Rozprawy doktorskie na temat „Compression de réseaux de neurones”
Utwórz poprawne odniesienie w stylach APA, MLA, Chicago, Harvard i wielu innych
Sprawdź 50 najlepszych rozpraw doktorskich naukowych na temat „Compression de réseaux de neurones”.
Przycisk „Dodaj do bibliografii” jest dostępny obok każdej pracy w bibliografii. Użyj go – a my automatycznie utworzymy odniesienie bibliograficzne do wybranej pracy w stylu cytowania, którego potrzebujesz: APA, MLA, Harvard, Chicago, Vancouver itp.
Możesz również pobrać pełny tekst publikacji naukowej w formacie „.pdf” i przeczytać adnotację do pracy online, jeśli odpowiednie parametry są dostępne w metadanych.
Przeglądaj rozprawy doktorskie z różnych dziedzin i twórz odpowiednie bibliografie.
Fernandez, Brillet Lucas. "Réseaux de neurones CNN pour la vision embarquée". Thesis, Université Grenoble Alpes, 2020. http://www.theses.fr/2020GRALM043.
Pełny tekst źródłaRecently, Convolutional Neural Networks have become the state-of-the-art soluion(SOA) to most computer vision problems. In order to achieve high accuracy rates, CNNs require a high parameter count, as well as a high number of operations. This greatly complicates the deployment of such solutions in embedded systems, which strive to reduce memory size. Indeed, while most embedded systems are typically in the range of a few KBytes of memory, CNN models from the SOA usually account for multiple MBytes, or even GBytes in model size. Throughout this thesis, multiple novel ideas allowing to ease this issue are proposed. This requires to jointly design the solution across three main axes: Application, Algorithm and Hardware.In this manuscript, the main levers allowing to tailor computational complexity of a generic CNN-based object detector are identified and studied. Since object detection requires scanning every possible location and scale across an image through a fixed-input CNN classifier, the number of operations quickly grows for high-resolution images. In order to perform object detection in an efficient way, the detection process is divided into two stages. The first stage involves a region proposal network which allows to trade-off recall for the number of operations required to perform the search, as well as the number of regions passed on to the next stage. Techniques such as bounding box regression also greatly help reduce the dimension of the search space. This in turn simplifies the second stage, since it allows to reduce the task’s complexity to the set of possible proposals. Therefore, parameter counts can greatly be reduced.Furthermore, CNNs also exhibit properties that confirm their over-dimensionment. This over-dimensionement is one of the key success factors of CNNs in practice, since it eases the optimization process by allowing a large set of equivalent solutions. However, this also greatly increases computational complexity, and therefore complicates deploying the inference stage of these algorithms on embedded systems. In order to ease this problem, we propose a CNN compression method which is based on Principal Component Analysis (PCA). PCA allows to find, for each layer of the network independently, a new representation of the set of learned filters by expressing them in a more appropriate PCA basis. This PCA basis is hierarchical, meaning that basis terms are ordered by importance, and by removing the least important basis terms, it is possible to optimally trade-off approximation error for parameter count. Through this method, it is possible to compress, for example, a ResNet-32 network by a factor of ×2 both in the number of parameters and operations with a loss of accuracy <2%. It is also shown that the proposed method is compatible with other SOA methods which exploit other CNN properties in order to reduce computational complexity, mainly pruning, winograd and quantization. Through this method, we have been able to reduce the size of a ResNet-110 from 6.88Mbytes to 370kbytes, i.e. a x19 memory gain with a 3.9 % accuracy loss.All this knowledge, is applied in order to achieve an efficient CNN-based solution for a consumer face detection scenario. The proposed solution consists of just 29.3kBytes model size. This is x65 smaller than other SOA CNN face detectors, while providing equal detection performance and lower number of operations. Our face detector is also compared to a more traditional Viola-Jones face detector, exhibiting approximately an order of magnitude faster computation, as well as the ability to scale to higher detection rates by slightly increasing computational complexity.Both networks are finally implemented in a custom embedded multiprocessor, verifying that theorical and measured gains from PCA are consistent. Furthermore, parallelizing the PCA compressed network over 8 PEs achieves a x11.68 speed-up with respect to the original network running on a single PE
Hubens, Nathan. "Towards lighter and faster deep neural networks with parameter pruning". Electronic Thesis or Diss., Institut polytechnique de Paris, 2022. http://www.theses.fr/2022IPPAS025.
Pełny tekst źródłaSince their resurgence in 2012, Deep Neural Networks have become ubiquitous in most disciplines of Artificial Intelligence, such as image recognition, speech processing, and Natural Language Processing. However, over the last few years, neural networks have grown exponentially deeper, involving more and more parameters. Nowadays, it is not unusual to encounter architectures involving several billions of parameters, while they mostly contained thousands less than ten years ago.This generalized increase in the number of parameters makes such large models compute-intensive and essentially energy inefficient. This makes deployed models costly to maintain but also their use in resource-constrained environments very challenging.For these reasons, much research has been conducted to provide techniques reducing the amount of storage and computing required by neural networks. Among those techniques, neural network pruning, consisting in creating sparsely connected models, has been recently at the forefront of research. However, although pruning is a prevalent compression technique, there is currently no standard way of implementing or evaluating novel pruning techniques, making the comparison with previous research challenging.Our first contribution thus concerns a novel description of pruning techniques, developed according to four axes, and allowing us to unequivocally and completely define currently existing pruning techniques. Those components are: the granularity, the context, the criteria, and the schedule. Defining the pruning problem according to those components allows us to subdivide the problem into four mostly independent subproblems and also to better determine potential research lines.Moreover, pruning methods are still in an early development stage, and primarily designed for the research community. Indeed, most pruning works are usually implemented in a self-contained and sophisticated way, making it troublesome for non-researchers to apply such techniques without having to learn all the intricacies of the field. To fill this gap, we proposed FasterAI toolbox, intended to be helpful to researchers, eager to create and experiment with different compression techniques, but also to newcomers, that desire to compress their neural network for concrete applications. In particular, the sparsification capabilities of FasterAI have been built according to the previously defined pruning components, allowing for a seamless mapping between research ideas and their implementation.We then propose four theoretical contributions, each one aiming at providing new insights and improving on state-of-the-art methods in each of the four identified description axes. Also, those contributions have been realized by using the previously developed toolbox, thus validating its scientific utility.Finally, to validate the applicative character of the pruning technique, we have selected a use case: the detection of facial manipulation, also called DeepFakes Detection. The goal is to demonstrate that the developed tool, as well as the different proposed scientific contributions, can be applicable to a complex and actual problem. This last contribution is accompanied by a proof-of-concept application, providing DeepFake detection capabilities in a web-based environment, thus allowing anyone to perform detection on an image or video of their choice.This Deep Learning era has emerged thanks to the considerable improvements in high-performance hardware and access to a large amount of data. However, since the decline of Moore's Law, experts are suggesting that we might observe a shift in how we conceptualize the hardware, by going from task-agnostic to domain-specialized computations, thus leading to a new era of collaboration between software, hardware, and machine learning communities. This new quest for more efficiency will thus undeniably go through neural network compression techniques, and particularly sparse computations
Dupont, Robin. "Deep Neural Network Compression for Visual Recognition". Electronic Thesis or Diss., Sorbonne université, 2023. http://www.theses.fr/2023SORUS565.
Pełny tekst źródłaThanks to the miniaturisation of electronics, embedded devices have become ubiquitous since the 2010s, performing various tasks around us. As their usage expands, there's an increasing demand for efficient data processing and decision-making. Deep neural networks are apt tools for this, but they are often too large and intricate for embedded systems. Therefore, methods to compress these networks without affecting their performance are crucial. This PhD thesis introduces two methods focused on pruning to compress networks, maintaining accuracy. The thesis first details a budget-aware method for compressing large neural networks using weight reparametrisation and a budget loss, eliminating the need for fine-tuning. Traditional pruning methods often use post-training indicators to cut weights, ignoring desired pruning rates. Our method incorporates a budget loss, directing pruning during training, enabling simultaneous topology and weight optimisation. By soft-pruning smaller weights via reparametrisation, we reduce accuracy loss compared to standard pruning. We validate our method on several datasets and architectures. Later, the thesis examines extracting efficient subnetworks without weight training. We aim to discern the optimal subnetwork topology within a large network, bypassing weight optimisation yet ensuring strong performance. This is realized with our Arbitrarily Shifted Log Parametrisation, a differentiable method for discrete topology sampling, facilitating masks' training to denote weight selection probability. Additionally, a weight recalibration technique, Smart Rescale, is presented. It boosts extracted subnetworks' performance and hastens their training. Our method identifies the best pruning rate in a single training cycle, averting exhaustive hyperparameter searches and various rate training. Through extensive tests, our technique consistently surpasses similar state-of-the-art methods, creating streamlined networks that achieve high sparsity without notable accuracy drops
Yvinec, Edouard. "Efficient Neural Networks : Post Training Pruning and Quantization". Electronic Thesis or Diss., Sorbonne université, 2023. http://www.theses.fr/2023SORUS581.
Pełny tekst źródłaDeep neural networks have grown to be the most widely adopted models to solve most computer vision and natural language processing tasks. Since the renewed interest, sparked in 2012, for these architectures, in machine learning, their size in terms of memory footprint and computational costs have increased tremendously, which has hindered their deployment. In particular, with the rising interest for generative ai such as large language models and diffusion models, this phenomenon has recently reached new heights, as these models can weight several billions of parameters and require multiple high-end gpus in order to infer in real-time. In response, the deep learning community has researched for methods to compress and accelerate these models. These methods are: efficient architecture design, tensor decomposition, pruning and quantization. In this manuscript, I paint a landscape of the current state-of-the art in deep neural networks compression and acceleration as well as my contributions to the field. First, I propose a general introduction to the aforementioned techniques and highlight their shortcomings and current challenges. Second, I provide a detailed discussion regarding my contributions to the field of deep neural networks pruning. These contributions led to the publication of three articles: RED, RED++ and SInGE. In RED and RED++, I introduced a novel way to perform data-free pruning and tensor decomposition based on redundancy reduction. On the flip side, in SInGE, I proposed a new importance-based criterion for data-driven pruning. This criterion was inspired by attribution techniques which consist in ranking inputs by their relative importance with respect to the final prediction. In SInGE, I adapted one of the most effective attribution technique to weight importance ranking for pruning. In the third chapter, I layout my contributions to the field of deep quantization: SPIQ, PowerQuant, REx, NUPES, and a best practice paper. Each of these methods address one of the previous limitations of post-training quantization. In SPIQ, PowerQuant and REx, I provide a solution to the granularity limitations of quantization, a novel non-uniform format which is particularly effective on transformer architectures and a technique for quantization decomposition which eliminates the need for unsupported bit-widths, respectively. In the two remaining articles, I provide significant improvements over existing gradient-based post-training quantization techniques, bridging the gap between such techniques and non-uniform quantization. In the last chapter, I propose a set of leads for future work which I believe to be the, current, most important unanswered questions in the field
Resmerita, Diana. "Compression pour l'apprentissage en profondeur". Thesis, Université Côte d'Azur, 2022. http://www.theses.fr/2022COAZ4043.
Pełny tekst źródłaAutonomous cars are complex applications that need powerful hardware machines to be able to function properly. Tasks such as staying between the white lines, reading signs, or avoiding obstacles are solved by using convolutional neural networks (CNNs) to classify or detect objects. It is highly important that all the networks work in parallel in order to transmit all the necessary information and take a common decision. Nowadays, as the networks improve, they also have become bigger and more computational expensive. Deploying even one network becomes challenging. Compressing the networks can solve this issue. Therefore, the first objective of this thesis is to find deep compression methods in order to cope with the memory and computational power limitations present on embedded systems. The compression methods need to be adapted to a specific processor, Kalray's MPPA, for short term implementations. Our contributions mainly focus on compressing the network post-training for storage purposes, which means compressing the parameters of the network without retraining or changing the original architecture and the type of the computations. In the context of our work, we decided to focus on quantization. Our first contribution consists in comparing the performances of uniform quantization and non-uniform quantization, in order to identify which of the two has a better rate-distortion trade-off and could be quickly supported in the company. The company's interest is also directed towards finding new innovative methods for future MPPA generations. Therefore, our second contribution focuses on comparing standard floating-point representations (FP32, FP16) to recently proposed alternative arithmetical representations such as BFloat16, msfp8, Posit8. The results of this analysis were in favor for Posit8. This motivated the company Kalray to conceive a decompressor from FP16 to Posit8. Finally, since many compression methods already exist, we decided to move to an adjacent topic which aims to quantify theoretically the effects of quantization error on the network's accuracy. This is the second objective of the thesis. We notice that well-known distortion measures are not adapted to predict accuracy degradation in the case of inference for compressed neural networks. We define a new distortion measure with a closed form which looks like a signal-to-noise ratio. A set of experiments were done using simulated data and small networks, which show the potential of this distortion measure
Cherkashyn, Valeriy. "Représentation adaptative d'images de télédétection à très haute résolution spatiale une nouvelle approche hybride (la décomposition pyramidale avec des réseaux de neurones)". Thèse, Université de Sherbrooke, 2011. http://hdl.handle.net/11143/5831.
Pełny tekst źródłaMitrica, Iulia. "Video compression of airplane cockpit screens content". Electronic Thesis or Diss., Institut polytechnique de Paris, 2021. http://www.theses.fr/2021IPPAT042.
Pełny tekst źródłaThis thesis addresses the problem of encoding the video of airplane cockpits.The cockpit of modern airliners consists in one or more screens displaying the status of the plane instruments (e.g., the plane location as reported by the GPS, the fuel level as read by the sensors in the tanks, etc.,) often superimposed over natural images (e.g., navigation maps, outdoor cameras, etc.).Plane sensors are usually inaccessible due to security reasons, so recording the cockpit is often the only way to log vital plane data in the event of, e.g., an accident.Constraints on the recording storage available on-board require the cockpit video to be coded at low to very low bitrates, whereas safety reasons require the textual information to remain intelligible after decoding. In addition, constraints on the power envelope of avionic devices limit the cockpit recording subsystem complexity.Over the years, a number of schemes for coding images or videos with mixed computer-generated and natural contents have been proposed. Text and other computer generated graphics yield high-frequency components in the transformed domain. Therefore, the loss due to compression may hinder the readability of the video and thus its usefulness. For example, the recently standardized Screen Content Coding (SCC) extension of the H.265/HEVC standard includes tools designed explicitly for screen contents compression. Our experiments show however that artifacts persist at the low bitrates targeted by our application, prompting for schemes where the video is not encoded in the pixel domain.This thesis proposes methods for low complexity screen coding where text and graphical primitives are encoded in terms of their semantics rather than as blocks of pixels.At the encoder side, characters are detected and read using a convolutional neural network.Detected characters are then removed from screen via pixel inpainting, yielding a smoother residual video with fewer high frequencies. The residual video is encoded with a standard video codec and is transmitted to the receiver side together with text and graphics semantics as side information.At the decoder side, text and graphics are synthesized using the decoded semantics and superimposed over the residual video, eventually recovering the original frame. Our experiments show that an AVC/H.264 encoder retrofitted with our method has better rate-distortion performance than H.265/HEVC and approaches that of its SCC extension.If the complexity constraints allow inter-frame prediction, we also exploit the fact that co-located characters in neighbor frames are strongly correlated.Namely, the misclassified symbols are recovered using a proposed method based on low-complexity model of transitional probabilities for characters and graphics. Concerning character recognition, the error rate drops up to 18 times in the easiest cases and at least 1.5 times in the most difficult sequences despite complex occlusions.By exploiting temporal redundancy, our scheme further improves in rate-distortion terms and enables quasi-errorless character decoding. Experiments with real cockpit video footage show large rate-distortion gains for the proposed method with respect to video compression standards
Verma, Sagar. "Deep Neural Network Modeling of Electric Motors". Electronic Thesis or Diss., université Paris-Saclay, 2023. http://www.theses.fr/2023UPAST088.
Pełny tekst źródłaThis thesis deals with the application of neural networks in solving electrical motor problems. Chapter 2 contributes to identifying a neural network that can learn the multivariate relationship between different electrical motor signals.The identified network is then used for speed-torque estimation from currents and voltages. Chapter 3 focuses on detecting and recovering from faulty measurements. Our method encompasses electrical sensor faults, mechanical faults, and temperature estimation.Chapter 4 then discusses the reliability of the speed-torque estimator in case of noisy currents and voltages. We presenta denoising method which allows our speed- torque estimator to be applicable in a realistic context. This is followed by an analysis of the adversarial robustness of the neural networks used in electrical motor tasks. The generalization capability of the speed-torque estimator is also briefly considered. In Chapter 5, we focus on the final roadblock in achieving real-world application of neural networks: computational requirements. We present the Subdifferential Inclusion for Sparsity (SIS) method to find the best sparse network from pretrained weights while maintaining original accuracy
Burel, Gilles. "RESEAUX DE NEURONES EN TRAITEMENT D'IMAGES - Des Modèles théoriques aux Applications Industrielles -". Phd thesis, Université de Bretagne occidentale - Brest, 1991. http://tel.archives-ouvertes.fr/tel-00101699.
Pełny tekst źródłatraitement du signal et de l'image. On se place d'emblée du point de vue de
l'industriel impliqué dans la recherche, c'est à dire que l'on s'intéresse à
des problèmes réalistes, sans pour autant négliger la recherche
théorique.
Dans une première partie, nous montrons
l'intérêt des réseaux de neurones comme source d'inspiration pour la
conception de nouveaux algorithmes. Nous proposons en particulier une
structure originale pour la prédiction, ainsi que de nouveaux algorithmes de
Quantification Vectorielle. Les propriétés des algorithmes existants sont
également éclaircies du point de vue théorique, et des méthodes de réglage
automatique de leurs paramètres sont proposées.
On montre ensuite les capacités des réseaux de neurones à traiter un vaste champ
d'applications d'intérêt industriel. Pour divers problèmes de traitement de
l'image et du signal (de la segmentation à la séparation de sources, en
passant par la reconnaissance de formes et la compression de données), on
montre qu'il est possible de développer à moindre coût une solution neuronale
efficace.
Boukli, Hacene Ghouthi. "Processing and learning deep neural networks on chip". Thesis, Ecole nationale supérieure Mines-Télécom Atlantique Bretagne Pays de la Loire, 2019. http://www.theses.fr/2019IMTA0153/document.
Pełny tekst źródłaIn the field of machine learning, deep neural networks have become the inescapablereference for a very large number of problems. These systems are made of an assembly of layers,performing elementary operations, and using a large number of tunable variables. Using dataavailable during a learning phase, these variables are adjusted such that the neural networkaddresses the given task. It is then possible to process new data.To achieve state-of-the-art performance, in many cases these methods rely on a very largenumber of parameters, and thus large memory and computational costs. Therefore, they are oftennot very adapted to a hardware implementation on constrained resources systems. Moreover, thelearning process requires to reuse the training data several times, making it difficult to adapt toscenarios where new information appears on the fly.In this thesis, we are first interested in methods allowing to reduce the impact of computations andmemory required by deep neural networks. Secondly, we propose techniques for learning on thefly, in an embedded context
Carvalho, Micael. "Deep representation spaces". Electronic Thesis or Diss., Sorbonne université, 2018. http://www.theses.fr/2018SORUS292.
Pełny tekst źródłaIn recent years, Deep Learning techniques have swept the state-of-the-art of many applications of Machine Learning, becoming the new standard approach for them. The architectures issued from these techniques have been used for transfer learning, which extended the power of deep models to tasks that did not have enough data to fully train them from scratch. This thesis' subject of study is the representation spaces created by deep architectures. First, we study properties inherent to them, with particular interest in dimensionality redundancy and precision of their features. Our findings reveal a strong degree of robustness, pointing the path to simple and powerful compression schemes. Then, we focus on refining these representations. We choose to adopt a cross-modal multi-task problem, and design a loss function capable of taking advantage of data coming from multiple modalities, while also taking into account different tasks associated to the same dataset. In order to correctly balance these losses, we also we develop a new sampling scheme that only takes into account examples contributing to the learning phase, i.e. those having a positive loss. Finally, we test our approach in a large-scale dataset of cooking recipes and associated pictures. Our method achieves a 5-fold improvement over the state-of-the-art, and we show that the multi-task aspect of our approach promotes a semantically meaningful organization of the representation space, allowing it to perform subtasks never seen during training, like ingredient exclusion and selection. The results we present in this thesis open many possibilities, including feature compression for remote applications, robust multi-modal and multi-task learning, and feature space refinement. For the cooking application, in particular, many of our findings are directly applicable in a real-world context, especially for the detection of allergens, finding alternative recipes due to dietary restrictions, and menu planning
Dugas, Alexandre. "Architecture de transformée de cosinus discrète sur deux dimensions sans multiplication et mémoire de transposition". Mémoire, Université de Sherbrooke, 2012. http://hdl.handle.net/11143/6174.
Pełny tekst źródłaMorozkin, Pavel. "Design and implementation of image processing and compression algorithms for a miniature embedded eye tracking system". Electronic Thesis or Diss., Sorbonne université, 2018. https://accesdistant.sorbonne-universite.fr/login?url=https://theses-intra.sorbonne-universite.fr/2018SORUS435.pdf.
Pełny tekst źródłaHuman-Machine Interaction (HMI) progressively becomes a part of coming future. Being an example of HMI, embedded eye tracking systems allow user to interact with objects placed in a known environment by using natural eye movements. The EyeDee™ portable eye tracking solution (developed by SuriCog) is an example of an HMI-based product, which includes Weetsy™ portable wire/wireless system (including Weetsy™ frame and Weetsy™ board), π-Box™ remote smart sensor and PC-based processing unit running SuriDev eye/head tracking and gaze estimation software, delivering its result in real time to a client’s application through SuriSDK (Software Development Kit). Due to wearable form factor developed eye-tracking system must conform to certain constraints, where the most important are low power consumption, low heat generation low electromagnetic radiation, low MIPS (Million Instructions per Second), as well as support wireless eye data transmission and be space efficient in general. Eye image acquisition, finding of the eye pupil ROI (Region Of Interest), compression of ROI and its wireless transmission in compressed form over a medium are very beginning steps of the entire eye tracking algorithm targeted on finding coordinates of human eye pupil. Therefore, it is necessary to reach the highest performance possible at each step in the entire chain. In contrast with state-of-the-art general-purpose image compression systems, it is possible to construct an entire new eye tracking application-specific image processing and compression methods, approaches and algorithms, design and implementation of which are the goal of this thesis
Mahé, Pierre. "Codage ambisonique pour les communications immersives". Thesis, La Rochelle, 2022. http://www.theses.fr/2022LAROS011.
Pełny tekst źródłaThis thesis takes place in the context of the spread of immersive content. For the last couple of years, immersive audio recording and playback technologies have gained momentum and have become more and more popular. New codecs are needed to handle those spatial audio formats, especially for communication applications. There are several ways to represent spatial audio scenes. In this thesis, we focused on First Order Ambisonic. The first part of our research focused on improving multi-monocoding by decorrelated each ambisonic signal component before the multi-mono coding. To guarantee signal continuity between frames, efficient quantization new mechanisms are proposed. In the second part of this thesis, we proposed a new coding concept using a power map to recreate the original spatial image. With this concept, we proposed two compressing methods. The first one is a post-processing focused on limiting the spatial distortion of the decoded signal. The spatial correction is based on the difference between the original and the decoded spatial image. This post-processing is later extended to a parametric coding method. The last part of this thesis presents a more exploratory method. This method studied audio signal compression by neural networks inspired by image compression models using variational autoencoders
Wenzek, Didier. "Construction de réseaux de neurones". Phd thesis, Grenoble INPG, 1993. http://tel.archives-ouvertes.fr/tel-00343569.
Pełny tekst źródłaTsopze, Norbert. "Treillis de Galois et réseaux de neurones : une approche constructive d'architecture des réseaux de neurones". Thesis, Artois, 2010. http://www.theses.fr/2010ARTO0407/document.
Pełny tekst źródłaThe artificial neural networks are successfully applied in many applications. But theusers are confronted with two problems : defining the architecture of the neural network able tosolve their problems and interpreting the network result. Many research works propose some solutionsabout these problems : to find out the architecture of the network, some authors proposeto use the problem domain theory and deduct the network architecture and some others proposeto dynamically add neurons in the existing networks until satisfaction. For the interpretabilityproblem, solutions consist to extract rules which describe the network behaviour after training.The contributions of this thesis concern these problems. The thesis are limited to the use of theartificial neural networks in solving the classification problem.In this thesis, we present a state of art of the existing methods of finding the neural networkarchitecture : we present a theoritical and experimental study of these methods. From this study,we observe some limits : difficulty to use some method when the knowledges are not available ;and the network is seem as ’black box’ when using other methods. We a new method calledCLANN (Concept Lattice-based Artificial Neural Network) which builds from the training dataa semi concepts lattice and translates this semi lattice into the network architecture. As CLANNis limited to the two classes problems, we propose MCLANN which extends CLANN to manyclasses problems.A new method of rules extraction called ’MaxSubsets Approach’ is also presented in thisthesis. Its particularity is the possibility of extracting the two kind of rules (If then and M-of-N)from an internal structure.We describe how to explain the MCLANN built network result aboutsome inputs
Voegtlin, Thomas. "Réseaux de neurones et auto-référence". Lyon 2, 2002. http://theses.univ-lyon2.fr/documents/lyon2/2002/voegtlin_t.
Pełny tekst źródłaThe purpose of this thesis is to present a class of unsupervised learning algorithms for recurrent networks. In the first part (chapters 1 to 4), I propose a new approach to this question, based on a simple principle: self-reference. A self-referent algorithm is not based on the minimization of an objective criterion, such as an error function, but on a subjective function, that depends on what the network has previously learned. An example of a supervised recurrent network where learning is self-referent is the Simple Recurrent Network (SRN) by Elman (1990). In the SRN, self-reference is applied to the supervised error back-propagation algorithm. In this aspect, the SRN differs from other generalizations of back-propagation to recurrent networks, that use an objective criterion, such as Back-Propagation Through Time, or Real-Time Recurrent Learning. In this thesis, I show that self-reference can be combined with several well-known unsupervised learning methods: the Self-Organizing Map (SOM), Principal Components Analysis (PCA), and Independent Components Analysis (ICA). These techniques are classically used to represent static data. Self-reference allows one to generalize these techniques to time series, and to define unsupervised learning algorithms for recurrent networks
Teytaud, Olivier. "Apprentissage, réseaux de neurones et applications". Lyon 2, 2001. http://theses.univ-lyon2.fr/documents/lyon2/2001/teytaud_o.
Pełny tekst źródłaCôté, Marc-Alexandre. "Réseaux de neurones génératifs avec structure". Thèse, Université de Sherbrooke, 2017. http://hdl.handle.net/11143/10489.
Pełny tekst źródłaJodouin, Jean-François. "Réseaux de neurones et traitement du langage naturel : étude des réseaux de neurones récurrents et de leurs représentations". Paris 11, 1993. http://www.theses.fr/1993PA112079.
Pełny tekst źródłaBrette, Romain. "Modèles Impulsionnels de Réseaux de Neurones Biologiques". Phd thesis, Université Pierre et Marie Curie - Paris VI, 2003. http://tel.archives-ouvertes.fr/tel-00005340.
Pełny tekst źródłaTardif, Patrice. "Autostructuration des réseaux de neurones avec retards". Thesis, Université Laval, 2007. http://www.theses.ulaval.ca/2007/24240/24240.pdf.
Pełny tekst źródłaMaktoobi, Sheler. "Couplage diffractif pour réseaux de neurones optiques". Thesis, Bourgogne Franche-Comté, 2020. http://www.theses.fr/2020UBFCD019.
Pełny tekst źródłaPhotonic networks with high performance can be considered as substrates for future computing systems. In comparison with electronics, photonic systems have substantial privileges, for instance the possibility of a fully parallel implementation of networks. Recently, neural networks have moved into the center of attention of the photonic community. One of the most important requirements for parallel large-scale photonic networks is to realize the connectivities. Diffraction is considered as a method to process the connections between the nodes (coupling) in optical neural networks. In the current thesis, we evaluate the scalability of a diffractive coupling in more details as follow:First, we begin with a general introductions for artificial intelligence, machine learning, artificial neural network and photonic neural networks. To establish a working neural network, learning rules are an essential part to optimize a configuration for obtaining a low error from the system, hence learning rules are introduced (Chapter 1). We investigate the fundamental concepts of diffractive coupling in our spatio-temporal reservoir. In that case, theory of diffraction is explained. We use an analytical scheme to provide the limits for the size of diffractive networks which is a part of our photonic neural network (Chapter 2). The concepts of diffractive coupling are investigated experimentally by two different experiments to confirm the analytical limits and to obtain maximum number of nodes which can be coupled in the photonic network (Chapter 3). Numerical simulations for such an experimental setup is modeled in two different schemes to obtain the maximum size of network numerically, which approaches a surface of 100 mm2 (Chapter 4). Finally, the complete photonic neural network is demonstrated. We design a spatially extended reservoir for 900 nodes. Consequently, our system generalizes the prediction for the chaotic Mackey–Glass sequence (Chapter 5)
Ouali, Jamel. "Architecture intégrée flexible pour réseaux de neurones". Grenoble INPG, 1991. http://www.theses.fr/1991INPG0035.
Pełny tekst źródłaBigot, Pascal. "Utilisation des réseaux de neurones pour la télégestion des réseaux techniques urbains". Lyon 1, 1995. http://www.theses.fr/1995LYO10036.
Pełny tekst źródłaKoiran, Pascal. "Puissance de calcul des réseaux de neurones artificiels". Lyon 1, 1993. http://www.theses.fr/1993LYO19003.
Pełny tekst źródłaGraïne, Slimane. "Inférence grammaticale régulière par les réseaux de neurones". Paris 13, 1994. http://www.theses.fr/1994PA132020.
Pełny tekst źródłaLe, Fablec Yann. "Prévision de trajectoires d'avions par réseaux de neurones". Toulouse, INPT, 1999. http://www.theses.fr/1999INPT034H.
Pełny tekst źródłaCorne, Christophe. "Parallélisation de réseaux de neurones sur architecture distribuée". Mulhouse, 1999. http://www.theses.fr/1999MULH0583.
Pełny tekst źródłaHe, Bing. "Estimation paramétrique du signal par réseaux de neurones". Lille 1, 2002. https://pepite-depot.univ-lille.fr/RESTREINT/Th_Num/2002/50376-2002-75.pdf.
Pełny tekst źródłaPompougnac, Hugo. "Spécification et compilation de réseaux de neurones embarqués". Electronic Thesis or Diss., Sorbonne université, 2022. http://www.theses.fr/2022SORUS436.
Pełny tekst źródłaIn this thesis, we propose an approach for the joint specification and compilation of both High-Performance Computing (HPC) and Real-Time Embedded (RTE) aspects of a system. Our approach is based on a formal, algorithmic and tooled integration between two formalisms underlying a large part of works in HPC and RTE fields: the SSA formalism and the synchronous dataflow language Lustre. The SSA formalism is a key component of many HPC compilers, including those used by Machine Learning frameworks such as TensorFlow or PyTorch. The Lustre language is a key component of implementation processes of critical embedded systems in avionics or rail transportation
Bénédic, Yohann. "Approche analytique pour l'optimisation de réseaux de neurones artificiels". Phd thesis, Université de Haute Alsace - Mulhouse, 2007. http://tel.archives-ouvertes.fr/tel-00605216.
Pełny tekst źródłaGatet, Laurent. "Intégration de Réseaux de Neurones pour la Télémétrie Laser". Phd thesis, Toulouse, INPT, 2007. http://oatao.univ-toulouse.fr/7595/1/gatet.pdf.
Pełny tekst źródłaRobitaille, Benoît. "Contrôle adaptatif par entraînement spécialisé de réseaux de neurones". Thesis, National Library of Canada = Bibliothèque nationale du Canada, 1997. http://www.collectionscanada.ca/obj/s4/f2/dsk2/ftp02/NQ35778.pdf.
Pełny tekst źródłaDucom, Jean-Christophe. "Codage temporel et apprentissage dans les réseaux de neurones". Aix-Marseille 1, 1996. http://www.theses.fr/1996AIX11041.
Pełny tekst źródłaBenaïm, Michel. "Dynamiques d'activation et dynamiques d'apprentissage des réseaux de neurones". Toulouse, ENSAE, 1992. http://www.theses.fr/1992ESAE0001.
Pełny tekst źródłaAupetit, Michaël. "Approximation de variétés par réseaux de neurones auto-organisés". Grenoble INPG, 2001. http://www.theses.fr/2001INPG0128.
Pełny tekst źródłaJiang, Fei. "Optimisation de la topologie de grands réseaux de neurones". Paris 11, 2009. http://www.theses.fr/2009PA112211.
Pełny tekst źródłaIn this dissertation, we present our study regarding the influence of the topology on the learning performances of neural networks with complex topologies. Three different neural networks have been investigated: the classical Self-Organizing Maps (SOM) with complex graph topology, the Echo States Network (ESN) and the Standard Model Features(SMF). In each case, we begin by comparing the performances of different topologies for the same task. We then try to optimize the topology of some neural network in order to improve such performance. The first part deals with Self-Organizing Maps, and the task is the standard classification of handwritten digits from the MNIST database. We show that topology has a small impact on performance and robustness to neuron failures, at least at long learning times. Performance may however be increased by almost 10% by artificial evolution of the network topology. In our experimental conditions, the evolved networks are more random than their parents, but display a more heterogeneous degree distribution. In the second part, we propose to apply CMA-ES, the state-of-the-art method in evolutionary continuous parameter optimization, to the evolutionary learning of the parameters of an Echo State Network (the Readout weights, of course, but also, Spectral Radius, Slopes of the neurons active function). First, a standard supervised learning problem is used to validate the approach and compare it to the original one. But the flexibility of Evolutionary optimization allows us to optimize not only the outgoing weights but also, or alternatively, other ESN parameters, sometimes leading to improved results. The classical double pole balancing control problem is used to demonstrate the feasibility of evolutionary reinforcement learning of ESN. We show that the evolutionary ESN obtain results that are comparable with those of the best topology-learning neuro-evolution methods. Finally, the last part presents our initial research of the SMF - a visual object recognition model which is inspired by the visual cortex. Two version based on SMF are applied to the PASCAL Visual multi-Object recognition Challenge (VOC2008). The long terms goal is to find the optimal topology of the SMF model, but the computation cost is however too expensive to optimize the complete topology directly. So as a first step, we apply an Evolutionary Algorithm to auto-select the feature used by the systems. We show that, for the VOC2008 challenge, with only 20% selected feature, the system can perform as well as with all 1000 randomly selected feature
Alvado, Ludovic. "Neurones artificiels sur silicium : une évolution vers les réseaux". Bordeaux 1, 2003. http://www.theses.fr/2003BOR12674.
Pełny tekst źródłaThis thesis describes a new approach for modelling biological neuron networks. This approach uses analogue specific integrated circuit (ASIC) in which Hodgkin-Huxley formalism as been implemented to integrate medium density artificial neural network, modelled at a biological realistic level. This thesis also deals with the component mismatches problem and the pertinent choice of optimized structure dedicated to network applications
Biela, Philippe. "Classification automatique d'observations multidimensionnelles par réseaux de neurones compétitifs". Lille 1, 1999. https://pepite-depot.univ-lille.fr/LIBRE/Th_Num/1999/50376-1999-469.pdf.
Pełny tekst źródłaElhor, Noureddine. "Suivi de fonctionnement d'une éolienne par réseaux de neurones". Lille 1, 2000. https://pepite-depot.univ-lille.fr/LIBRE/Th_Num/2000/50376-2000-57.pdf.
Pełny tekst źródłaLa projection non lineaire offre une visualisation plane des donnees alors que les comparaisons entre les sorties estimees et reelles generent des residus permettant de verifier si le spectre courant a devie ou non du fonctionnement normal memorise par le reseau. Pour valider notre etude, nous avons teste les performances du reseau sur des donnees en fonctionnement normal acquis dans des situations differentes de celles de la base d'apprentissage et sur des situations correspondant a des defauts simules type balourd. Ce type de defauts se manifeste par l'apparition d'une frequence et ses harmoniques dans le spectre d'energie. Dans les deux situations, les resultats obtenus par le reseau ont ete quantifies et se sont reveles satisfaisants. Nous avons exploite, par ailleurs, un reseau modulaire a apprentissage supervise pour la discrimination entre deux situations : le fonctionnement normal et la presence d'un defaut type balourd d'une amplitude minimale fixee. Les performances du reseau ont ete testees sur des defauts d'amplitudes differentes. Souvent les methodes de diagnostic sont appliquees sur des bancs d'essais dans des conditions de laboratoire controlees. Notre demarche est d'autant plus importante que nous surveillons une machine reelle en pleine production
Chakik, Fadi El. "Maximum d'entropie et réseaux de neurones pour la classification". Grenoble INPG, 1998. http://www.theses.fr/1998INPG0091.
Pełny tekst źródłaDemartines, Pierre. "Analyse de données par réseaux de neurones auto-organisés". Grenoble INPG, 1994. http://www.theses.fr/1994INPG0129.
Pełny tekst źródłaOussar, Yacine. "Réseaux d'ondelettes et réseaux de neurones pour la modélisation statique et dynamique de processus". Phd thesis, Université Pierre et Marie Curie - Paris VI, 1998. http://pastel.archives-ouvertes.fr/pastel-00000677.
Pełny tekst źródłaBissery, Christophe. "La détection centralisée des fuites sur les réseaux d'eau potable par réseaux de neurones". Lyon, INSA, 1994. http://www.theses.fr/1994ISAL0112.
Pełny tekst źródłaFor few years, under the influence of the urban environment, the perception of dysfunction risk in technical systems and in particular in water supply networks has changed. The lack of risk doesn't exist and it's necessary to learn how to manage it. It's in this way that appears the need of centralized leakage detection on water supply networks, leaks that represent an important part of the dysfunction risk of water supply. This study proposes a centralized leakage detection system using a computerized neural network approach. The building method of learning bases and the sensors localization method are pointed out and developed. This study has showed that on a realistic network model results obtained with the centralized leakage detection system using a computerized neural network approach allowed experimentations on real networks. The study ends on the presentation of the working priorities for these real experimentations (and in particular the need of hourly water consumption previsions)
Basterrech, Sebastián. "Apprentissage avec les réseaux de neurones aléatoires et les machines de calcul avec réservoir de neurones". Rennes 1, 2012. http://www.theses.fr/2012REN1S178.
Pełny tekst źródłaSince the 1980s a new computational model merging concepts from neural networks and queuing theory was developed. The model was introduced under the name of Random Neural Networks (RNNs), inside the field of Neural Networks. In this thesis, a first contribution consists of an adaptation of quasi-Newton optimisation methods for training the RNN model. In the last decade, a new computational paradigm was introduced in the field of Machine Learning, under the name of Reservoir Computing (RC). One of the pioneers and most diffused RC methods is the Echo State Network (ESN) model. Here, we propose a method based on topographic maps to initialise the ESN procedure. Another contribution of the thesis is the introduction of a new RC model called the Echo State Queueing Network (ESQN), where we use ideas coming from RNNs for the design of the reservoir. An ESQN consists of an ESN where the reservoir has a new dynamics inspired by recurrent RNNs. In this thesis, we position the ESQN method in the global Machine Learning area, and provide examples of their use and performances. Finally, we propose a method for real–time estimation of Speech Quality using the learning tools above described. Audio quality in the Internet can be strongly affected by network conditions. As a consequence, many techniques to evaluate it have been developed. In particular, the ITU-T adopted in 2001 a technique called Perceptual Evaluation of Speech Quality (PESQ) to automatically measuring speech quality. PESQ is a well-known and widely used procedure, providing in general an accurate evaluation of perceptual quality by comparing the original and received voice sequences. The thesis provides a procedure for estimating PESQ output working only with measures taken on the network state and using some properties of the communication system, without any original signal. The experimental results obtained prove the capability of our approach to give good estimations of the speech quality in a real–time context
Krauth, Werner. "Physique statistique des réseaux de neurones et de l'optimisation combinatoire". Phd thesis, Université Paris Sud - Paris XI, 1989. http://tel.archives-ouvertes.fr/tel-00011866.
Pełny tekst źródłaPersonnaz, Léon. "Etude des réseaux de neurones formels : conception, propriétés et applications". Paris 6, 1986. http://www.theses.fr/1986PA066569.
Pełny tekst źródłaMercier, David. "Hétéro-association de signaux audio-vidéo par réseaux de neurones". Rennes 1, 2003. http://www.theses.fr/2003REN10009.
Pełny tekst źródłaPuechmorel, Stéphane. "Réseaux de neurones et optimisation globale en analyse temps-fréquence". Toulouse, INPT, 1994. http://www.theses.fr/1994INPT105H.
Pełny tekst źródła