De nouvelles preuves suggèrent que les dysfonctionnements des circuits sous-jacents aux symptômes et aux déficits cognitifs des maladies psychiatriques pourraient être causés par une altération des paramètres d'équilibre d’excitation/inhibition (E/I). Cependant, les preuves physiologiques directes de cette hypothèse à partir de données électrophysiologiques et de neuro-imagerie non invasives sont jusqu'à présent rares. Pour apporter un soutien supplémentaire à l’hypothèse de l’équilibre E/I, la présente étude a appliqué une approche avancée de microscopie holographique numérique (MHN) pour examiner la dynamique des systèmes excitateurs/inhibiteurs suite à une stimulation glutamatergique dans des réseaux de neurones à différents stades de maturation neuronale. Cette approche fournissant une mesure approximative très précise des variations de mouvement de l’eau dans les cellules permet d’étudier certains processus physiologiques, tels que ceux reliés à l’activité neuronale. Cette étude a ainsi permis d’améliorer les connaissances sur la dynamique de la réponse neuronale induite par le glutamate, notamment en la caractérisant dans des cultures de neurones corticaux primaires de rats postnataux. L’activation des neurones engendrée par le glutamate, le principal neurotransmetteur excitateur, a révélé des changements plus ou moins persistants de la morphologie et des propriétés intracellulaires des neurones. De plus, les différentes réponses obtenues indiquent que le glutamate engendre des mécanismes d’activation et des processus de régulation du volume neuronal distincts d’un neurone à l’autre, probablement dépendant de l’état d’excitabilité de ce dernier qui résulte de l’interaction complexe des neurones inhibiteurs et excitateurs. Ainsi, la régulation de l’équilibre E/I de réseaux neuronaux pourrait potentiellement être reflétée par la proportion des différentes réponses de phase induites lors de stimulation de réseaux neuronaux au glutamate. / New evidences suggest that circuit dysfunctions underlying symptoms and cognitive deficits of psychiatric disorders may be caused by impaired excitation/inhibition equilibrium parameters (E/I). However, direct physiological evidences supporting this hypothesis from non-invasive electrophysiological and neuroimaging remain scarce. To provide additional support concerning the E/I balance hypothesis, this study uses an advanced digital holographic microscopy (DHM) approach to explore the dynamics of excitatory/inhibitory systems following glutamatergic stimulation in neural networks at different stages of neuronal maturation. This approach provides a very accurate approximate measurement of the water movement variations in cells allowing to study certain specific physiological processes, such as those related to neuronal activity. This study improves the knowledge regarding the dynamics of the glutamate-induced neuronal response, especially by characterizing it in cultures of primary cortical neurons of postnatal rats. The activation of neurons induced by glutamate, which is the main excitatory neurotransmitter, revealed more or less permanent changes in the morphology and intracellular properties of neurons. Moreover, the various responses obtained indicate that glutamate generates different neuronal activation mechanisms and neuronal volume regulation processes from a neuron to another, probably depending to the excitability state of the neuron that results from the complex interaction of inhibitory and excitatory neurons. Thus, the E/I balance regulation of neural networks could potentially be reflected by the proportion of different phase responses induced during glutamate neural network stimulation.

L'apprentissage et la mémoire aux niveaux cellulaire et moléculaire se caractérisent par la modulation de la force synaptique en recrutant et relocalisant des protéines synaptiques à l'échelle nanométrique. La plupart des études portant sur les mécanismes de la plasticité synaptique se sont concentrées sur des synapses spécifiques, manquant ainsi d'une vue d'ensemble de la diversité des changements de force synaptique et de la réorganisation des protéines dans les circuits neuronaux. Nous utilisons une combinaison d'imagerie fonctionnelle et à super résolution dans des cultures dissociées d'hippocampe et des outils d'intelligence artificielle pour classifier la diversité de synapses en fonction de leurs caractéristiques fonctionnelles et organisationnelles. Nous avons mesuré l'activité synaptique en utilisant la microscopie à grand champ pour enregistrer des événements calciques dans des neurones exprimant le senseur calcique fluorescent GCaMP6f. Nous avons développé une approche d'apprentissage profond pour détecter et segmenter ces événements calciques. Nous montrons la modulation de l'amplitude et de la fréquence des événements calciques en fonction de l'activité neuronale. En outre, nous avons classifié les synapses actives et nous avons identifié un recrutement différentiel de certains types de synapses en fonction du paradigme de plasticité utilisé. Comme l'organisation des protéines synaptiques à l'intérieur de domaines nanométriques des synapses joue un rôle central dans la force et la plasticité synaptiques, nous résolvons l'organisation des protéines d'échafaudage présynaptiques (Bassoon, RIM1/2) et postsynaptiques (PSD95, Homer1c) en utilisant la nanoscopie STED (Déplétion par émission stimulée). Nous avons quantifié l'organisation synaptique à l'aide d'une analyse statistique de la distance entre objets basée sur Python (pySODA). Nous montrons que les stimuli induisant la plasticité modifient de manière différentielle l'organisation de ces protéines. En particulier, les protéines PSD95 et Bassoon présentent un changement d'organisation dépendant d'un traitement induisant une potentiation synaptique ou une dépression synaptique. De plus, à l'aide d'approches d'apprentissage automatique non supervisées, nous révélons la riche diversité des sous-types de protéines synaptiques présentant un remodelage différentiel. Pour étudier le lien entre l'architecture des protéines synaptiques et la force synaptique, nous avons combiné l'imagerie fonctionnelle et l'imagerie à super-résolution. Nous avons donc utilisé une approche d'apprentissage automatique pour optimiser les paramètres d'imagerie des cellules vivantes pour l'imagerie à haute résolution et nous avons combiné cela avec l'optimisation des paramètres de déblocage du glutamate pour sonder les signaux calciques correspondants. Notre approche permet de caractériser la population de synapses en fonction de leur taux d'activité et de leur organisation de protéines synaptiques et devrait fournir une base pour explorer davantage les divers mécanismes moléculaires de la plasticité synaptique. / Learning and memory at the cellular and molecular levels are characterized by modulation of synaptic strength, involving the recruitment and re-localization of proteins within specific nanoscale synaptic domains. Most studies investigating the mechanisms of synaptic plasticity have been focussed on specific synapses, lacking a broad view of the diversity of synaptic changes in strength and protein re-organization across neural circuits. We use a combination of functional and super-resolution optical imaging in dissociated hippocampal cultures and artificial intelligence tools to classify the diversity of synapses, based on their functional and organizational characteristics. We measured synaptic activity using wide field microscopy to record miniature synaptic calcium transients (MSCTs) in neurons expressing the fluorescent calcium sensor GCaMP6f. We developed a deep learning approach to detect and segment these calcium events. Our results show that the amplitude and frequency of miniature calcium events are modulated by prior levels of circuit activity. In addition, we classified active synapses and identify differential recruitment of certain calcium dynamics depending on the plasticity paradigm used. To link the nanoscale organization of synaptic proteins with synaptic strength and plasticity, we optically resolved the organization of presynaptic (Bassoon, RIM1/2) and postsynaptic (PSD95, Homer1c) scaffolding proteins using STED (Stimulated Emission Depletion) nanoscopy. Using Python-based statistical object distance analysis (pySODA), we show that plasticity-inducing stimuli differentially alter the spatial organization of these proteins. In particular, PSD95 and Bassoon proteins show a treatment-dependent change in organization, associated either with synaptic potentiation or depression. Furthermore, using unsupervised machine learning approaches, we reveal the rich diversity of synaptic protein subtypes exhibiting differential remodeling. To investigate further the link between synaptic protein architecture and synaptic function, we aimed to combine functional and super-resolution imaging. We therefore used a machine learning approach to optimize live-cell imaging parameters for time-lapse imaging and combined this with the optimization of glutamate uncaging parameters to probe corresponding calcium signals. Our approach allows to characterize the population of synapses in terms of their activity rate and synaptic protein organization, providing a basis for further exploring the diverse molecular mechanisms of synaptic plasticity.

NOTICE EN COURS DE TRAITEMENT / L'imagerie cérébrale englobe un éventail de techniques qui permettent la collecte de données neurobiologiques abondantes présentant de l'hétérogénéité dans leur composition chimique. Pour analyser et décrire leur complexité, de nombreuses mesures morphométriques sont extraites afin de caractériser les événements observés. Cependant, sur la base de ces caractéristiques morphométriques, les données semblent souvent homogènes lors de l'analyse. Pour saisir et comprendre la diversité de ces phénomènes biologiques, nous avons choisi d'utiliser des méthodes computationnelles, notamment la réduction de dimension des données et le regroupement. Dans cette thèse, nous présenterons deux exemples d'application. La première partie est consacrée à l'étude de l'hétérogénéité des cellules en migration dans le cerveau en fonction de leur dynamique migratoire. La migration cellulaire est un phénomène important dans le développement du cerveau, notamment dans le cadre des troubles neurodéveloppementaux. Les précurseurs neuronaux, appelés neuroblastes, changent de formes lors de leur migration. Il existe deux phases pour ce processus, une phase stationnaire et une phase migratoire. L'objectif de cette étude est de déterminer si ces populations de neuroblastes peuvent être séparées sur la base de leurs propriétés migratoires mais également d'utiliser des méthodes d'analyses statistiques pour trouver les différentes sous-populations afin de déterminer lesquelles sont communes. Enfin, nous avons étudié les propriétés migratoires de ces différentes populations des neuroblastes en venant perturber la migration à l'aide de modifications génétique ou environnementale. La seconde partie porte sur l'étude de la plasticité structurelle, qui fait référence à la capacité qu'ont deux neurones à former une connexion, appelée synapse, qui peut se renforcer ou s'affaiblir. Ces changements synaptiques sont essentiels pour les processus d'apprentissage et de mémoire. En examinant des images de dendrites du bulbe olfactif prises avec un microscope confocal, on observe des protrusions sur la surface de la dendrite qui servent à recevoir les entrées synaptiques. Pour analyser ces images, nous avons développé un pipeline computationnel destiné à prétraiter les images et extraire les épines dendritiques. À la suite de la reconstruction 3D de la dendrite, nous avons extrait les épines et calculé plusieurs métriques, telles que la longueur et la surface de l'épine, des indicateurs couramment utilisés dans l'analyse des épines dendritiques. En procédant à une réduction de la dimensionnalité du jeu de données et à son partitionnement, nous avons relié la morphologie de chacune de ces sous-populations à leurs propriétés structurelles. Enfin, nous avons comparé le groupe contrôle et le groupe expérimental dans le cas de trois expériences olfactives, deux tâches de renforcement, et une de déprivation, qui ont conduit à des changements de plasticité. Les résultats montrent que la morphologie des épines ou leurs densités sont affectées par ces différentes conditions. En résumé, nous avons développé des outils computationnels permettant de révéler l'hétérogénéité des neurones en développement en fonction de leur dynamique migratoire et de leurs propriétés structurelles. / Brain imaging encompasses a range of techniques that enable the collection of abundant neurobiological data that presents heterogeneity in their chemical composition. To analyse and describe their complexity, numerous morphometric metrics are extracted to characterise the observed events. However, based on these morphometric features, the data often appear homogeneous during analysis. To grasp and understand the diversity of these biological phenomena, we have chosen to use computational methods including dimension reduction of data and clustering. In this thesis, we will present two application examples. The first part is devoted to the study of the heterogeneity of migrating neuronal cells based on their migratory dynamics. Cell migration is an important phenomenon in brain development, particularly in the context of neurodevelopmental disorders. Neuronal precursors, called neuroblasts, change shape during their migration. There are two phases for this process, so-called stationary phase and a migratory phase. The aim of this study is to determine whether neuroblasts can be separated to different sub-populations based on their migratory properties and to use statistical analysis methods to find the different subpopulations and determine which ones are common. Finally, we have studied the migratory properties of these different neuroblast populations by disrupting migration using genetic or environmental modifications. The second part focuses on the study of synaptic plasticity, which refers to the capacity of two neurons to form a connection, called a synapse, which can strengthen or weaken. These changes are central to the synaptic remodelling that occurs during the learning and memory phase. From images of dendrites, taken with a confocal microscope in the olfactory bulb, we have set up a computational pipeline to perform image pre-processing and then extract dendritic spines, which are protrusions on the surface of the dendrite that serve to receive synaptic inputs. After 3D reconstruction of the dendrite, these spines are extracted, and several metrics are calculated, including the length and surface area of the spine, which are standard metrics in spine analysis. After dimension reduction of the dataset and clustering, we have linked the morphology of each of these subpopulations to their structural properties. Finally, we have compared the control group and the experimental group in the case of three experiments that led to plasticity changes. The results show that the morphology of spines or their densities are affected by these different conditions. In summary, we have developed computational tools that reveal the heterogeneity of developing neurons based on their migratory dynamics and structural properties.

In the realm of natural language programming, translating free-form sentences in natural language into a functional, machine-executable program remains difficult due to the following 4 challenges. First, the inherent ambiguity of natural languages. Second, the high-level verbose nature in user descriptions. Third, the complexity in the sentences and Fourth, the invalid or semantically unclear sentences. Our first solution is a Large Language Model (LLM) based Artificial Intelligence driven assistant to process free-form sentences and decompose them into sequences of simplified, unambiguous sentences that abide by a set of rules, thereby stripping away the complexities embedded within the original sentences. These resulting sentences are then used to generate the code. We applied the proposed approach to a set of free-form sentences written by middle-school students for describing the logic behind video games. More than 60% of the free-form sentences containing these problems were sufficiently converted to sequences of simple unambiguous object-oriented sentences by our approach. Next, the thesis also presents "IntentGuide," a neuro-symbolic integration framework to enhance the clarity and executability of human intentions expressed in freeform sentences. IntentGuide effectively integrates the rule-based error detection capabilities of symbolic AI with the powerful adaptive learning abilities of Large Language Model to convert ambiguous or complex sentences into clear, machine-understandable instructions. The empirical evaluation of IntentGuide performed on natural language sentences written by middle school students for designing video games, reveals a significant improvement in error correction and code generation abilities compared to previous approach, attaining an accuracy rate of 90%. / Master of Science / Imagine if you could talk to machines in everyday language and they could understand exactly what you meant, turning your words into programs that do exactly what you describe. That's the goal of the thesis. We've developed a system that helps machines make sense of the kind of free-form language that people, especially students, use when they describe what they want a computer to do. Understanding and converting everyday language into computer code is a complex challenge, primarily because the way we naturally speak can be vague, overly detailed, or just complex. This thesis presents a new tool using artificial intelligence that helps break down and simplify these sentences. By transforming them into clearer, rulefollowing instructions, this tool makes it easier for machines to understand and execute the tasks we describe. The technology was tested using descriptions from middle-school students on how video games should work. Over 60% of these complex or unclear descriptions were sufficiently converted into straightforward instructions that a machine could use. Additionally, a new system called "IntentGuide" was introduced, combining traditional AI methods with advanced language models to improve how effectively machines can interpret and act on human instructions. This improved system showed a 90% accuracy in understanding and correcting errors in the students' game descriptions, marking a significant step forward in helping computers better understand us.

Titre de l'écran-titre (visionné le 29 novembre 2023) / L'apprentissage automatique a considérablement progressé dans le domaine de la recherche en neurosciences, mais son application pose des défis en raison des différences entre les principes biologiques du cerveau et les méthodes traditionnelles d'apprentissage automatique. Dans ce contexte, le projet présenté propose NeuroTorch, un pipeline convivial d'apprentissage automatique spécialement conçu pour les neuroscientifiques, afin de relever ces défis. Les objectifs clés de ce projet sont de fournir une librairie d'apprentissage profond adaptée aux neurosciences computationnelles, d'implémenter l'algorithme eligibility trace forward propagation (e-prop) pour sa plausibilité biologique, de comparer les réseaux de neurones continus et à impulsions en termes de résilience, et d'intégrer un pipeline d'apprentissage par renforcement. Le projet se divise en plusieurs parties. Tout d'abord, la théorie des dynamiques neuronales, des algorithmes d'optimisation et des fonctions de transformation d'espaces sera développée. Ensuite, l'attention sera portée sur la conception du pipeline NeuroTorch, incluant l'implémentation de l'algorithme e-prop. Les résultats de la prédiction de séries temporelles d'activité neuronale chez le poisson-zèbre seront présentés, ainsi que des observations sur la résilience à l'ablation des réseaux obtenus. Enfin, une section sera consacrée à l'exploration du pipeline d'apprentissage par renforcement de NeuroTorch et à la validation de son architecture dans l'environnement LunarLander de Gym. En résumé, les modèles à impulsions de NeuroTorch ont atteint des précisions de 96,37%, 85,58% et 74,16% respectivement sur les ensembles de validation MNIST, Fashion-MNIST et Heidelberg. De plus, les dynamiques leaky-integrate-and-fire with explicit synaptic current - low pass filter (SpyLIF-LPF) et Wilson-Cowan ont été entraînées avec succès à l'aide de l'algorithme e-prop sur des données neuronales expérimentales du ventral habenula du poisson-zèbre, obtenant respectivement des valeurs de pVar de 0,97 et 0,96. Les résultats concernant la résilience indiquent que l'application de la loi de Dale améliore la robustesse des modèles en termes d'ablation hiérarchique. Enfin, grâce au pipeline d'apprentissage par renforcement de NeuroTorch, différents types d'agents inspirés des neurosciences ont atteint le critère de réussite dans l'environnement LunarLander de Gym. Ces résultats soulignent la pertinence et l'efficacité de NeuroTorch pour les applications en neurosciences computationnelles. / Machine learning has made significant advancements in neuroscience research, but its application presents challenges due to the differences between the biological principles of the brain and traditional machine learning methods. In this context, the presented project proposes NeuroTorch, a comprehensive machine learning pipeline specifically designed for neuroscientists to address these challenges. The key objectives of this project are to provide a deep learning library tailored to computational neuroscience, implement the eligibility trace forward propagation (e-prop) algorithm for biological plausibility, compare continuous and spiking neural networks in terms of resilience, and integrate a reinforcement learning pipeline. The project is divided into several parts. Firstly, the theory of neural dynamics, optimization algorithms, and space transformation functions will be developed. Next focus will be on the design of the NeuroTorch pipeline, including the implementation of the e-prop algorithm. Results of predicting a time series of neuronal activity in zebrafish will be presented, along with observations on the resilience to network ablations obtained. Finally, a section will be dedicated to exploring the NeuroTorch reinforcement learning pipeline and validating its architecture in the LunarLander environment of Gym. In summary, NeuroTorch spiking models achieved accuracies of 96.37%, 85.58%, and 74.16% on the MNIST, Fashion-MNIST, and Heidelberg validation sets, respectively. Furthermore, the leaky-integrate-and-fire with explicit synaptic current - low pass filter (SpyLIF-LPF) and Wilson-Cowan dynamics were successfully trained using the e-prop algorithm on experimental neuronal data from the ventral habenula of zebrafish, achieving pVar values of 0.97 and 0.96, respectively. Results regarding resilience indicate that the application of the Dale law improves the robustness of models in terms of hierarchical ablation. Lastly, through the NeuroTorch reinforcement learning pipeline, different types of neuroscience-inspired agents successfully met the success criterion in the Gym's LunarLander environment. These results highlight the relevance and effectiveness of NeuroTorch for applications in computational neuroscience.

The diploma thesis is focused on the concept of the psychomotor development in context of remaining primary reflexes in children. The author draws on the knowledge of developmental psychology and neurophysiology that some of developmentally earlier stages must under optimal circumstances and maturation of a child disappear and be replaced by ontogenetically newer forms. Research confirms continuity of persistent primary reflexes, psychomotor development and learning and behavioral problems. The thesis aims to investigate whether inhibition of primary reflexes using Neuro-developmental therapy improves the condition of children with delayed psychomotor development, including their school grades. The author in the introductory chapters focusses on normal psychomotor development of children, in particular gross motor skills, fine motor skills, sensory skills, including proprioception and kinesthesia and their importance for learning ability in school age. The next part of the thesis is focused on particular primary reflexes and their impact on psychomotor development. The author pays special attention to the climbing stage in infancy which she considers as an important milestone in the optimal psychomotor development. Case studies shows, that skipping any developmental stage can be considered as a...

La sclérose en plaques (SEP) est une maladie neuro-inflammatoire chronique, première cause de handicap chez le jeune adulte. Actuellement, aucun traitement ne freine l'aggravation des symptômes liée aux formes progressives. Bien que connue, l'implication des macrophages et de la microglie dans la démyélinisation et l'atteinte axonale doit être plus finement caractérisée. Ce d'autant plus que la plasticité fonctionnelle de ces cellules suggère une réponse spécifique selon la pathologie, la localisation des lésions et le stade évolutif de la maladie. Ce travail de thèse a consisté en une caractérisation moléculaire des programmes d'activation macrophagique/ microgliale dans deux types d'altérations tissulaires du système nerveux central des patients SEP : les zones partiellement démyélinisées bordant les plaques de la moelle épinière et les lésions corticales. Cette étude a été réalisée sur des tissus post-mortem de patients atteints de formes progressives, formes dans lesquelles les lésions médullaires et corticales sont nombreuses et impliquées dans le handicap progressif et irréversible. Nous avons identifié des spécificités moléculaires caractérisant l'activation macrophagique/microgliale au cours de la SEP en comparant, par une approche in silico, les profils caractérisés à ceux observés dans des pathologies neuro-dégénératives à composantes inflammatoires, la maladie d'Alzheimer et de Parkinson notamment. Dans l'ensemble, ces résultats suggèrent que l'activation chronique des macrophages/cellules microgliales contribue à l'extension à bas bruit des lésions médullaires et corticales pendant la phase progressive de la SEP et proposent de nouvelles cibles thérapeutiques / Multiple sclerosis (MS) is a chronic neuro-inflammatory disease and the most common cause of chronic neurological disability in young adults. No treatment is currently available to prevent the aggravation of symptoms in the progressive forms of the disease. The involvement of macrophages and microglia in demyelination and axonal injury is well known but need to be further characterized. Especially, the high level of functional plasticity harboured by macrophages/ microglia suggests that these cells engage specific activation programs depending on the disease, its evolution stage and the localization of lesions. In this context, this phD thesis was essentially aimed to characterize macrophage/microglia activation programs in two categories of tissue alterations observed in the post-mortem central nervous system from MS patients: 1) partially demyelinated areas at the border of spinal cord plaques and 2) cortical lesions. These two particular types of lesions are both highly frequent in progressive forms of MS and suspected to be involved in chronic and irreversible neurological disability. Using an in silico approach, the macrophage/microglia activation programs identified in MS were then compared to those observed in neurodegenerative and inflammatory disorders such as Alzheimer's disease and Parkinson's disease. Overall, our results suggest that the chronic activation of macrophages and microglia largely contributes to the slow and chronic expansion of MS lesions in progressive forms of the disease. Our work also proposes new therapeutic targets

Uma vez que sistemas de inferência neuro-fuzzy adaptativos são aproximadores universais que podem ser usados em aplicações de aproximação de funções e de previsão, este trabalho tem por objetivo determinar seus melhores parâmetros e suas melhores arquiteturas com o propósito de se executar previsão de demanda de energia elétrica no curtíssimo prazo em subestações de distribuição. Isto pode possibilitar o desenvolvimento de controles automáticos de carga mais eficientes para sistemas elétricos de potência. As entradas do sistema são séries temporais de demanda de energia elétrica, compostas por dados mensurados em intervalos de cinco minutos ao longo de sete dias em subestações localizadas em cidades do interior do estado de São Paulo. Diversas configurações de entrada e diferentes arquiteturas foram examinadas para se fazer a previsão de um passo a frente. Os resultados do sistema de inferência neuro-fuzzy adaptativo frente às abordagens encontradas na literatura foram promissores. / Since adaptive neuro-fuzzy inference systems are universal approximators that can be used in functions approximation and forecasting applications, this work has the objective to determine their best parameters and best architectures with the purpose to execute very short term load forecasting in distribution substations. This can allow the development of more efficient load automatic control for power systems. The system inputs are load demand time series, which are composed of data measured at each five minutes interval, during seven days, from substations located in cities from São Paulo state countryside. Several input configurations and different architectures were examined to make a prediction aiming one step forecasting. The adaptive neuro-fuzzy inference system results in comparison with other approaches found in literature were promising.

[pt] Esta dissertação investiga uma nova metodologia, baseada em técnicas inteligentes, para a redução das perdas comerciais relativas ao fornecimento de energia elétrica. O objetivo deste trabalho é apresentar um modelo de inteligência computacional capaz de identificar irregularidades na medição de demanda e consumo de energia elétrica, considerando as características sazonais não lineares das curvas de carga das unidades consumidoras, características essas que são difíceis de se representar em modelos matemáticos. A metodologia é baseada em três etapas: categorização, para agrupar unidades consumidoras em classes similares; classificação para descobrir relacionamentos que expliquem o perfil da irregularidade no fornecimento de energia elétrica e que permitam prever a classe de um padrão desconhecido; e extração de conhecimento sob a forma de regras fuzzy interpretáveis. O modelo resultante foi denominado Sistema de Classificação de Unidades Consumidoras de Energia Elétrica. O trabalho consistiu em três partes: um estudo sobre os principais métodos de categorização e classificação de padrões; definição e implementação do Sistema de Classificação de Unidades Consumidoras de Energia Elétrica; e o estudo de casos. No estudo sobre os métodos de categorização foi feito um levantamento bibliográfico da área, resultando em um resumo das principais técnicas utilizadas para esta tarefa, as quais podem ser divididas em algoritmos de categorização hierárquicos e não hierárquicos. No estudo sobre os métodos de classificação foram feitos levantamentos bibliográficos dos sistemas Neuro-Fuzzy que resultaram em um resumo sobre as arquiteturas, algoritmos de aprendizado e extração de regras fuzzy de cada modelo analisado. Os modelos Neuro-Fuzzy foram escolhidos devido a sua capacidade de geração de regras lingüísticas. O Sistema de Classificação de Unidades Consumidoras de Energia Elétrica foi definido e implementado da seguinte forma: módulo de categorização, baseado no algoritmo Fuzzy C-Means (FCM); e módulo de classificação baseado nos Sistemas Neuro-Fuzzy NEFCLASS e NFHB-Invertido. No primeiro módulo, foram utilizadas algumas medidas de desempenho como o FPI (Fuzziness Performance Index), que estima o grau de nebulosidade (fuziness) gerado por um número específico de clusters, e a MPE (Modified Partition Entropy), que estima o grau de desordem gerado por um número específico de clusters. Para validação do número ótimo de clusters, aplicou-se o critério de dominância segundo o método de Pareto. No módulo de classificação de unidades consumidoras levou-se em consideração a peculiaridade de cada sistema neuro-fuzzy, além da análise de desempenho comparativa (benchmarking) entre os modelos. Além do objetivo de classificação de padrões, os Sistemas Neuro-Fuzzy são capazes de extrair conhecimento em forma de regras fuzzy interpretáveis expressas como: SE x é A e y é B então padrão pertence à classe Z. Realizou-se um amplo estudo de casos, abrangendo unidades consumidoras de atividades comerciais e industriais supridas em baixa e média tensão. Os resultados encontrados na etapa de categorização foram satisfatórios, uma vez que as unidades consumidoras foram agrupadas de forma natural pelas suas características de demanda máxima e consumo de energia elétrica. Conforme o objetivo proposto, esta categorização gerou um número reduzido de agrupamentos (clusters) no espaço de busca, permitindo que o treinamento dos sistemas Neuro-Fuzzy fosse direcionado para o menor número possível de grupos, mas com elevada representatividade sobre os dados. Os resultados encontrados com os modelos NFHB-Invertido e NEFCLASS mostraram-se, na maioria dos casos, superiores aos melhores resultados encontrados pelos modelos matemáticos comumente utilizados. O desempenho dos modelos NFHB-Invertido e NEFCLASS, em relação ao te / [en] This dissertation investigates a new methodology based on intelligent techniques for commercial losses reduction in electrical energy supply. The objective of this work is to present a model of computational intelligence able to identify irregularities in consumption and demand electrical measurements, regarding the non-linearity of the consumers seasonal load curve which is hard to represent by mathematical models. The methodology is based on three stages: clustering, to group consumers of electric energy into similar classes; patterns classification, to discover relationships that explain the irregularities profile and that determine the class for an unknown pattern; and knowledge extraction in form of interpretable fuzzy rules. The resulting model was entitled Electric Energy Consumers Classification System. The work consisted of three parts: a bibliographic research about main methods for clustering and patterns classification; definition and implementation of the Electric Energy Consumers Classification System; and case studies. The bibliographic research of clustering methods resulted in a survey of the main techniques used for this task, which can be divided into hierarchical and non-hierarchical clustering algorithms. The bibliographic research of classification methods provided a survey of the architectures, learning algorithms and rules extraction of the neuro-fuzzy systems. Neuro-fuzzy models were chosen due to their capacity of generating linguistics rules. The Electric Energy Consumers Classification System was defined and implemented in the following way: a clustering module, based on the Fuzzy CMeans (FCM) algorithm; and classification module, based on NEFCLASS and Inverted-NFHB neuro-fuzzy sytems. In the first module, some performance metrics have been used such as the FPI (Fuzziness Performance Index), which estimates the fuzzy level generated by a specific number of clusters; and the MPE (Modified Partition Entropy) that estimates disorder level generated by a specific number of clusters. The dominance criterion of Pareto method was used to validate optimal number of clusters. In the classification module, the peculiarities of each neuro-fuzzy system as well as performance comparison of each model were taken into account. Besides the patterns classification objective, the neuro-Fuzzy systems were able to extract knowledge in form of interpretable fuzzy rules. These rules are expressed by: IF x is A and y is B then the pattern belongs to Z class. The cases studies have considered industrial and commercial consumers of electric energy in low and medium tension. The results obtained in the clustering step were satisfactory, since consumers have been clustered in a natural way by their electrical consumption and demand characteristics. As the proposed objective, the system has generated an optimal low number of clusters in the search space, thus directing the learning step of the neuro-fuzzy systems to a low number of groups with high representation over data. The results obtained with Inverted-NFHB and NEFCLASS models, in the majority of cases, showed to be superior to the best results found by the mathematical methods commonly used. The performance of the Inverted-NFHB and NEFCLASS models concerning to processing time was also very good. The models converged to an optimal classification solution in a processing time inferior to a minute. The main objective of this work, that is the non- technical power losses reduction, was achieved by the assertiveness increases in the identification of the cases with measuring irregularities. This fact made possible some reduction in wasting with workers and effectively improved the billing.

Cette thèse s'inscrit dans le contexte de l'étude des circuits neuromorphiques utilisant des dispositifs memristifs comme synapses. Son objectif principal est d'évaluer les mérites d'une nouvelle classe de mémoires organiques développées au LICSEN (CEA Saclay/IRAMIS) et, plus particulièrement, leur adéquation avec les propositions d'implémentation et les règles d'apprentissage proposées par l'équipe NanoArchi de l'IEF (Univ. Paris-Sud, Orsay). Les memristors étudiés sont basés sur l'electro-greffage en films minces de complexes organiques redox pour la formation de jonctions métal/molécules/métal robustes et scalables. Outre la fabrication de memristors, le travail inclut d'importants efforts de caractérisation électrique (vitesse, non-volatilité, scalabilité, robustesse, etc.) visant d'une part à étudier les mécanismes de commutation dans ces nouveaux matériaux memristifs organiques, et d'autres part, à évaluer leur potentiel en tant que synapses. Cette thèse présente également une étude préparatoire à la réalisation d'un démonstrateur de circuit mixte de type réseaux de neurones combinant nano-memristors et électronique conventionnelle (programmabilité des dispositifs en mode impulsionnel, réalisation d'assemblées de dispositifs, variabilité). De plus, la démonstration de la compatibilité de ces memristors avec la propriété STDP (Spike Timing Dependent Plasticity) ainsi que de l’apprentissage d’un « réflexe conditionné » ouvrent la voie aux apprentissages non-supervisés. / This PhD project takes place in the context of the study of neuromorphic circuits using memristor devices as synapses. The main objective is to evaluate a new class of organic memories developed at LICSEN (CEA Saclay/IRAMIS) and particularly their compatibility with the learning rules and the implementation strategy proposed by the Nanoarchi group at IEF (Univ. Paris-Sud, Orsay). These new memristors are based on the electro-grafting of organic redox complexes thin films to form robust and scalable metal/molecules/metal junctions. In addition to memristor fabrication, this work includes detailed electrical characterization studies (speed, retention property, scalability, robustness, etc.) aiming at, on the one hand, establishing the commutation mechanism in these new memristors and, on the other hand, evaluating their potential as synapses. This work also proposes a preparatory study of a neural-network type mixed-circuit demonstrator combining nano-memristors and conventional electronic (programmability of devices by spikes, fabrication of assemblies of memristors, variability). Moreover the demonstration of the compatibility of such memristors with the STDP (Spike Timing Dependent Plasticity) property and of the learning of a “conditioned reflex” opens the way to future unsupervised learning studies.