Les nano-oscillateurs à transfert de spin sont des composants radiofréquences magnétiques non-linéaires, nanométrique, de faible consommation en énergie et accordables en fréquence. Ce sont aussi potentiellement des candidats prometteurs pour l’élaboration de larges réseaux d’oscillateurs couplés. Ces derniers peuvent être utilisés dans les architectures neuromorphiques qui nécessitent des assemblées très denses d’unités de calcul complexes imitant les neurones biologiques et comportant des connexions ajustables entre elles. L’approche neuromorphique permet de pallier aux limitations des ordinateurs actuels et de diminuer leur consommation en énergie. En effet pour résoudre des tâches cognitives telles que la reconnaissance vocale, le cerveau fonctionne bien plus efficacement en terme d’énergie qu’un ordinateur classique. Au vu du grand nombre de neurone dans le cerveau (100 milliards) une puce neuro-inspirée requière des oscillateurs de très petite taille tels que les nano-oscillateurs à transfert de spin. Récemment, une première démonstration de calcul neuromorphique avec un unique nano-oscillateur à transfert de spin a été établie. Cependant, pour aller au-delà, il faut démontrer le calcul neuromorphique avec plusieurs nano-oscillateurs et pouvoir réaliser l’apprentissage. Une difficulté majeure dans l’apprentissage des réseaux de nano-oscillateurs est qu’il faut ajuster le couplage entre eux. Dans cette thèse, en exploitant l'accordabilité en fréquence des nano-oscillateurs magnétiques, nous avons démontré expérimentalement l'apprentissage des nano-oscillateurs couplés pour classifier des voyelles prononcées avec un taux de reconnaissance de 88%. Afin de réaliser cette tache de classification, nous nous sommes inspirés de la synchronisation des taux d’activation des neurones biologiques et nous avons exploité la synchronisation des nano-oscillateurs magnétiques à des stimuli micro-ondes extérieurs. Les taux de reconnaissances observés sont dus aux fortes accordabilités et couplage intermédiaire des nano-oscillateurs utilisés. Enfin, afin de réaliser des taches plus difficiles nécessitant de larges réseaux de neurones, nous avons démontré numériquement qu’un réseau d’une centaine de nano-oscillateurs magnétiques peut être conçu avec les contraintes standards de nano-fabrication. / Spin-torque nano-oscillators are non-linear, nano-scale, low power consumption, tunable magnetic microwave oscillators which are promising candidates for building large networks of coupled oscillators. Those can be used as building blocks for neuromorphic hardware which requires high-density networks of neuron-like complex processing units coupled by tunable connections. The neuromorphic approach allows to overcome the limitation of nowadays computers and to reduce their energy consumption. Indeed, in order to perform cognitive tasks as voice recognition or image recognition, the brain is much more efficient in terms of energy consumption. Due to the large number of required neurons (100 billions), a neuromorphic chip requires very small oscillators such as spin-torque nano-oscillators to emulate neurons. Recently a first demonstration of neuromorphic computing with a single spin-torque nano-oscillator was established, allowing spoken digit recognition with state of the art performance. However, to realize more complex cognitive tasks, it is still necessary to demonstrate a very important property of neural networks: learning an iterative process through which a neural network can be trained using an initial fraction of the inputs and then adjusting internal parameters to improve its recognition or classification performance. One difficulty is that training networks of coupled nano-oscillators requires tuning the coupling between them. Here, through the high frequency tunability of spin-torque nano-oscillators, we demonstrate experimentally the learning ability of coupled nano-oscillators to classify spoken vowels with a recognition rate of 88%. To realize this classification task, we took inspiration from the synchronization of rhythmic activity of biological neurons and we leveraged the synchronization of spin-torque nano-oscillators to external microwave stimuli. The high experimental recognition rates stem from the weak-coupling regime and the high tunability of spin-torque nano-oscillators. Finally, in order to realize more difficult cognitive tasks requiring large neural networks, we show numerically that arrays of hundreds of spin-torque nano-oscillators can be designed with the constraints of standard nano-fabrication techniques.

<p>Bio-inorganic structures have been found to exhibit impressive optical and mechanical properties, such as control of light and/or high fracture strength. Certain species of diatoms (single-celled algae) form siliceous microshells (frustules) with organized structures that affect the transmission of light or fracture strengths. It has been found that <i>Coscinodiscus wailesii</i> diatoms have frustules with a quasi-regular hexagonal pattern of pores, which act as micro-lenses. In terms of mechanical strength, <i>Fragilariopsis kerguelensis</i> diatom SiO₂ frustules have been observed to exhibit impressive compressive and tensile fracture stress values. In this study, shape-preserving chemical conversion (using gas/solid reactions) is used to transform biogenic structures (diatom frustules) into high IR refractive index or ultrahigh specific strength materials. High-fidelity MgO/Si, Mg₂Si, Ca₂Si, MgO/Ti, and Ti replicas are successfully synthesized and characterized by SEM, EDX, XRD, and TEM. Focal point imaging experiments are used to show that focusing behavior of MgO/Si and Mg₂Si replicas can be enhanced in the IR range upon conversion into higher index replicas. Mechanical properties of SiO₂ frustules, MgO/Ti replicas, and Ti replicas have been measured by using in-situ and ex-situ indentation, which revealed that the mechanical properties can be enhanced by the shape-preserved chemical conversion of Bio-inorganic structures.</p><p><br></p>

L’un des principaux verrous scientifique rencontré au cours du développement de catalyseurs d’oxydation bio-inspirés concerne l’étape de réduction du métal pour permettre l’activation du dioxygène. Pour essayer de lever ce verrou, nous avons développé un système macromoléculaire composé d’un polymère hydrosoluble dans lequel deux types de cofacteurs sont incorporés ; (1) des cofacteurs redox capables de collecter des électrons issus d’un réducteur en solution, et (2) des cofacteurs catalytiques capables d’activer le dioxygène. De façon à permettre l’incorporation de ces cofacteurs au sein du polymère, ce dernier a été modifié avec différents groupement chimiques qui ont étés quantifiés par RMN du proton dans l’eau. Par ailleurs, la synthèse de différents complexes métalliques, connus pour être de bons catalyseurs d’oxydation, tels que des métalloporphyrines ou des complexes mononucléaire et binucléaire de fer et de cuivre, a été réalisée. Ces catalyseurs ont d’abord été étudiés avec H2O2 dans l’eau en présence ou en absence de polymère. En parallèle, l’incorporation de la FMN par interactions électrostatiques au sein du polymère a permis de générer un système capable de collecter les électrons de NADH en solution. Cette réduction s’est avérée 4 000 fois plus rapide que la réduction sans polymère modifié. Cette réductase artificielle (FMN+ PEI modifié) a ensuite été démontrée capable de réduire très efficacement les porphyrines de manganèse (III) ainsi que d’autres complexes métalliques. Au cours de l’étude, la capacité de ce système à séparer les électrons provenant de NADH a également été mis en avant. Finalement, cette réductase artificielle a été associée à différents catalyseurs métalliques afin d’étudier leur activité sur la réaction d’oxydation du thioanisole, ou d’autres substrats, par activation réductrice du dioxygène / A major scientific lock encountered during the development of bio-inspired oxidation catalysts is the metal reduction step to allow activation of dioxygen. In this optic, we have developped a macromolecular system composed of a water-soluble polymer in which two kinds of cofactors are incorporated; (1) redox cofactors capable of collecting electrons from a reducing agent in solution, and (2) catalytic cofactors capable of activating oxygen. In order to allow the incorporation of these co-factors within the polymer, the latter one has been modified by various chemical groups which have been quantified by proton NMR in water. Furthermore, the synthesis of various metal complexes, known as good oxidation catalysts, such as metalloporphyrins or mononuclear and dinuclear complexes of iron and copper was performed.These catalysts were first studied with H2O2 in water in the presence or the absence of polymer. In parallel, the incorporation of FMN by electrostatic interactions within the polymer has generated a system capable of collecting the electrons of NADH in solution. This reduction was found 4 000 times faster than the reduction without modified polymer. This artificial reductase (FMN + PEI modified) was then demonstrated to very efficiently reduce manganese porphyrins as well as other metal complexes. During this study, the ability of the system to split electron pairs collected from NADH has also been demonstrated. Finally, this artificial reductase has been associated with various metal catalysts in order to study their catalytic activity for various oxidation reaction using dioxygen.

<div>An estimated 20 million tons of adhesives are used globally each year, and the amount is continually increasing. Glues are used in nearly every economic sector but are largely consumed by key external drivers of the industry including construction and transportation equipment to replace mechanical fasteners. Many of these applications require specific functionality, like moisture resistance, desirable mechanical properties, or low toxicity. However, specific features usually occur at the expense of adhesive strength, and there is no “one size fits all” adhesive. The search for more practical and stronger glues has contributed to the development of biomimetic adhesives. Marine mussels and other sea creatures produce biological adhesives that stick well underwater. By using nature as an inspiration for better glues, we can combine stronger bonding and additional functionality into one adhesive system. Introducing the same catechol moiety used by marine organisms into synthetic polymers has allowed us to produce adhesives stronger than commercial glues in both dry and wet environments.</div><div><br></div><div>While many of these biomimetic polymer adhesives have been prepared, few have made it to market. Here, multiple biomimetic polymer adhesives are studied and optimized for different applications to provide the next step towards commercialization. The adhesives were tailored for use on different surfaces and conditions through formulation or polymer design. Structure-function studies have showed how surface energy influences optimal adhesion with catechol-containing polymers for applications in bonding dissimilar substrates while maintaining desired mechanical properties. Multiple adhesive systems were studied in mice to assess toxicity and determine viability as potential surgical glues. Underwater formulation and application methods were also pursued to improve product development strategies for offering a competitive advantage as an underwater glue. In addition to these practical-use modifications of the adhesives, industry research and market analysis was conducted to provide insight into further applications to pursue. A cost analysis led to creating new synthetic strategies for cost-reduction and scale-up, both of which are essential in the commercialization of a catechol-containing polymer adhesive.<br></div>

La criticité du réchauffement climatique incite à chercher des solutions pour réduire les émissions de dioxyde de carbone (CO₂). Le développement de catalyseurs qui peuvent aider à capturer, activer, réduire et valoriser le CO₂ est au cœur de ce défi. Cette thèse a répondu à cet appel en développant des mimétismes moléculaires inspirés de la Nature, dans le cadre plus large de la photosynthèse artificielle. Au début il s'agissait de suivre le parcours d'un photon de lumière visible et de déterminer comment il peut réduire la molécule de CO₂. Ensuite afin de réaliser des catalyseurs plus efficaces, de nouvelles molécules ont été synthétisées en s’inspirant de l’enzyme CO déshydrogénase (CODH) qui présente des performances exceptionnelles pour la réduction du CO₂. Enfin, une autre propriété du CODH a conduit à une validation de principe pour la valorisation immédiate du CO photo-produit dans la synthèse des liaisons amides marqués, une caractéristique courante des médicaments. / The criticality of global warming urges for the advancement of science to reduce carbon dioxide (CO₂) emissions in the atmosphere. At the heart of this challenge is the development of sustainable catalysts that can help capture, activate, reduce, and eventually valorize CO₂. This PhD work tried to respond to this call by developing molecular mimics inspired by natural systems in the larger scheme of artificial photosynthesis. Firstly, it involved tracking the journey of a photon of visible light and how it is transformed to a reducing power able to reduce CO₂. Secondly, in search for more efficient and stable catalysts, new mimics were synthesized inspired by the exceptional performance of CO dehydrogenase enzymes (CODH) in reducing CO₂. Lastly, further understanding of CODH also led to a proof-of-concept that directly valorizes the photo-produced CO for the synthesis of isotopically-labelled amide bonds, a common motif in pharmaceutically-relevant drugs.

No description available.

Accurate models of visual system are key for understanding how our brains process visual information. In recent years, deep neural networks (DNN) have been rapidly gaining traction in this domain. However, only few studies attempted to incorporate known anatomical properties of visual system into standard DNN architectures adapted from the general machine learning field, to improve their interpretability and performance on visual data. In this thesis, we optimize a recent biologically inspired deep learning architecture designed for analysis of population data recorded from mammalian primary visual cortex when presented with natural images as stimuli. We reimplement this prior modeling in existing neuroscience focused deep learning framework NDN3 and assess it in terms of stability and sensitivity to hyperparameters and architecture fine-tuning. We proceed to extend the model with components of various DNN models, analysing novel combinations and techniques from classical computer vision deep learning, comparing their effectiveness against the bio-inspired components. We were able to identify modifications that greatly increase the stability of the model while securing moderate improvement in overall performance. Furthermore, we document the importance of small hyperparameters adjustments versus...

No description available.

The domain of Intrusion Detection Systems (IDS) has witnessed growing interest in recent years due to the escalating threats posed by cyberattacks. As Internet of Things (IoT) becomes increasingly integrated into our every day lives, we widen our attack surface and expose more of our personal lives to risk. In the same way the Human Immune System (HIS) safeguards our physical self, a similar solution is needed to safeguard our digital self. This thesis presents the Artificial Immune inspired Intrusion Detection System (AIS-IDS), an IDS modeled after the HIS. This thesis proposes an architecture for AIS-IDS, instantiates an AIS-IDS model for evaluation, conducts a robust set of experiments to ascertain the efficacy of the AIS-IDS, and answers key research questions aimed at evaluating the validity of the AIS-IDS. Finally, two expansions to the AIS-IDS are proposed with the goal of further infusing the HIS into AIS-IDS design.

Tesis por compendio / [ES] La utilización de sistemas para el tratamiento eficiente de grandes volúmenes de información ha crecido en popularidad durante los últimos años. Esto conlleva el desarrollo de nuevas tecnologías, métodos y algoritmos, que permitan un uso eficiente de las infraestructuras. El tratamiento de grandes volúmenes de información no está exento de numerosos problemas y retos, algunos de los cuales se tratarán de mejorar. Dentro de las posibilidades actuales debemos tener en cuenta la evolución que han tenido los sistemas durante los últimos años y las oportunidades de mejora que existan en cada uno de ellos. El primer sistema de estudio, el Grid, constituye una aproximación inicial de procesamiento masivo y representa uno de los primeros sistemas distribuidos de tratamiento de grandes conjuntos de datos. Participando en la modernización de uno de los mecanismos de acceso a los datos se facilita la mejora de los tratamientos que se realizan en la genómica actual. Los estudios que se presentan están centrados en la transformada de Burrows-Wheeler, que ya es conocida en el análisis genómico por su capacidad para mejorar los tiempos en el alineamiento de cadenas cortas de polinucleótidos. Esta mejora en los tiempos, se perfecciona mediante la reducción de los accesos remotos con la utilización de un sistema de caché intermedia que optimiza su ejecución en un sistema Grid ya consolidado. Esta caché se implementa como complemento a la librería de acceso estándar GFAL utilizada en la infraestructura de IberGrid. En un segundo paso se plantea el tratamiento de los datos en arquitecturas de Big Data. Las mejoras se realizan tanto en la arquitectura Lambda como Kappa mediante la búsqueda de métodos para tratar grandes volúmenes de información multimedia. Mientras que en la arquitectura Lambda se utiliza Apache Hadoop como tecnología para este tratamiento, en la arquitectura Kappa se utiliza Apache Storm como sistema de computación distribuido en tiempo real. En ambas arquitecturas se amplía el ámbito de utilización y se optimiza la ejecución mediante la aplicación de algoritmos que mejoran los problemas en cada una de las tecnologías. El problema del volumen de datos es el centro de un último escalón, por el que se permite mejorar la arquitectura de microservicios. Teniendo en cuenta el número total de nodos que se ejecutan en sistemas de procesamiento tenemos una aproximación de las magnitudes que podemos obtener para el tratamiento de grandes volúmenes. De esta forma, la capacidad de los sistemas para aumentar o disminuir su tamaño permite un gobierno óptimo. Proponiendo un sistema bioinspirado se aporta un método de autoescalado dinámico y distribuido que mejora el comportamiento de los métodos comúnmente utilizados frente a las circunstancias cambiantes no predecibles. Las tres magnitudes clave del Big Data, también conocidas como V's, están representadas y mejoradas: velocidad, enriqueciendo los sistemas de acceso de datos por medio de una reducción de los tiempos de tratamiento de las búsquedas en los sistemas Grid bioinformáticos; variedad, utilizando sistemas multimedia menos frecuentes que los basados en datos tabulares; y por último, volumen, incrementando las capacidades de autoescalado mediante el aprovechamiento de contenedores software y algoritmos bioinspirados. / [CA] La utilització de sistemes per al tractament eficient de grans volums d'informació ha crescut en popularitat durant els últims anys. Açò comporta el desenvolupament de noves tecnologies, mètodes i algoritmes, que aconsellen l'ús eficient de les infraestructures. El tractament de grans volums d'informació no està exempt de nombrosos problemes i reptes, alguns dels quals es tractaran de millorar. Dins de les possibilitats actuals hem de tindre en compte l'evolució que han tingut els sistemes durant els últims anys i les ocasions de millora que existisquen en cada un d'ells. El primer sistema d'estudi, el Grid, constituïx una aproximació inicial de processament massiu i representa un dels primers sistemes de tractament distribuït de grans conjunts de dades. Participant en la modernització d'un dels mecanismes d'accés a les dades es facilita la millora dels tractaments que es realitzen en la genòmica actual. Els estudis que es presenten estan centrats en la transformada de Burrows-Wheeler, que ja és coneguda en l'anàlisi genòmica per la seua capacitat per a millorar els temps en l'alineament de cadenes curtes de polinucleòtids. Esta millora en els temps, es perfecciona per mitjà de la reducció dels accessos remots amb la utilització d'un sistema de memòria cau intermèdia que optimitza la seua execució en un sistema Grid ja consolidat. Esta caché s'implementa com a complement a la llibreria d'accés estàndard GFAL utilitzada en la infraestructura d'IberGrid. En un segon pas es planteja el tractament de les dades en arquitectures de Big Data. Les millores es realitzen tant en l'arquitectura Lambda com a Kappa per mitjà de la busca de mètodes per a tractar grans volums d'informació multimèdia. Mentre que en l'arquitectura Lambda s'utilitza Apache Hadoop com a tecnologia per a este tractament, en l'arquitectura Kappa s'utilitza Apache Storm com a sistema de computació distribuït en temps real. En ambdós arquitectures s'àmplia l'àmbit d'utilització i s'optimitza l'execució per mitjà de l'aplicació d'algoritmes que milloren els problemes en cada una de les tecnologies. El problema del volum de dades és el centre d'un últim escaló, pel qual es permet millorar l'arquitectura de microserveis. Tenint en compte el nombre total de nodes que s'executen en sistemes de processament tenim una aproximació de les magnituds que podem obtindre per al tractaments de grans volums. D'aquesta manera la capacitat dels sistemes per a augmentar o disminuir la seua dimensió permet un govern òptim. Proposant un sistema bioinspirat s'aporta un mètode d'autoescalat dinàmic i distribuït que millora el comportament dels mètodes comunment utilitzats enfront de les circumstàncies canviants no predictibles. Les tres magnituds clau del Big Data, també conegudes com V's, es troben representades i millorades: velocitat, enriquint els sistemes d'accés de dades per mitjà d'una reducció dels temps de tractament de les busques en els sistemes Grid bioinformàtics; varietat, utilitzant sistemes multimèdia menys freqüents que els basats en dades tabulars; i finalment, volum, incrementant les capacitats d'autoescalat per mitjà de l'aprofitament de contenidors i algoritmes bioinspirats. / [EN] The use of systems for the efficient treatment of large data volumes has grown in popularity during the last few years. This has led to the development of new technologies, methods and algorithms to efficiently use of infrastructures. The Big Data treatment is not exempt from numerous problems and challenges, some of which will be attempted to improve. Within the existing possibilities, we must take into account the evolution that systems have had during the last years and the improvement that exists in each one. The first system of study, the Grid, constitutes an initial approach of massive distributed processing and represents one of the first treatment systems of big data sets. By researching in the modernization of the data access mechanisms, the advance of the treatments carried out in current genomics is facilitated. The studies presented are centred on the Burrows-Wheeler Transform, which is already known in genomic analysis for its ability to improve alignment times of short polynucleotids chains. This time, the update is enhanced by reducing remote accesses by using an intermediate cache system that optimizes its execution in an already consolidated Grid system. This cache is implemented as a GFAL standard file access library complement used in IberGrid infrastructure. In a second step, data processing in Big Data architectures is considered. Improvements are made in both the Lambda and Kappa architectures searching for methods to process large volumes of multimedia information. For the Lambda architecture, Apache Hadoop is used as the main processing technology, while for the Kappa architecture, Apache Storm is used as a real time distributed computing system. In both architectures the use scope is extended and the execution is optimized applying algorithms that improve problems for each technology. The last step is focused on the data volume problem, which allows the improvement of the microservices architecture. The total number of nodes running in a processing system provides a measure for the capacity of processing large data volumes. This way, the ability to increase and decrease capacity allows optimal governance. By proposing a bio-inspired system, a dynamic and distributed self-scaling method is provided improving common methods when facing unpredictable workloads. The three key magnitudes of Big Data, also known as V's, will be represented and improved: speed, enriching data access systems by reducing search processing times in bioinformatic Grid systems; variety, using multimedia data less used than tabular data; and finally, volume, increasing self-scaling capabilities using software containers and bio-inspired algorithms. / Herrera Hernández, J. (2020). Optimización de arquitecturas distribuidas para el procesado de datos masivos [Tesis doctoral]. Universitat Politècnica de València. https://doi.org/10.4995/Thesis/10251/149374 / Compendio