ligne d'arythmies cardiaques." Rennes 1, 2005. https://tel.archives-ouvertes.fr/tel-00011942v2.
Full textPortet, François Cordier Marie-Odile Carrault Guy. "Pilotages d'algorithmes pour la reconnaissance en ligne d'arythmies cardiaques." [S.l.] : [s.n.], 2005. ftp://ftp.irisa.fr/techreports/theses/2005/portet.pdf.
Full textJeunesse, Christophe. "Collaboration et interculturalité dans la formation en ligne. Contribution à l'écologie de l'apprenance." Thesis, Paris 10, 2009. http://www.theses.fr/2009PA100161/document.
Full textThe research concerns the study of specificities connected to the online collaborative learning in a multicultural context. It is situated at the intersection of the reflections carried out on the conceptual fields dealing with the motivation, with the culture, with the gender and with the distance training mediatized by the educational technologies. The context of this study lies in and online university training gathering 249 European and African students, all French-speaking people, in an adult continuing training and working within remote collaborative learning plan. My reflection was driven by the questioning about the way the students lived the online collaboration in multicultural context, in particular on the difficulties shown by the Africans while at the same time they seemed to present a more positive attitude than their western peers towards this method of training. Several successive investigations (preliminary, quantitative and qualitative among representative samples of the students) make it possible to bring a certain number of answers to the question of research as well as additional details. The culture, in particular the sociotechnical environment of the learners, provides an additional variable well to be taken into account with regard to the gender necessary to decode attitudes and behavior of learners who are involved in a online collaborative training. A reflection around the dimensions of the “learnance” (learning readiness) and the transactional distance also allows to understand better the relations between the actors of the training and the necessary adaptations of the training design in such a context
Yibokou, Kossi Seto. "Apprentissage informel de l'anglais en ligne : quelles conséquences sur la prononciation des étudiants français ?" Thesis, Strasbourg, 2019. https://publication-theses.unistra.fr/restreint/theses_doctorat/2019/Yibokou_Kossi_Seto_2019_ED520.pdf.
Full textThis work is part of the online informal learning of English and explores practices related to various sources of exposure of a sample of students from the University of Strasbourg. The data collected, based on a pronunciation test, a perception test and a survey, show inter- and intra-individual variability inherent to the complexity of the system in which participants evolve. With regards to Received Pronunciation and General American accents, acoustic analyses of pronunciation elements highlight oral productions composed of mixtures of characteristics of the two accents and those of the French language. The perturbation of speech production, implemented through fast speech variation, indicates a resistivity of the system for certain sounds/sequences of sounds. Results also show that television series are the most influential activities among those that promote vocal imitation and allow phonetic-phonological appropriation
Rivera-Santos, Miguel. "Les déterminants de l'apprentissage entre partenaires dans les alliances : Elaboration d'un modèle théorique et étude empirique sur le secteur du commerce en ligne." Jouy-en Josas, HEC, 2003. http://www.theses.fr/2003EHEC0003.
Full textRochd, El Mehdi. "Modèles probabilistes de consommateurs en ligne : personnalisation et recommandation." Thesis, Aix-Marseille, 2015. http://www.theses.fr/2015AIXM4086.
Full textResearch systems have facilitated access to information available on the web using mechanisms for collecting, indexing and storage of heterogeneous content. They generate data resulting from the activity of users on Internet (queries, logfile). The next step is to analyze the data using data mining tools in order to improve the response’s quality of these systems, or to customize the response based on users’ profiles. Some actors, such as the company Marketshot, are positioned as intermediaries between consumers and professionals. Indeed, they link potential buyers with the leading brands and distribution networks through their websites. For such purposes, these intermediaries have developed effective portals, and have stored large volumes of data related to the activity of users on their websites. These data repositories are exploited to respond positively to the needs of users as well as those of professionals who seek to understand the behavior of their customers and anticipate their purchasing actions. My thesis comes within the framework of searching through the data collected from the web. The idea is to build models that explain the correlation between the activities of users on websites of aid for the purchase, and sales trends of products in « real life ». In fact, my research concerns probabilistic learning, in particular Topic Models. It involves modeling the users’ behavior from uses of trader websites
Nissen, Elke. "Apprendre une langue en ligne dans une perspective actionnelle : effets de l'interaction sociale." Phd thesis, Université Louis Pasteur - Strasbourg I, 2003. http://tel.archives-ouvertes.fr/edutice-00001449.
Full textNous menons deux observations contrôlées des personnes se formant par ce dispositif – soit en groupe tutoré, soit en individuel tutoré. L'analyse porte sur un pré- et un post-test, des tâches réalisées durant la phase d'apprentissage, des formulaires auto-administrés et une étude de l'interaction. Les observations ne confirment que partiellement les hypothèses : pendant la phase d'apprentissage, les groupes restreints évoluent différemment, et ce probablement en raison des phénomènes de leadership qui s'y développent. Les apprenants affichent de meilleures performances durant la phase d'apprentissage que lors des tests, mais leurs résultats entre le pré- et le post-test ne s'améliorent pas sensiblement. Une supériorité d'un apprentissage en groupe tutoré par rapport à un apprentissage individuel tutoré ne peut pas être démontrée, du moins pour la tâche et les contextes observés. Plus que la modalité d'apprentissage (en groupe vs. en individuel), le facteur déterminant pour l'apprentissage semble être la présence d'un tuteur.
Yao, Ziwen. "Régulateur adaptatif robuste pour les liaisons de transport a courant continu en haute tension." Vandoeuvre-les-Nancy, INPL, 1993. http://www.theses.fr/1993INPL051N.
Full textGomont, Jacques. "Mise au point d'un procédé de filtration/expression continu applicable à la déshydratation en ligne de suspensions solide-liquide concentrées." Compiègne, 1986. http://www.theses.fr/1986COMPI251.
Full textCaillault, Emilie. "Architecture et Apprentissage d'un Système Hybride Neuro-Markovien pour la Reconnaissance de l'Écriture Manuscrite En-Ligne." Phd thesis, Université de Nantes, 2005. http://tel.archives-ouvertes.fr/tel-00084061.
Full textPoisson, Émilie. "Architecture et apprentissage d'un système hybride neuro-markovien pour la reconnaissance de l'écriture manuscrite en-ligne." Nantes, 2005. http://www.theses.fr/2005NANT2082.
Full textThis thesis deals with the study, the conception, the development and the test of an online unconstrained handwriting word recognition system for an omni-writer application. The proposed system is based on a hybrid architecture including on the one hand, a neural convolutional network (TDNN and/or SDNN), and on the other hand Hidden Markov Models (HMM). The neural network has a global vision and works at the character level, while the HMM works on a more local description and allows the extension from the character level to the word level. The system was first dedicated for processing isolated characters (digits, lowercase letters, uppercase letters). This architecture has been optimized in terms of performances and size. The second part of this work concerns the extension to the word level. In this case, we have defined a global training scheme directly at the word level. It allows to insure the global convergence of the system. It relies on an objective function that combines two main criteria: one based on generative models (typically by maximum likelihood estimation) and the second one based on discriminant criteria (maximum mutual information). Several results are presented on MNIST, IRONOFF and UNIPEN databases. They show the influence of the main parameters of the system, either in terms of topologies, information sources, and training models (number of states, criteria weighting, duration)
Garcin, Claudine. "Pratiques participatives, apprentissage et développement professionnel sur Internet : Le cas de la communauté en ligne "Moodle"." Thesis, Aix-Marseille, 2014. http://www.theses.fr/2014AIXM3007/document.
Full textThe research is based on the framework of Activity Theory and the Social Learning Theory. This thesis addresses the practices of the Internet users who invest work and time in the improvement and the design of "Moodle", the platform for online education.Even if their main objective is not developing their knowledge, their activity requires creating, diffusing and acquiring certain types of knowledge and consequently developing their professional skills. Since information circulates within the virtual communities on the Internet, the selected ethnographic method considers "Moodle" activity as a situated and social activity generating learning process. It is based on both a questionnaire on the practices of the involved people (the Moodlers) and an analysis of the written traces that they produce on the social Web. The outcomes, on the one hand, show how the "Moodlers" manage their activities to learn mainly in a know-how perspective. On the other hand, it appears that the professional development depends on a collective and an interactional dynamics which is not determined by the institutional framework
Moulet, Lucie. "Modélisation de l'apprenant avec une approche par compétences dans le cadre d'environnement d'apprentissage en ligne." Paris 6, 2011. http://www.theses.fr/2011PA066636.
Full textMainsant, Marion. "Apprentissage continu sous divers scénarios d'arrivée de données : vers des applications robustes et éthiques de l'apprentissage profond." Electronic Thesis or Diss., Université Grenoble Alpes, 2023. http://www.theses.fr/2023GRALS045.
Full textThe human brain continuously receives information from external stimuli. It then has the ability to adapt to new knowledge while retaining past events. Nowadays, more and more artificial intelligence algorithms aim to learn knowledge in the same way as a human being. They therefore have to be able to adapt to a large variety of data arriving sequentially and available over a limited period of time. However, when a deep learning algorithm learns new data, the knowledge contained in the neural network overlaps old one and the majority of the past information is lost, a phenomenon referred in the literature as catastrophic forgetting. Numerous methods have been proposed to overcome this issue, but as they were focused on providing the best performance, studies have moved away from real-life applications where algorithms need to adapt to changing environments and perform, no matter the type of data arrival. In addition, most of the best state of the art methods are replay methods which retain a small memory of the past and consequently do not preserve data privacy.In this thesis, we propose to explore data arrival scenarios existing in the literature, with the aim of applying them to facial emotion recognition, which is essential for human-robot interactions. To this end, we present Dream Net - Data-Free, a privacy preserving algorithm, able to adapt to a large number of data arrival scenarios without storing any past samples. After demonstrating the robustness of this algorithm compared to existing state-of-the-art methods on standard computer vision databases (Mnist, Cifar-10, Cifar-100 and Imagenet-100), we show that it can also adapt to more complex facial emotion recognition databases. We then propose to embed the algorithm on a Nvidia Jetson nano card creating a demonstrator able to learn and predict emotions in real-time. Finally, we discuss the relevance of our approach for bias mitigation in artificial intelligence, opening up perspectives towards a more ethical AI
Zimmer, Matthieu. "Apprentissage par renforcement développemental." Thesis, Université de Lorraine, 2018. http://www.theses.fr/2018LORR0008/document.
Full textReinforcement learning allows an agent to learn a behavior that has never been previously defined by humans. The agent discovers the environment and the different consequences of its actions through its interaction: it learns from its own experience, without having pre-established knowledge of the goals or effects of its actions. This thesis tackles how deep learning can help reinforcement learning to handle continuous spaces and environments with many degrees of freedom in order to solve problems closer to reality. Indeed, neural networks have a good scalability and representativeness. They make possible to approximate functions on continuous spaces and allow a developmental approach, because they require little a priori knowledge on the domain. We seek to reduce the amount of necessary interaction of the agent to achieve acceptable behavior. To do so, we proposed the Neural Fitted Actor-Critic framework that defines several data efficient actor-critic algorithms. We examine how the agent can fully exploit the transitions generated by previous behaviors by integrating off-policy data into the proposed framework. Finally, we study how the agent can learn faster by taking advantage of the development of his body, in particular, by proceeding with a gradual increase in the dimensionality of its sensorimotor space
Lefort, Mathieu. "Apprentissage spatial de corrélations multimodales par des mécanismes d'inspiration corticale." Phd thesis, Université Nancy II, 2012. http://tel.archives-ouvertes.fr/tel-00756687.
Full textBouker, Mohamed Ali. "Les communautés d'apprentissage professionnelles en ligne : un moyen de développement professionnel d'acteurs en éducation." Doctoral thesis, Université Laval, 2017. http://hdl.handle.net/20.500.11794/27437.
Full textIn this study, we examined the online professional learning community (PLC) approach used to support the professional development of education stakeholders. Specifically, we sought to identify the instigators and promoters of these online PLCs, the needs expressed regarding training, the means deployed for effective outcomes, and the issues interfering with professional growth within the PLC. We conducted semi-structured interviews with teachers, principals, education consultants, and facilitators who were members of online PLCs (N = 10) and onsite PLCs (N = 39) in two Canadian provinces, namely Québec and New Brunswick. Several questions were asked during the course of this research: Who is responsible for integrating information and communication technologies (ICTs) in the PLCs? What impact do ICTs have in the PLCs and how are they deployed? How is the collaborative process experienced between principals, education consultants, teachers and training instructors in an online PLC? What are the roles of these education stakeholders and how do online PLCs contribute to their professional development? What type of training do online PLC members need? And finally, which challenges and issues experienced by PLC members interfere with proper functioning and performance? Our conceptual framework was inspired by elements from four recognized theoretical models: Wenger’s social learning theory (2005), Huberman’s professional life cycle model for networking teachers (1995), Engeström’s human activity theory (1994), and Daele’s teacher professional development model in online communities (2004). Our results show that teachers and principals were most likely to instigate ICT implementation in PLCs. The development of ICT skills facilitated collaboration between the PLC members. Indeed, the latter expressed both their interest and need for ICT training to continue improving collaboration and communication within their community, and they concurred that integrating technologies would be beneficial to the PLC’s structure and performance. In addition to agreeing that developing computer skills would help their work as a PLC, our study’s respondents also emphasized a need for this training to further their professional development. Our results show that in online PLCs. time factors and having to adapt to new ICTs represented major issues hampering the advancement of these communities. Regarding their work within the PLC, our respondents stated various needs. On one hand were the professional aspects, namely, educational, pedagogical, and didactic needs, and on the other hand were needs relative to their participation in tasks within the PLC, such as keeping informed on certain subjects, being supervised, and benefiting from continuing education throughout their time in the PLC. Finally, our results reveal two significant issues hindering performance outcomes in these PLCs: financial concerns and the lack of available technologies in certain remote area schools.
Chollet, Antoine. "Apprentissage et mobilisation de compétences managériales des joueurs de jeux de rôle en ligne massivement multijoueurs (MMORPG)." Thesis, Montpellier, 2015. http://www.theses.fr/2015MONTD049/document.
Full textMedia relate instances where MMORPG players are being recruited to responsibility positions, thanks to their managerial skills, acquired through playing. Do MMORPG players really develop such skills though playing, and if so, under what conditions? To explore these research questions, we're basing our works on the Social Learning Theory as well as the Social Cognitive Theory, both resulting from of Albert Bandura's researches. Literature reviewing as well as an exploratory qualitative study (13 players and older MMORPG players) led us to propose a managerial skill learning structural model of the MMORPG player. Two analyses were realized.The first one, in an exploratory aim, allowing to refine measuring tools, saw 414 questionnaires being validated (on 707 collected). The second one, with a confirmatory aim, allowing to verify hypothesis, saw 2 628 questionnaires being validated (on 3 690 collected). Once we've drawn the MMORPG player's profile, we're showing that there are managerial skill learning phenomena perceived by the players that are developed then mastered in MMORPG, under specific conditions linked to the game's environment, as well as the internal state of the player. The proposed model is thus validated.Conclusions of this research offer possibilities for players as well as organizations in various domains, such as recruitment or training, by benefiting MMORPG's potential. Longitudinal studies would deserve to be done in order to explore the MMORPG's player learning evolution, and confirm our results
Caraguel, Valérie. "Appropriation des technologies et apprentissage dans un environnement en e-learning : le rôle du tutorat en ligne." Thesis, Aix-Marseille, 2013. http://www.theses.fr/2013AIXM1106/document.
Full textThe purpose of this research is to contribute to a deeper understanding of tutor-learner interactions within the framework of a supportive process of appropriation of technology and learning in the context of e-learning. In addition to e-learning, the literature review focuses on three theoretical fields: learning, appropriation, and e-tutoring. This leads us to question the modes of e-tutoring and their evolution; and the media coverage of the tutor in the learners’ process of learning. A case study, performed at Aix Marseille University, enables us to identify elements of responses to previous questions. The results show that the tutor is seen primarily in a supporting role, and that he may be the social link between the platform and the learners. Beyond these activities, we found that the e-tutors may also have two other e-tutoring functions: knowledge manager and facilitator of peer-tutoring. This allows us ultimately to suggest that a goal of e-learning systems is to set up an e-tutoring system, while leaving the emergence and fostering of self-organization to the learners themselves. Our research also shows that the goal of "making technology transparent" is reached when the acquisition of technology by students is initiated and promoted from the upstream phases of the teaching process. Finally, while there was a concern that e-learning replaces the need for teachers, we found that the role of the latter is reinforced, although still evolving: in other words, the role of the teacher changes; the e-tutor emerges! They become the critical link between electronics and learning, between " e " and " learning "!
Caraguel, Valérie. "Appropriation des technologies et apprentissage dans un environnement en e-learning : le rôle du tutorat en ligne." Electronic Thesis or Diss., Aix-Marseille, 2013. http://www.theses.fr/2013AIXM1106.
Full textThe purpose of this research is to contribute to a deeper understanding of tutor-learner interactions within the framework of a supportive process of appropriation of technology and learning in the context of e-learning. In addition to e-learning, the literature review focuses on three theoretical fields: learning, appropriation, and e-tutoring. This leads us to question the modes of e-tutoring and their evolution; and the media coverage of the tutor in the learners’ process of learning. A case study, performed at Aix Marseille University, enables us to identify elements of responses to previous questions. The results show that the tutor is seen primarily in a supporting role, and that he may be the social link between the platform and the learners. Beyond these activities, we found that the e-tutors may also have two other e-tutoring functions: knowledge manager and facilitator of peer-tutoring. This allows us ultimately to suggest that a goal of e-learning systems is to set up an e-tutoring system, while leaving the emergence and fostering of self-organization to the learners themselves. Our research also shows that the goal of "making technology transparent" is reached when the acquisition of technology by students is initiated and promoted from the upstream phases of the teaching process. Finally, while there was a concern that e-learning replaces the need for teachers, we found that the role of the latter is reinforced, although still evolving: in other words, the role of the teacher changes; the e-tutor emerges! They become the critical link between electronics and learning, between " e " and " learning "!
Belley, Sophie. "Conception de systèmes d'analyse et de contrôle en ligne pour une unité de production de solution d'enrobage en continu pour comprimés pharmaceutiques." Mémoire, Université de Sherbrooke, 2014. http://hdl.handle.net/11143/5941.
Full textDuneau, Laurent. "Etude et réalisation d'un système adaptatif pour la reconnaissance en ligne de mots manuscrits." Compiègne, 1994. http://www.theses.fr/1994COMP7665.
Full textZimmer, Matthieu. "Apprentissage par renforcement développemental." Electronic Thesis or Diss., Université de Lorraine, 2018. http://www.theses.fr/2018LORR0008.
Full textReinforcement learning allows an agent to learn a behavior that has never been previously defined by humans. The agent discovers the environment and the different consequences of its actions through its interaction: it learns from its own experience, without having pre-established knowledge of the goals or effects of its actions. This thesis tackles how deep learning can help reinforcement learning to handle continuous spaces and environments with many degrees of freedom in order to solve problems closer to reality. Indeed, neural networks have a good scalability and representativeness. They make possible to approximate functions on continuous spaces and allow a developmental approach, because they require little a priori knowledge on the domain. We seek to reduce the amount of necessary interaction of the agent to achieve acceptable behavior. To do so, we proposed the Neural Fitted Actor-Critic framework that defines several data efficient actor-critic algorithms. We examine how the agent can fully exploit the transitions generated by previous behaviors by integrating off-policy data into the proposed framework. Finally, we study how the agent can learn faster by taking advantage of the development of his body, in particular, by proceeding with a gradual increase in the dimensionality of its sensorimotor space
Portet, François. "Pilotage d'algorithmes pour la reconnaissance en ligne d'arythmies cardiaques." Phd thesis, Université Rennes 1, 2005. http://tel.archives-ouvertes.fr/tel-00011942.
Full textMarcastel, Alexandre. "Allocation de puissance en ligne dans un réseau IoT dynamique et non-prédictible." Thesis, Cergy-Pontoise, 2019. http://www.theses.fr/2019CERG0995/document.
Full textOne of the key challenges in Internet of Things (IoT) networks is to connect numerous, heterogeneous andautonomous devices. These devices have different types of characteristics in terms of: application, computational power, connectivity, mobility or power consumption. These characteristics give rise to challenges concerning resource allocation such as: a) these devices operate in a highly dynamic and unpredictable environments; b) the lack of sufficient information at the device end; c) the interference control due to the large number of devices in the network. The fact that the network is highly dynamic and unpredictable implies that existing solutions for resource allocation are no longer relevant because classical solutions require a perfect or statistical knowledge of the network. To address these issues, we use tools from online optimization and machine learning. In the online optimization framework, the device only needs to have strictly causal information to define its online policy. In order to evaluate the performance of a given online policy, the most commonly used notion is that of the regret, which compares its performance in terms of loss with a benchmark policy, i.e., the best fixed strategy computed in hindsight. Otherwise stated, the regret measures the performance gap between an online policy and the best mean optimal solution over a fixed horizon. In this thesis, we focus on an online power minimization problem under rate constraints in a dynamic IoT network. To address this issue, we propose a regret-based formulation that accounts for arbitrary network dynamics, using techniques used to solve the multi-armed bandit problem. This allows us to derive an online power allocation policy which is provably capable of adapting to such changes, while relying solely on strictly causal feedback. In so doing, we identify an important tradeoff between the amount of feedback available at the transmitter side and the resulting system performance. We first study the case in which the device has access to a vector, either the gradient or an unbiased estimated of the gradient, as information feedback. To limit the feedback exchange in the network our goal is to reduce it as mush as possible. Therefore, we study the case in which the device has access to only a loss-based information (scalar feedback). In this case, we propose a second online algorithm to determine an efficient and adaptative power allocation policy
Kers, Hagberg Anna. "À la recherche des facteurs de réussite au lycée : Apprentissage de langues en ligne au lycée - une étude." Thesis, Högskolan Dalarna, Franska, 2014. http://urn.kb.se/resolve?urn=urn:nbn:se:du-13851.
Full textBaudoin, Emmanuel. "Facteurs de suivi et apprentissages individuels des salariés dans des parcours e-learning : quatre études de cas chez un constructeur automobile." Paris 9, 2010. https://portail.bu.dauphine.fr/fileviewer/index.php?doc=2010PA090057.
Full textDarwiche, Domingues Omar. "Exploration en apprentissage par renforcement : au-delà des espaces d'états finis." Thesis, Université de Lille (2022-....), 2022. http://www.theses.fr/2022ULILB002.
Full textReinforcement learning (RL) is a powerful machine learning framework to design algorithms that learn to make decisions and to interact with the world. Algorithms for RL can be classified as offline or online. In the offline case, the algorithm is given a fixed dataset, based on which it needs to compute a good decision-making strategy. In the online case, an agent needs to efficiently collect data by itself, by interacting with the environment: that is the problem of exploration in reinforcement learning. This thesis presents theoretical and practical contributions to online RL. We investigate the worst-case performance of online RL algorithms in finite environments, that is, those that can be modeled with a finite amount of states, and where the set of actions that can be taken by an agent is also finite. Such performance degrades as the number of states increases, whereas in real-world applications the state set can be arbitrarily large or continuous. To tackle this issue, we propose kernel-based algorithms for exploration that can be implemented for general state spaces, and for which we provide theoretical results under weak assumptions on the environment. Those algorithms rely on a kernel function that measures the similarity between different states, which can be defined on arbitrary state-spaces, including discrete sets and Euclidean spaces, for instance. Additionally, we show that our kernel-based algorithms are able to handle non-stationary environments by using time-dependent kernel functions, and we propose and analyze approximate versions of our methods to reduce their computational complexity. Finally, we introduce a scalable approximation of our kernel-based methods, that can be implemented with deep reinforcement learning and integrate different representation learning methods to define a kernel function
Zaninotti, Marion. "Planification en ligne de la stratégie de navigation pour un drone autonome en environnement urbain." Electronic Thesis or Diss., Toulouse, ISAE, 2024. http://www.theses.fr/2024ESAE0063.
Full textUAVs can now be used for various applications, including service robotics, exploration and monitoring of environments, precision agriculture, as well as search and rescue missions. The need for autonomous navigation for UAVs is therefore becoming increasingly important. Many UAVs use GPS (Global Positioning System) for localization. However, in urban environments, the measured position can be inaccurate or even unavailable, which can compromise mission safety.In this context, the problem of efficient and safe navigation for an autonomous UAV, under uncertain GNSS availability, has been modeled as a POMDP (Partially Observable Markov Decision Process). Nevertheless, planning in such a complex model suffers from high computational cost and yields insufficient results under real-time constraints.Recently, research has focused on integrating offline learning to guide online planning. Inspired by the state-of-the-art CAMP (Context-specific Abstract Markov Decision Process) modeling, we propose a method that involves learning a constraint to be imposed during online planning for this problem. Imposing this constraint allows for an abstraction of the state space by restricting the UAV's navigation to a corridor within the environment.We then generalize this method to all SSP (Stochastic Shortest Path) problems with dead ends. The weight assigned to safety versus path efficiency is learned, a global path is planned based on this weight, and the constraint is derived from this global path. The resolution thus relies on a hybrid approach combining global path planning and online path planning.Afterward, we apply this generalized method to the FrozenLake problem, where an agent seeks a path across a frozen lake to reach a goal while avoiding holes, and to the initial UAV navigation problem. The results of all the experiments demonstrate that using such a method can improve the quality of solutions obtained through online planning, particularly for complex navigation environments and missions
Mckim, Kerrie. "Sites Internet : approches par les tâches et apprentissage du lexique en langue étrangère." Grenoble, 2010. http://www.theses.fr/2010GRENL001.
Full textThis dissertation lies within the fields of Second Language Acquisition and Computer-Assisted Language Learning integrating a task-based learning approach while focusing on lexical acquisition. This study was developed following the observation that websites are frequently used in the foreign language class yet little research has been done to examine the impact that the use of l websites has on the students’ acquisition of the language. The purpose of this research is to examine the impact of the websites on lexical learning in the foreign language classroom. The participants in this study are university students enrolled in a conversational language class, Their language level range from advanced beginners to advanced intermediate learners. Websites are integrated into the course curriculum through the use of Blackboard (a course management system) and studied outside of class to prepare for in class communicative tasks. The study presents a description of the communicative tasks carried out by the students as well as the Internet websites used in the execution of the communicative tasks. Comparisons are then made between the language produced by the learners during the tasks and the lexical items present on the websites
Al, Hajj Mohamad Rami. "Reconnaissance hors ligne de mots manuscrits cursifs par l'utilisation de systèmes hybrides et de techniques d'apprentissage automatique." Paris, ENST, 2007. http://www.theses.fr/2007ENST0020.
Full textThe automatic offline recognition of handwritten words improves human-machine interaction. It is already used in many business office applications dealing with the automatic processing of documents such as automatic post sorting, and the verification and recognition of bank check amounts. The off line recognition of cursive handwritten words remains an open problem due to difficulties such as :handwriting normalization, word segmentation into compound components and the modeling of these components. The main objective of this thesis, is to propose, design, and implement a system for the automatic offline recognition of Arabic handwritten words. The proposed approach is analytical without explicit segmentation of words into compound characters, and it is based on the stochastic HMM approach (Hidden Markov models). The method is composed of two stages : a recognition stage based on different features, and a combination stage of three HMM-based classifiers. Each individual HMM classifier uses a sliding window with a specific inclination. Different combining strategies are tested, among them the Sum rule, the Majority Vote rule and the Borda Count rule. The best combination strategy consists of using a neural network-based combining classifier. The combination of these classifiers can better cope with the writing inclination, the erroneous positions of diacritical marks and points, and the overlapping of consecutive characters in handwritten words. The reference system based on the proposed method has shown best performance at the competition organized at ICDAR 2005, where a set of state-of art systems were compared and tested on the IFN/ENIT benchmark database
Péret, Laurent. "Recherche en ligne pour les Processus Décisionnels de Markov : application à la maintenance d'une constellation de satellites." Phd thesis, Toulouse, INPT, 2004. https://hal.science/tel-04603802.
Full textCruz, José Marcio Martins da. "Contribution au classement statistique mutualisé de messages électroniques (spam)." Paris, ENMP, 2011. https://pastel.archives-ouvertes.fr/pastel-00637173.
Full textSince the 90's, different machine learning methods were investigated and applied to the email classification problem (spam filtering), with very good but not perfect results. It was always considered that these methods are well adapted to filter messages to a single user and not filter to messages of a large set of users, like a community. Our approach was, at first, look for a better understanding of handled data, with the help of a corpus of real messages, before studying new algorithms. With the help of a logistic regression classifier with online active learning, we could show, empirically, that with a simple classification algorithm coupled with a learning strategy well adapted to the real context it's possible to get results which are as good as those we can get with more complex algorithms. We also show, empirically, with the help of messages from a small group of users, that the efficiency loss is not very high when the classifier is shared by a group of users