Дисертації з теми "Compression de réseaux neuronaux"
Оформте джерело за APA, MLA, Chicago, Harvard та іншими стилями
Ознайомтеся з топ-50 дисертацій для дослідження на тему "Compression de réseaux neuronaux".
Біля кожної праці в переліку літератури доступна кнопка «Додати до бібліографії». Скористайтеся нею – і ми автоматично оформимо бібліографічне посилання на обрану працю в потрібному вам стилі цитування: APA, MLA, «Гарвард», «Чикаго», «Ванкувер» тощо.
Також ви можете завантажити повний текст наукової публікації у форматі «.pdf» та прочитати онлайн анотацію до роботи, якщо відповідні параметри наявні в метаданих.
Переглядайте дисертації для різних дисциплін та оформлюйте правильно вашу бібліографію.
Foucher, Christophe. "Analyse et amélioration d'algorithmes neuronaux et non neuronaux de quantification vectorielle pour la compression d'images." Rennes 1, 2002. http://www.theses.fr/2002REN10120.
Повний текст джерелаDupont, Robin. "Deep Neural Network Compression for Visual Recognition." Electronic Thesis or Diss., Sorbonne université, 2023. http://www.theses.fr/2023SORUS565.
Повний текст джерелаThanks 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
Leconte, Louis. "Compression and federated learning : an approach to frugal machine learning." Electronic Thesis or Diss., Sorbonne université, 2024. http://www.theses.fr/2024SORUS107.
Повний текст джерела“Intelligent” devices and tools are gradually becoming the standard, as the implementation of algorithms based on artificial neural networks is experiencing widespread development. Neural networks consist of non-linear machine learning models that manipulate high-dimensional objects and obtain state-of-the-art performances in various areas, such as image recognition, speech recognition, natural language processing, and recommendation systems.However, training a neural network on a device with lower computing capacity can be challenging, as it can imply cutting back on memory, computing time or power. A natural approach to simplify this training is to use quantized neural networks, whose parameters and operations use efficient low-bit primitives. However, optimizing a function over a discrete set in high dimension is complex, and can still be prohibitively expensive in terms of computational power. For this reason, many modern applications use a network of devices to store individual data and share the computational load. A new approach, federated learning, considers a distributed environment: Data is stored on devices and a centralized server orchestrates the training process across multiple devices.In this thesis, we investigate different aspects of (stochastic) optimization with the goal of reducing energy costs for potentially very heterogeneous devices. The first two contributions of this work are dedicated to the case of quantized neural networks. Our first idea is based on an annealing strategy: we formulate the discrete optimization problem as a constrained optimization problem (where the size of the constraint is reduced over iterations). We then focus on a heuristic for training binary deep neural networks. In this particular framework, the parameters of the neural networks can only have two values. The rest of the thesis is about efficient federated learning. Following our contributions developed for training quantized neural network, we integrate them into a federated environment. Then, we propose a novel unbiased compression technique that can be used in any gradient based distributed optimization framework. Our final contributions address the particular case of asynchronous federated learning, where devices have different computational speeds and/or access to bandwidth. We first propose a contribution that reweights the contributions of distributed devices. Then, in our final work, through a detailed queuing dynamics analysis, we propose a significant improvement to the complexity bounds provided in the literature onasynchronous federated learning.In summary, this thesis presents novel contributions to the field of quantized neural networks and federated learning by addressing critical challenges and providing innovative solutions for efficient and sustainable learning in a distributed and heterogeneous environment. Although the potential benefits are promising, especially in terms of energy savings, caution is needed as a rebound effect could occur
Yvinec, Edouard. "Efficient Neural Networks : Post Training Pruning and Quantization." Electronic Thesis or Diss., Sorbonne université, 2023. http://www.theses.fr/2023SORUS581.
Повний текст джерелаDeep 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
Louis, Thomas. "Conventionnel ou bio-inspiré ? Stratégies d'optimisation de l'efficacité énergétique des réseaux de neurones pour environnements à ressources limitées." Electronic Thesis or Diss., Université Côte d'Azur, 2025. http://www.theses.fr/2025COAZ4001.
Повний текст джерелаIntegrating artificial intelligence (AI) algorithms directly into satellites presents numerous challenges. These embedded systems, which are heavily limited in energy consumption and memory footprint, must also withstand interference. This systematically requires the use of system-on-chip (SoC) solutions to combine two so-called “heterogeneous” systems: a versatile microcontroller and an energy-efficient computing accelerator (such as an FPGA or ASIC). To address the challenges related to deploying such architectures, this thesis focuses on optimizing and deploying neural networks on heterogeneous embedded architectures, aiming to balance energy consumption and AI performance.In Chapter 2 of this thesis, an in-depth study of recent compression techniques for feedforward neural networks (FNN) like MLPs or CNNs was conducted. These techniques, which reduce the computational complexity and memory footprint of these models, are essential for deployment in resource-constrained environments. Spiking neural networks (SNN) were also explored. These bio-inspired networks can indeed offer greater energy efficiency compared to FNNs.In Chapter 3, we adapted and developed innovative quantization methods to reduce the number of bits used to represent the values in a spiking network. This allowed us to compare the quantization of SNNs and FNNs, to understand and assess their respective trade-offs in terms of losses and gains. Reducing the activity of an SNN (e.g., the number of spikes generated during inference) directly improves the energy efficiency of SNNs. To this end, in Chapter 4, we leveraged knowledge distillation and regularization techniques. These methods reduce the spiking activity of the network while preserving its accuracy, ensuring effective operation of SNNs on resource-limited hardware.In the final part of this thesis, we explored the hybridization of SNNs and FNNs. These hybrid networks (HNN) aim to further optimize energy efficiency while enhancing performance. We also proposed innovative multi-timestep networks, which process information with different latencies across layers within the same SNN. Experimental results show that this approach enables a reduction in overall energy consumption while maintaining performance across a range of tasks.This thesis serves as a foundation for deploying future neural network applications in space. To validate our methods, we provide a comparative analysis on various public datasets (CIFAR-10, CIFAR-100, MNIST, Google Speech Commands) as well as on a private dataset for cloud segmentation. Our approaches are evaluated based on metrics such as accuracy, energy consumption, or SNN activity. This research extends beyond aerospace applications. We have demonstrated the potential of quantized SNNs, hybrid neural networks, and multi-timestep networks for a variety of real-world scenarios where energy efficiency is critical. This work offers promising prospects for fields such as IoT devices, autonomous vehicles, and other systems requiring efficient AI deployment
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.
Повний текст джерелаSince 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
Resmerita, Diana. "Compression pour l'apprentissage en profondeur." Thesis, Université Côte d'Azur, 2022. http://www.theses.fr/2022COAZ4043.
Повний текст джерелаAutonomous 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
Fernandez, Brillet Lucas. "Réseaux de neurones CNN pour la vision embarquée." Thesis, Université Grenoble Alpes, 2020. http://www.theses.fr/2020GRALM043.
Повний текст джерелаRecently, 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
Verma, Sagar. "Deep Neural Network Modeling of Electric Motors." Electronic Thesis or Diss., université Paris-Saclay, 2023. http://www.theses.fr/2023UPAST088.
Повний текст джерелаThis 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
Metz, Clément. "Codages optimisés pour la conception d'accélérateurs matériels de réseaux de neurones profonds." Electronic Thesis or Diss., université Paris-Saclay, 2023. http://www.theses.fr/2023UPAST190.
Повний текст джерелаNeural networks are an important component of machine learning tools because of their wide range of applications (health, energy, defence, finance, autonomous navigation, etc.). The performance of neural networks is greatly influenced by the complexity of their architecture in terms of the number of layers, neurons and connections. But the training and inference of ever-larger networks translates to greater demands on hardware resources and longer computing times. Conversely, their portability is limited on embedded systems with low memory and/or computing capacity.The aim of this thesis is to study and design methods for reducing the hardware footprint of neural networks while preserving their performance as much as possible. We restrict ourselves to convolution networks dedicated to computer vision by studying the possibilities offered by quantization. Quantization aims to reduce the hardware footprint, in terms of memory, bandwidth and computation operators, by reducing the number of bits in the network parameters and activations.The contributions of this thesis consist of a new post-training quantization method based on the exploitation of spatial correlations of network parameters, an approach facilitating the learning of very highly quantized networks, and a method aiming to combine mixed precision quantization and lossless entropy coding.The contents of this thesis are essentially limited to algorithmic aspects, but the research orientations were strongly influenced by the requirement for hardware feasibility of our solutions
Lerogeron, Hugo. "Approximation de Dynamic Time Warping par réseaux de neurones pour la compression de signaux EEG et l'analyse de l'insomnie induite par le COVID long." Electronic Thesis or Diss., Normandie, 2023. http://www.theses.fr/2023NORMR098.
Повний текст джерелаThis manuscript presents the work carried out within the framework of the CIFRE thesis conducted in partnership between LITIS and Saagie and which is part of the PANDORE-IA project in association with the VIFASOM sleep center.Electroencephalographic (EEG) signals are very useful in helping experts identify various abnormalities like sleep disorders. Recently, the community has shown great interest in long COVID and its various impacts on sleep. However, these signals are voluminous: compression allows reducing storage and transfer costs. Recent compression approaches are based on autoencoders that use a cost function to learn. It is usually the Mean Squared Error (MSE), but there are metrics more suited to time series, particularly Dynamic Time Warping (DTW). However, DTW is not differentiable and thus can not be used as a loss for end-to-end learning.To solve this problem, we propose in this thesis two approaches to approximate DTW based on neural networks. The first approach uses a Siamese network to project the signals so that the Euclidean distance of the projected signals is as close as possible to the DTW of the original signals. The second approach attempts to predict directly the DTW value. We show that these approaches are faster than other differentiable approximations of DTW while obtaining results similar to DTW in query or classification on sleep data.We then demonstrate that the Siamese approximation can be used as a cost function for learning a sleep signal compression system based on an autoencoder. We justify the choice of the network architecture by the fact that it allows us to vary the compression rate. We evaluate this compression system by classification on the compressed and then reconstructed signals, and show that the usual measures of compression quality do not allow for a proper assessment of a compression system's ability to retain discriminative information. We show that our DTW approximations yield better performance on the reconstructed data than conventional compression algorithms and other reconstruction losses.Finally, to study the impact of long COVID on insomnia, we collect and provide the community with a dataset named COVISLEEP, containing polysomnographies of individuals who developed chronic insomnia after COVID infection, and of those suffering from chronic insomnia but who have not been infected by the virus. We compare various state-of-the-art approaches for sleep staging, and use the best one for learning the detection of long COVID. We highlight the difficulty of the task, especially due to the high variability among patients. This offers a complex dataset to the community that allows for the development of more effective methods
Carvalho, Micael. "Deep representation spaces." Electronic Thesis or Diss., Sorbonne université, 2018. http://www.theses.fr/2018SORUS292.
Повний текст джерелаIn 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
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.
Повний текст джерелаIn 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
Linhares, Pontes Elvys. "Compressive Cross-Language Text Summarization." Thesis, Avignon, 2018. http://www.theses.fr/2018AVIG0232/document.
Повний текст джерелаThe popularization of social networks and digital documents increased quickly the informationavailable on the Internet. However, this huge amount of data cannot be analyzedmanually. Natural Language Processing (NLP) analyzes the interactions betweencomputers and human languages in order to process and to analyze natural languagedata. NLP techniques incorporate a variety of methods, including linguistics, semanticsand statistics to extract entities, relationships and understand a document. Amongseveral NLP applications, we are interested, in this thesis, in the cross-language textsummarization which produces a summary in a language different from the languageof the source documents. We also analyzed other NLP tasks (word encoding representation,semantic similarity, sentence and multi-sentence compression) to generate morestable and informative cross-lingual summaries.Most of NLP applications (including all types of text summarization) use a kind ofsimilarity measure to analyze and to compare the meaning of words, chunks, sentencesand texts in their approaches. A way to analyze this similarity is to generate a representationfor these sentences that contains the meaning of them. The meaning of sentencesis defined by several elements, such as the context of words and expressions, the orderof words and the previous information. Simple metrics, such as cosine metric andEuclidean distance, provide a measure of similarity between two sentences; however,they do not analyze the order of words or multi-words. Analyzing these problems,we propose a neural network model that combines recurrent and convolutional neuralnetworks to estimate the semantic similarity of a pair of sentences (or texts) based onthe local and general contexts of words. Our model predicted better similarity scoresthan baselines by analyzing better the local and the general meanings of words andmulti-word expressions.In order to remove redundancies and non-relevant information of similar sentences,we propose a multi-sentence compression method that compresses similar sentencesby fusing them in correct and short compressions that contain the main information ofthese similar sentences. We model clusters of similar sentences as word graphs. Then,we apply an integer linear programming model that guides the compression of theseclusters based on a list of keywords. We look for a path in the word graph that has goodcohesion and contains the maximum of keywords. Our approach outperformed baselinesby generating more informative and correct compressions for French, Portugueseand Spanish languages. Finally, we combine these previous methods to build a cross-language text summarizationsystem. Our system is an {English, French, Portuguese, Spanish}-to-{English,French} cross-language text summarization framework that analyzes the informationin both languages to identify the most relevant sentences. Inspired by the compressivetext summarization methods in monolingual analysis, we adapt our multi-sentencecompression method for this problem to just keep the main information. Our systemproves to be a good alternative to compress redundant information and to preserve relevantinformation. Our system improves informativeness scores without losing grammaticalquality for French-to-English cross-lingual summaries. Analyzing {English,French, Portuguese, Spanish}-to-{English, French} cross-lingual summaries, our systemsignificantly outperforms extractive baselines in the state of the art for all these languages.In addition, we analyze the cross-language text summarization of transcriptdocuments. Our approach achieved better and more stable scores even for these documentsthat have grammatical errors and missing information
Mahé, Pierre. "Codage ambisonique pour les communications immersives." Thesis, La Rochelle, 2022. http://www.theses.fr/2022LAROS011.
Повний текст джерелаThis 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
Rouzier, Sophie. "Réseaux neuronaux et modularité." Grenoble INPG, 1998. http://www.theses.fr/1998INPG0032.
Повний текст джерелаCohen, Floriane. "Architectures dynamiques des réseaux neuronaux in vitro." Thesis, Paris Sciences et Lettres (ComUE), 2018. https://tel.archives-ouvertes.fr/tel-02512337.
Повний текст джерелаThe function of the nervous system relies on the establishment of complex neuronal circuitry. During development, axon branching allows each neuron to establish synaptic contacts with multiple targets and is essential to the assembly of highly interconnected networks. Therefore, understanding the mechanisms underlying the control of neuronal branching is crucial in the study of neuronal circuit development.In this thesis, we investigated this phenomenon by imposing morphological constraints to neurons through the use of different chemical micropatterning techniques. Using static micropatterns, we explored branching behavior in a wide range of geometries with a focus on the influence of branching angle. In parallel, we have also worked on the development of a dynamic patterning technique based on spontaneous adsorption of comb-like derivatives of poly-L-lysine to form switchable patterns on highly cell-repellent surfaces, with the aim of creating a platform allowing for spatio-temporally controlled generation of neurite branches
Abdelazim, Gamil. "Applications de réseaux neuronaux à l'optimisation combinatoire." Paris 9, 1992. https://portail.bu.dauphine.fr/fileviewer/index.php?doc=1992PA090029.
Повний текст джерелаDeffuant, Guillaume. "Réseaux connexionnistes auto-construits." Paris, EHESS, 1992. http://www.theses.fr/1992EHES0303.
Повний текст джерелаVoegtlin, Thomas. "Réseaux de neurones et auto-référence." Lyon 2, 2002. http://theses.univ-lyon2.fr/documents/lyon2/2002/voegtlin_t.
Повний текст джерелаThe 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
Addor, Jean-Bernard. "Étude des réseaux neuronaux auto-organisés de type LINSKER et analyse multifractale de signaux neuronaux." Thesis, National Library of Canada = Bibliothèque nationale du Canada, 1998. http://www.collectionscanada.ca/obj/s4/f2/dsk1/tape10/PQDD_0017/MQ55567.pdf.
Повний текст джерелаRonald, Edmund. "Apprentissage évolutionniste des réseaux neuromimétiques." Palaiseau, Ecole polytechnique, 1997. http://www.theses.fr/1997EPXX0048.
Повний текст джерелаMeunier, Claire. "Etude de la neuromodulation des réseaux neuronaux du cortex." Thesis, Paris 11, 2013. http://www.theses.fr/2013PA11T089/document.
Повний текст джерелаThe cortex is crucial for processes such as sensory perception, cognition and memory. Cortical organization is based on neuronal networks composed of excitatory (E) and inhibitory (I) neurons which target layer 5 pyramidal neurons. Dysfunctions of such networks result in psychiatric pathologies including major depression and schizophrenia. Regulations of cortical activity also involve neuromodulators such as serotonin, dopamine, D-serine and glycine. The current body of work decipher the interactions of the effects of 5-HT1A-, D1-, D2-, NMDA- and Glycine-receptors activation on the E-I balance and synaptic plasticity. The electrophysiological data that I have generated in the prefrontal cortex show that concomitant activation of 5-HT1A- and D1-receptors downregulates the induction of LTD whilst 5-HT1A coupled to D2-receptors activation promotes LTD induction, via a common modulation of GSK3β. I also collected data from the visual cortex, showing that D-serine is the co-agonist NMDA-receptor in this brain region and is, as such, required for LTP-induction. Glycine was instead found to act on dendritic Glycine-receptors, resulting in a shunt, which altered dendritic integration and thus turned LTP to a LTD-like effect at the somatic level
Lion, Marc. "Filtrage adaptatif par réseaux neuronaux : application à la trajectographie." École nationale supérieure de l'aéronautique et de l'espace (Toulouse ; 1972-2007), 2000. http://www.theses.fr/2000ESAE0003.
Повний текст джерелаAissa, Wafa. "Réseaux de modules neuronaux pour un raisonnement visuel compositionnel." Electronic Thesis or Diss., Paris, HESAM, 2023. http://www.theses.fr/2023HESAC033.
Повний текст джерелаThe context of this PhD thesis is compositional visual reasoning. When presented with an image and a question pair, our objective is to have neural networks models answer the question by following a reasoning chain defined by a program. We assess the model's reasoning ability through a Visual Question Answering (VQA) setup.Compositional VQA breaks down complex questions into modular easier sub-problems.These sub-problems include reasoning skills such as object and attribute detection, relation detection, logical operations, counting, and comparisons. Each sub-problem is assigned to a different module. This approach discourages shortcuts, demanding an explicit understanding of the problem. It also promotes transparency and explainability.Neural module networks (NMN) are used to enable compositional reasoning. The framework is based on a generator-executor framework, the generator learns the translation of the question to its function program. The executor instantiates a neural module network where each function is assigned to a specific module. We also design a neural modules catalog and define the function and the structure of each module. The training and evaluations are conducted using the pre-processed GQA dataset cite{gqa}, which includes natural language questions, functional programs representing the reasoning chain, images, and corresponding answers.The research contributions revolve around the establishment of an NMN framework for the VQA task.One primary contribution involves the integration of vision and language pre-trained (VLP) representations into modular VQA. This integration serves as a ``warm-start" mechanism for initializing the reasoning process.The experiments demonstrate that cross-modal vision and language representations outperform uni-modal ones. This utilization enables the capture of intricate relationships within each individual modality while also facilitating alignment between different modalities, consequently enhancing overall accuracy of our NMN.Moreover, we explore various training techniques to enhance the learning process and improve cost-efficiency. In addition to optimizing the modules within the reasoning chain to collaboratively produce accurate answers, we introduce a teacher-guidance approach to optimize the intermediate modules in the reasoning chain. This ensures that these modules perform their specific reasoning sub-tasks without taking shortcuts or compromising the reasoning process's integrity. We propose and implement several teacher-guidance techniques, one of which draws inspiration from the teacher-forcing method commonly used in sequential models. Comparative analyses demonstrate the advantages of our teacher-guidance approach for NMNs, as detailed in our paper [1].We also introduce a novel Curriculum Learning (CL) strategy tailored for NMNs to reorganize the training examples and define a start-small training strategy. We begin by learning simpler programs and progressively increase the complexity of the training programs. We use several difficulty criteria to define the CL approach. Our findings demonstrate that by selecting the appropriate CL method, we can significantly reduce the training cost and required training data, with only a limited impact on the final VQA accuracy. This significant contribution forms the core of our paper [2].[1] W. Aissa, M. Ferecatu, and M. Crucianu. Curriculum learning for compositional visual reasoning. In Proceedings of VISIGRAPP 2023, Volume 5: VISAPP, 2023.[2] W. Aissa, M. Ferecatu, and M. Crucianu. Multimodal representations for teacher-guidedcompositional visual reasoning. In Advanced Concepts for Intelligent Vision Systems, 21st International Conference (ACIVS 2023). Springer International Publishing, 2023.[3] D. A. Hudson and C. D. Manning. GQA: A new dataset for real-world visual reasoning and compositional question answering. 2019
Teytaud, Olivier. "Apprentissage, réseaux de neurones et applications." Lyon 2, 2001. http://theses.univ-lyon2.fr/documents/lyon2/2001/teytaud_o.
Повний текст джерелаLagacherie, Hervé. "L'analyse des données cliniques et biologiques par les réseaux neuronaux." Bordeaux 2, 1997. http://www.theses.fr/1997BOR2P091.
Повний текст джерелаBasulaim, Khaled. "Application des réseaux neuronaux en visionique : traitement d'images et classification." Le Havre, 1995. http://www.theses.fr/1995LEHA0002.
Повний текст джерелаMessai, Nadhir. "Surveillance du trafic urbain et interurbain à base de modèles neuronaux." Besançon, 2003. http://www.theses.fr/2003BESA2033.
Повний текст джерелаThe minimization of the effects of the nonrecurring congestions is a capital challenge that the management traffic centres try to raise using traffic monitoring tools. However, the algorithms employed are generally based on some heuristics and they are incompatible with the operational constraints of exploitation. This work presents a monitoring approach based on neural networks models and gives some arguments in favour of the use of neural networks (NN) for traffic modelling and monitoring. In the context of traffic modelling, NN are used in a simple and systematic way to extract fundamental diagrams (DF) which do not require any calibration effort. In addition, we formulate initialization algorithms which avoid the convergence towards local minima which are very far away from the global minimum. NN are also employed as models that predict the traffic flow. This modelling approach, which is based on a systematic validation procedure, has the advantage to keep away from the theoretical and the experimental difficulties related to the validation of the existing models. Once the modelling phase is achieved, we were interested in the monitoring problems. In this context, we propose, in a first stage, an algorithm based on the static model (DF). The incidents detection results from a segmentation of the space debit/density in four areas and the classification of the measured data in one of these areas. In order to take the edge off the limitations of this static algorithm, we were interested in a monitoring approach, which is based on the analysis of residues resulting from the comparison of real data with those obtained from the prediction model. Furthermore, we develop an alternative procedure which mixes four recurring networks. Lastly, the problem of traffic supervision on a network is tackled. Modular and hierarchical schemas which deal with this difficulty are proposed. These schemas are based on Petri nets and they have the potential to allow a temporal and a spatial observation of incidents and congestions propagations. The modeling and monitoring proposed approaches are tested and compared on two real systems: an urban crossroads of the town of Nancy and a portion of the Californian freeway I-880
Morel, Hervé. "Diagnostic du rotor principal de l'hélicoptère : analyse de signatures vibratoires par réseaux de neurones." Paris, ENSAM, 2007. http://www.theses.fr/2007ENAM0001.
Повний текст джерелаAziz, Mohammed, and Abdelaziz Bensrhair. "Apprentissage de réseaux de neurones impulsionnels. Application à des systèmes sensorimoteurs." INSA de Rouen, 2005. http://www.theses.fr/2005ISAM0005.
Повний текст джерелаQuélavoine, Régis. "Etude de l'apprentissage et des structures des réseaux de neurones multicouches pour l'analyse de données." Avignon, 1997. http://www.theses.fr/1997AVIG0002.
Повний текст джерелаReyes, Salgado Gerardo. "Connaissances de haut niveau dans les systèmes hybrides neuro-symboliques." Grenoble INPG, 2001. http://www.theses.fr/2001INPG0047.
Повний текст джерелаMONNET, MARIE-LAURE. "Interaction des proprietes cellulaires et des proprietes synaptiques dans la dynamique neuronale." Palaiseau, Ecole polytechnique, 1997. http://www.theses.fr/1997EPXX0009.
Повний текст джерелаHuguet, Stéphane. "Application de classificateurs aux données d'émission acoustique : identification de la signature acoustique des mécanismes d'endommagement dans les composites à matrice polymère." Lyon, INSA, 2002. http://theses.insa-lyon.fr/publication/2002ISAL0087/these.pdf.
Повний текст джерелаComposites. Unidirectional samples tested in off-axis tensile tests, associated with tensile tests on resin and microcomposite, allow the identification of AE signatures of the three damage modes in composite materials: matrix cracking, fiber/matrix decohesion and fiber fracture. Statistical analysis techniques, with Keywords nearest neighbors and neural networks (Kohonen map) were used to build an automatic recognition tool able to distinguish between the signals from those damage modes. This methodology was successfully also applied on +-55 degrees glass-fiber/epoxy-resin laminates
Seube, Nicolas. "Régulation de systèmes contrôlés avec contraintes sur l'état par réseaux de neurones." Paris 9, 1992. https://portail.bu.dauphine.fr/fileviewer/index.php?doc=1992PA090013.
Повний текст джерелаGuérin-Dugué, Anne. "Crasy : un Calculateur de Réseaux Adaptatifs SYstolique : application au calcul neuromimétique." Grenoble INPG, 1987. http://www.theses.fr/1987INPG0082.
Повний текст джерелаKlein, Jacques-Olivier. "Contribution a l'etude de l'adequation algorithme-architecture : machines de boltzmann et circuits analogiques cellulaires." Paris 11, 1995. http://www.theses.fr/1995PA112009.
Повний текст джерелаLe, Fablec Yann. "Prévision de trajectoires d'avions par réseaux de neurones." Toulouse, INPT, 1999. http://www.theses.fr/1999INPT034H.
Повний текст джерелаSéjourné, Julien. "Dynamique des phases de mémoire et réseaux neuronaux chez Drosophila melanogaster." Phd thesis, Paris 6, 2009. http://pastel.archives-ouvertes.fr/pastel-00567093.
Повний текст джерелаHenniges, Philippe. "PSO pour l'apprentissage supervisé des réseaux neuronaux de type fuzzy ARTMAP." Mémoire, École de technologie supérieure, 2006. http://espace.etsmtl.ca/508/1/HENNIGES_Pihilippe.pdf.
Повний текст джерелаGrava, Cristian. "Compensation de mouvement par réseaux neuronaux cellulaires : application en imagerie médicale." Lyon, INSA, 2003. http://theses.insa-lyon.fr/publication/2003ISAL0096/these.pdf.
Повний текст джерелаThis work concern motion estimation and compensation in image sequences. The originality of this thesis consist in the proposed solutions for a fast implementation of the interpolation and motion compensation on an existing hardware structure based on Cellular Neural Networks (CNN). To improve the precision of classical motion estimation methods, we developed markovian approaches, taking into account the discontinuities in the motion field. The deterministic algorithms implemented for the minimization of the maximum a Posteriori energy are ICM and mean-field annealing. The motion compensation implemented on CNN is based on the motion field, already estimated. This is facilitated by the similitude between the neighbourhood in the Markov sense and the physical structure of the CNN. The performances of the proposed algorithms were studied on images from echography and X-ray tomography. The time-gain is several orders (until 10~) greater and they are an alternative to conventional solutions
Remm, Jean-François. "Extraction de connaissances par réseaux neuronaux : application au domaine du radar." Nancy 1, 1996. http://www.theses.fr/1996NAN10366.
Повний текст джерелаDaoudi, Mohamed. "Classification interactive multidimensionnelle par les réseaux neuronaux et la morphologie mathématique." Lille 1, 1993. http://www.theses.fr/1993LIL10134.
Повний текст джерелаNargeot, Romuald. "Plasticité des réseaux neuronaux stomatogastriques des crustacés : étude électrophysiologique et pharmacologique." Bordeaux 1, 1995. http://www.theses.fr/1995BOR10514.
Повний текст джерелаParey, Christine. "Logique majoritaire trivalente et réseaux neuronaux : application à l'analyse de fiabilité." Paris 11, 1988. http://www.theses.fr/1988PA112201.
Повний текст джерелаMinaburo, Villar Ana Carolina. "Compression des en-têtes sur les réseaux bas-débit." Rennes 1, 2003. http://www.theses.fr/2003REN10144.
Повний текст джерелаMakkaoui, Leila. "Compression d'images dans les réseaux de capteurs sans fil." Phd thesis, Université de Lorraine, 2012. http://tel.archives-ouvertes.fr/tel-00795503.
Повний текст джерелаMakkaoui, Leila. "Compression d'images dans les réseaux de capteurs sans fil." Electronic Thesis or Diss., Université de Lorraine, 2012. http://www.theses.fr/2012LORR0416.
Повний текст джерелаThe increasing development of Wireless Camera Sensor Networks today allows a wide variety of applications with different objectives and constraints. However, the common problem of all the applications of sensor networks remains the vulnerability of sensors nodes because of their limitation in material resources, the most restricting being energy. Indeed, the available wireless technologies in this type of networks are usually a low-power, short-range wireless technology and low power hardware resources (CPU, battery). So we should meet a twofold objective: an efficient solution while delivering outstanding image quality on reception. This thesis concentrates mainly on the study and evaluation of compression methods dedicated to transmission over wireless camera sensor networks. We have suggested a new image compression method which decreases the energy consumption of sensors and thus maintains a long network lifetime. We evaluate its hardware implementation using experiments on real camera sensor platforms in order to show the validity of our propositions, by measuring aspects such as the quantity of memory required for the implantation program of our algorithms, the energy consumption and the execution time. We then focus on the study of the hardware features of our proposed method of synthesis of the compression circuit when implemented on a FPGA and ASIC chip prototype
Casasnovas, Béatrice. "Différenciation et maturation des réseaux neuronaux au cours de l'ontogenèse : étude anatomique, électrophysiologique et pharmacologique du système nerveux stomatogastrique des crustacés." Bordeaux 1, 1996. http://www.theses.fr/1996BOR10569.
Повний текст джерела