Die vorliegende Arbeit beschäftigt sich mit Mechanismen, die in Einzelzellmodellen zu einer frequenzabhängigen Informationsübertragung führen können. Um dies zu untersuchen, werden Methoden aus der theoretischen Physik (Statistische Physik) und der Informationstheorie angewandt. Die Informationsfilterung in mehreren stochastischen Neuronmodellen, in denen unterschiedliche Mechanismen zur Informationsfilterung führen können, werden numerisch und, falls möglich, analytisch untersucht. Die Bandbreite der betrachteten Modelle erstreckt sich von reduzierten strombasierten ’Integrate-and-Fire’ (IF) Modellen bis zu biophysikalisch realistischeren leitfähigkeitsbasierten Modellen. Anhand numerischer Untersuchungen wird aufgezeigt, dass viele Varianten der IF-Neuronenmodelle vorzugsweise Information über langsame Anteile eines zeitabhängigen Eingangssignals übertragen. Der einfachste Vertreter der oben genannten Klasse der IF-Neuronmodelle wird dahingehend erweitert, dass ein Konzept von neuronalem ’Gedächtnis’, vermittelst positiver Korrelationen zwischen benachbarten Intervallen aufeinander- folgender Spikes, integriert wird. Dieses Model erlaubt eine analytische störungstheoretische Untersuchung der Auswirkungen positiver Korrelationen auf die Informationsfilterung. Um zu untersuchen, wie sich sogenannte ’unterschwelligen Resonanzen’ auf die Signalübertragung auswirken, werden Neuronenmodelle mit verschiedenen Nichtlinearitäten anhand numerischer Computersimulationen analysiert. Abschließend wird die Signalübertragung in einem neuronalen Kaskadensystem, bestehend aus linearen und nichtlinearen Elementen, betrachtet. Neuronale Nichtlinearitäten bewirken eine gegenläufige Abhängigkeit (engl. "trade-off") zwischen qualitativer, d.h. frequenzselektiver, und quantitativer Informations-übertragung, welche in allen von mir untersuchten Modellen diskutiert wird. Diese Arbeit hebt die Gewichtigkeit von Nichtlinearitäten in der neuronalen Informationsfilterung hervor. / Neurons transmit information about time-dependent input signals via highly non-linear responses, so-called action potentials or spikes. This type of information transmission can be frequency-dependent and allows for preferences for certain stimulus components. A single neuron can transmit either slow components (low pass filter), fast components (high pass filter), or intermediate components (band pass filter) of a time-dependent input signal. Using methods developed in theoretical physics (statistical physics) within the framework of information theory, in this thesis, cell-intrinsic mechanisms are being investigated that can lead to frequency selectivity on the level of information transmission. Various stochastic single neuron models are examined numerically and, if tractable analytically. Ranging from simple spiking models to complex conductance-based models with and without nonlinearities, these models include integrator as well as resonator dynamics. First, spectral information filtering characteristics of different types of stochastic current-based integrator neuron models are being studied. Subsequently, the simple deterministic PIF model is being extended with a stochastic spiking rule, leading to positive correlations between successive interspike intervals (ISIs). Thereafter, models are being examined which show subthreshold resonances (so-called resonator models) and their effects on the spectral information filtering characteristics are being investigated. Finally, the spectral information filtering properties of stochastic linearnonlinear cascade neuron models are being researched by employing different static nonlinearities (SNLs). The trade-off between frequency-dependent signal transmission and the total amount of transmitted information will be demonstrated in all models and constitutes a direct consequence of the nonlinear formulation of the models.

Description des différentes étapes de la conception d'un microprocesseur pour le contrôle des automatismes de sécurité, en particulier pour les systèmes de transport. Ce microprocesseur est autotestable, c'est-à-dire capable de détecter ses propres erreurs. La conception du circuit est basée sur les hypothèses de pannes au niveau analytique dans la technologie NMOS. Les blocs fonctionnels «Strongly Fault Secure» et les contrôleurs «Strongly Code Disjoint» sont à la base des circuits «Self-checking», dits autotestables. Le circuit COBRA démontre la faisabilité d'un microprocesseur autotestable. COBRA gère indépendamment 19 signaux différents, date des événements externes, mesure des fréquences, surveille 14 entrées logiques et possède 7 sorties indépendantes. Le programme d'application de COBRA est contenu dans une mémoire morte programmable externe de 16 Koctets adressés par 14 bits multiplexés sur le bus interne de 8 bits. COBRA contient également une liaison série, une mémoire à accès direct de 64 octets et 3 temporisateurs de 14 bits indépendants ainsi qu'une unité arithmétique et logique de 8 bits, COBRA exécute un jeu de 43 instructions

To measure the speed of a vehicle, the revolution of a wheel or a rigid axle is traditionally used. Therefore non corrigible systematic errors occur which are caused by slip, spin and by the change of wheel diameter due to fretting. Train control and traction systems require new robust as well as precise methods of speed measurement. Because of their physical properties, Doppler-radar-sensors attached to the vehicle and measuring ground speed are first choice for this range of applications. Currently used sensors cannot fulfil the high demands under all operating conditions, because they are unable to completely compensate the various interferences and systematic deviations. This is the starting point of this dissertation. Two independent diverse methods with optimised reliability and accuracy must be used to meet all requirements. Limited resources of the embedded digital signal processor system under real-time conditions have to be taken into account. According to the boundary conditions, the introductory chapters critically discuss the frequency analysis methods currently used and describe starting points for further development. This leads to the design of a new, robust, wide-band spectral analysis which combines techniques of the dyadic wavelet transformation with the fast Fourier transformation. At the same time a new frame procedure and general model for the estimation of motion parameters is developed which features short delays. The disadvantages of the block-based discrete spectral analysis applied over continuous approaches are extensively compensated. The block structure of spectral data enables the selective use of new knowledge-based spectral filters for the compensation of the remaining intense interferences which are typical of this kind of application. / Die Fahrzeuggeschwindigkeitsmessung über die Drehzahl eines Rades weist in Schlupf- und Schleuderzuständen erhebliche systematische Abweichungen auf. Deshalb erfordern Zugbeeinflussungs- und Antriebssysteme neue gleichzeitig robuste und präzise Geschwindigkeitsmessmethoden. Die Mikrowellensensorik unter Nutzung des Dopplereffekts zwischen Fahrzeug und Gleisbett wird wegen ihrer physikalischen Eigenschaften für dieses Einsatzgebiet favorisiert. Bisherige Sensorapplikationen erfüllen aber die hohen Ansprüche nicht in allen Betriebszuständen. Hier setzt die in dieser Arbeit beschriebene Sensorentwicklung auf. Zwei getrennt hergeleitete und nach Zuverlässigkeit und Genauigkeit optimierte neue Verfahren können bei gleichzeitiger Anwendung die gestellten Anforderungen erfüllen. Dabei müssen auch die beschränkten Ressourcen des eingebetteten digitalen Signalverarbeitungssystems unter Echtzeitbedingungen berücksichtigt werden. Entsprechend dieser Randbedingungen findet einleitend eine kritische Betrachtung bestehender Frequenzanalysemethoden statt und Ansätze für die Weiterentwicklung werden herausgearbeitet. Einerseits führt dies zur Konstruktion einer neuen störunempfindlichen Weitbereichsspektralzerlegung, welche Ansätze der dyadischen Wavelettransformation mit der Diskreten Fourier-Transformation verbindet. Andererseits wird ein neues Rahmenverfahren für die verzögerungsarme Schätzung der Bewegungsparameter des Fahrzeuges aufgrund seines physikalischen Bewegungsmodells hergeleitet und mit einem hochgenauen Frequenzauswerteverfahren kombiniert. Beide Verfahren basieren auf blockweisen diskreten Spektralzerlegungen, deren prinzipielle Nachteile gegenüber kontinuierlichen Ansätzen weitgehend kompensiert werden können. Durch die Blockorganisation lassen sich neuartige wissensbasierte Spektralfilter selektiv zur Unterdrückung starker bahnanwendungstypischer Störeinflüsse einsetzen.

Nous traitons des signaux électromyographiques intramusculaires (signaux iEMG) relevés dans les muscles de l'avant-bras. Les signaux iEMG représentent une image de la commande du système nerveux central vers les muscles. Ils se composent d'une superposition de trains d'ondelettes, chaque ondelette code un groupe de fibres musculaires et son taux de mise à feu code l'effort produit par ce groupe. L'objectif est d'extraire de façon séquentielle des informations du signal iEMG. Nous espérons que ces informations se révèleront utiles pour la commande d'une prothèse d'avant-bras. En premier lieu, nous modélisons un train d'impulsions comme une chaîne de Markov et nous discutons des lois pouvant caractériser le temps entre deux impulsions. La loi de Weibull discrète a retenu notre attention. Nous avons mis en place une méthode d'estimation en ligne de ses paramètres. En second lieu, nous modélisons le signal iEMG par un modèle de Markov caché s'appuyant sur le modèle de train d'impulsions ci-dessus. La mise en place d'un filtre bayésien nous permet de propager séquentiellement une estimation bayésienne des paramètres du modèle de Markov caché, en particulier la forme des ondelettes et leur taux de mise à feu. Nous proposons finalement une méthode d'estimation du nombre de trains d'ondelettes, un paramètre discret du modèle. Nous validons les méthodes et algorithmes proposés sur des signaux simulés et des signaux iEMG.

This dissertation discusses logistical aspects of the Persians invasion of Greece; the Athenian need for timber for building warships; supply problems in their assault on Syracuse; and the march of Alexander's army from Macedonia into Asia. The amount of cereals needed by the Persian and Greek armies and navies is calculated from modern nutritional data and an estimate of the numbers of combatants. The location and size of the Persian food dumps; the excavation of the Athos canal; and the ships and materials needed to build the bridges of boats are considered. The Athenian need for ship-timber led to the costly occupation of Amphipolis. An assured supply of cereals was one motive for the disastrous Sicilian Expedition. The Athenian fleet was an inefficient long-range support for an army which had to protect its non-combatant sailors. This was realised by Alexander the Great, who crossed the Hellespont without naval support. / Classics and Modern European Languages / M.A. (Classics)

No description available.

CORTEZ, Ana Isabel Ribeiro Parente. Memória descarrilhadas: o trem na cidade do Crato. 2008. 245 f. Dissertação (Mestrado em História) - Universidade Federal do Ceará, Departamento de História, Programa de Pós-Graduação em História Social, Fortaleza-ce, 2008. / Submitted by Maria Josineide Góis (josineide@ufc.br) on 2012-06-19T16:53:08Z No. of bitstreams: 1 2008_Dis_AIRPCortez.pdf: 5401271 bytes, checksum: 8456eb03096d4737033602b79ff016d4 (MD5) / Approved for entry into archive by Maria Josineide Góis(josineide@ufc.br) on 2012-06-25T12:42:48Z (GMT) No. of bitstreams: 1 2008_Dis_AIRPCortez.pdf: 5401271 bytes, checksum: 8456eb03096d4737033602b79ff016d4 (MD5) / Made available in DSpace on 2012-06-25T12:42:48Z (GMT). No. of bitstreams: 1 2008_Dis_AIRPCortez.pdf: 5401271 bytes, checksum: 8456eb03096d4737033602b79ff016d4 (MD5) Previous issue date: 2008 / Memórias Descarriladas interpreta a construção de narrativas a partir de lembranças do trem na cidade do Crato. Na oralidade, os entrevistados inventaram e reinventaram suas memórias sobre a máquina férrea engendrando inúmeras relações que precipitaram seus enredos para fora dos trilhos. O descarrilamento destas memórias permitiu a construção de inúmeras relações com outros aspectos. Nesse sentido, a cidade do Crato é redimensionada e surge múltipla, formulada de acordo com as referências pessoais de quem a descreve; assim como, as viagens, os momentos na estação e os acidentes férreos são interpretados a partir de constantes relações com o vivido e o esperado. Nessas narrativas o trem é reconstruído, transformado através de cores, cheiros, sons, e sabores.

Trabalhos prévios mostram que a variabilidade do gelo marinho antártico em diversas escalas temporais está intimamente relacionada a mecanismos de teleconexões trópicos-extratrópicos. Com base nesta hipótese, este trabalho pretende estabelecer a resposta da passagem dos trens de ondas em latitudes médias, associados a fenômenos de escala intrasazonal (20-100 dias) como a Oscilação de Madden-Julian (Madden-Julian Oscillation ou MJO), nas camadas superiores do Oceano Austral (OA) e impactos no gelo marinho antártico. O período investigado neste estudo é 19892007, com ênfase no inverno e sobre o mar de Ross, localizado no setor Pacífico austral (região diretamente afetada pela passagem dos trens de ondas de latitudes médias). Composições com defasagens (lag composites) de anomalias intrasazonais da tensão de cisalhamento do vento (zonal e meridional) mostram que correntes oceânicas são geradas em resposta a essa forçante atmosférica sobre o oceano no setor Pacífico austral. O transporte zonal e meridional de massa na camada de Ekman oceânica indica que divergência de massa nessa camada precede os eventos extremos intrasazonais de retração do gelo marinho em Ross (EIR). Em contraste, convergência precede períodos de eventos extremos intrasazonais de expansão do gelo marinho em Ross (EIE). A divergência (convergência) de massa na camada de Ekman associada com anomalias intrasazonais do bombeamento de Ekman resulta em ressurgência (subsidência) que precede a ocorrência de EIR (EIE). Alguns trabalhos mostram que águas intermediárias antárticas, que são relativamente mais quentes no inverno em relação às águas superficiais que estão próximas ao ponto de congelamento (ou congeladas), são dirigidas para a superfície do oceano pelo bombeamento de Ekman e ocasionam o derretimento do gelo marinho. Anomalias do transporte meridional de calor na camada de Ekman oceânica mostram que durante os EIR (EIE), calor é transportado para dentro (fora) do mar de Ross entre 15 e 8 dias (12 e 8 dias) precedentes aos EIR (EIE). Anomalias intrasazonais do fluxo de calor na interface ar-mar mostram que precedendo o dia de observação dos EIR (EIE) o fluxo de calor é direcionado da atmosfera para o oceano (do oceano para a atmosfera), sendo essa configuração associada a um ganho (perda) de calor no oceano superior em Ross. Em todas as composições, observa-se a mudança de fase das anomalias nos dias posteriores (lags positivos) ao dia dos EIG e são consistentes com a propagação do modo conhecido como Pacific-South-American (PSA), identificado nesse trabalho por meio de anomalias intrasazonais da altura geopotencial em 200 hPa. Além disso, uma diferença notada em alguns casos nas lag-composities é que em períodos de MJO ativa, as anomalias parecem estar mais deslocadas para o sul do que em períodos de MJO inativa. Em períodos de MJO inativa foram observados 15 (13) eventos de EIR (EIE), enquanto que, em períodos de MJO ativa observou-se 25 (24) eventos de EIR (EIE). Observa-se ainda que há uma maior quantidade de ciclones quando a MJO está presente. Por exemplo, o número de ciclones com duração a partir de 12 horas para períodos sem MJO foi igual a 146 para os EIR e 130 para os EIE. Já o número de ciclones para períodos com MJO foi igual a 311 para os EIR e 278 para os EIE. Com isso, observa-se claramente o papel da MJO na circulação de latitudes média e possíveis associações com o gelo marinho, pois é sabido que a atividade ciclônica está relacionada à advecção de massas de ar sobre o gelo marinho, além da advecção do próprio gelo marinho. Para investigar em detalhes a interação oceano-atmosfera-gelo marinho foram examinados casos persistentes de EIR e EIE. Os casos mais persistentes de EIR (EIE) tiveram durações de 34 e 30 (26 e 25) dias, sendo esses os casos analisados. Mostra-se que as anomalias intrasazonais da circulação atmosférica em baixos níveis (em 850 hPa) estão associadas a advecções quentes (frias) na proporção de aproximadamente 0,5 1 (0,1 1) m.s-1 em períodos precedentes aos EIR (EIE). No geral, anomalias em latitudes médias da circulação atmosférica ciclônicas (anti-ciclônicas) e divergência (convergência) das correntes oceânicas superficiais aparecem relacionadas à ressurgência (subsidência) da ordem de 0,1 0,3 m2.s-1 em algumas pêntadas anteriores a pêntada que corresponde ao início dos supercasos de EIR (EIE). Os padrões mudam de fase com o tempo, o que sugere a propagação de um padrão de onda em escala intrasazonal. Estas anomalias mostram-se abrangendo dimensões espaciais que compreendem grande parte do setor Pacífico austral, incluindo o mar de Ross. Impactos associados a estas anomalias podem ser verificados diretamente no mar de Ross através das análises da concentração do gelo marinho em Ross. Anomalias intrasazonais negativas (positivas) da concentração do gelo marinho predominam sobre o campo do gelo marinho do mar de Ross nas primeiras pêntadas a partir daquela que indica o início dos casos persistentes de EIR (EIE). Durante os períodos de EIR, as anomalias intrasazonais negativas da concentração do gelo mostram-se da ordem de aproximadamente 5% a 10% no interior do mar de Ross e entre 15% a 30% nas bordas do gelo marinho de Ross. Já durante os períodos de EIE, as anomalias intrasazonais positivas da concentração do gelo marinho em Ross mostram-se da ordem de 10% a 30% nas bordas do gelo marinho do mar de Ross. Esse resultado mostra que a resposta do gelo marinho aos padrões atmosféricos e oceânicos em escala intrasazonal possui uma defasagem entre 5 e 1 pêntada(s). No geral, este trabalho cumpriu o objetivo de verificar as respostas do oceano às anomalias da circulação atmosférica e impactos associados no gelo marinho, em escala intrasazonal. / Previous works show that antarctic sea ice variability on several time-scales is close related to tropics-extratropics teleconections mechanisms. Based on this hyphotesis, this work intend to verify the responses in oceanic upper layers of Austral Ocean on intraseasonal time-scale (20-100 days) phenomenom and impacts on sea ice due to anomalous atmospheric circulation associated to the Madden-Julian Oscillation (MJO). The period analysed in this study is from 1989 to 2007, with emphasis on winter season and on Ross Sea (located at austral Pacific sector). Lag composities of zonal and meridional intraseasonal wind stress anomalies show that oceanic currents are generated as a response of these atmospheric forcings on austral Pacific sector. Zonal and meridional mass transport on oceanic Ekman layer, which are perpendicular and to the left of wind stress at Southern Hemisphere (SH), indicate that Ekman mass divergence precedes intraseasonal Ross sea ice extreme retraction (EIR). In contrast, convergence precedes the periods of extreme intraseasonal Ross sea ice expantion (EIE). Divergence (convergence) on oceanic Ekman layer associated to intraseasonal Ekman pumping anomalies results in upwelling (downwelling) wich precedes the occurrence of EIR (EIE). Some works have already shown that intermediate antarctic waters, wich are relatively warmer in the wintertime when compared to superficial waters that are next to the freezing point (or freezed), are headed to ocean surface due to Ekman pumping, generating sea ice melt. Intraseasonal anomalies of sea-air heat flux show that days before EIR (EIE) occurrences, the flux is headed from atmosphere to the ocean (from ocean to the atmosphere), which configuration is associated to the earn (loss) of heat at Ross upper ocean. In all compositions, the change of anomalies phase on the days before EI occurrence (positive lags) is clearly noticed and is consistent to the propagations of the mode known as Pacific South American (PSA), revealed in intraseasonal anomalies of geopotencial height at 200 hPa. Furthermore, in some cases (as in the case of Ekman pumping and Sverdrup transport) the anomalies seem to be deplaced southward in active MJO periods than in inactive MJO periods. In inactive MJO periods were observed 15 (13) EIR (EIE) events, while in active MJO periods were observed 25 (24) EIR (EIE) events. Furthermore, the number of cyclones during EIR periods was bigger than during EIE periods. In addiction, more cyclones were observed when MJO is active. For example, the number of cyclones with duration of 6 (12) hours without MJO was equal to 174 (146) during EIR events and 169 (130) during EIE events. However, the number of cyclones with active MJO was equal to 393 (311) in EIR events and 364 (278) in EIE events. In order to investigate in details the interaction between ocean-atmosphere-sea ice, it was examinated persistents cases of EIR and EIE events. The cases more persistents of EIR (EIE) events had durations of 34 and 30 (26 and 25) days, which were the analised cases. It was observed that intraseasonal anomalies of atmospheric circulation at lower levels (in 850 hPa) and intraseasonal anomalies of superficial ocean currents were associates to hot (cold) advection during periods before EIR (EIR) events. In general, the medium latitude cyclonic (anticyclonic) anomalies of atmospheric circulation and divergence (convergence) of superficial ocean currents seem to be linked to upwelling (downwelling) in some pentads before the pentad which is correspondent to the beggining of EIR (EIE) supercases. The patterns observed change their phases along the time, suggesting the propagation of extratropical intraseasonal wave train pattern. Negatives (positives) intraseasonal anomalies of sea ice concentration were observed above Ross Sea in the first pentads after the beggining of EIR (EIE) persistents cases. This result shows that sea ice response to atmospheric and to oceanic patterns on intraseasonal time-scales has a lag between 5 and 1 pentad(s). In general, this work contributed to better understand the oceanic responses due to anomalies in atmospheric circulation and related impacts on sea ice, on intraseasonal time-scale.

I investigate the spoken dialogue of four Indian novels in English: Mulk Raj Anand's Untouchable (1935), Khushwant Singh's Train to Pakistan (1956), Rasipuram Krishnaswami Narayan's The World of Nagaraj (1990), and Rohinton Mistry's Family Matters (2002). Roger Fowler has said that literature, as a form of discourse, articulates ideology; it is through linguistic criticism (combination of literary criticism and linguistic analyses) that the ideologies in a literary text are uncovered. Shobhana Chelliah in her study of Indian novels in English concludes that the authors use Indian English (IndE) as a device to characterize buffoons and villains. Drawing upon Fowler's and Chelliah's framework, my investigation employs linguistic criticism of the four novels to expose the ideologies reflected in the use of fictionalized English in the Indian context. A quantitative inquiry based on thirty-five IndE features reveals that the authors appropriate these features, either to a greater or lesser degree, to almost all their characters, suggesting that IndE functions as the mainstream variety in these novels and creating an illusion that the authors are merely representing the characters' unique Indian worldviews. But within this dialect range, the appropriation of higher percentages of IndE features to specific characters or groups of characters reveal the authors' manipulation of IndE as a counter-realist and ideological device to portray deviant and defective characters. This subordinating of IndE as a substandard variety of English functions as the dominant ideology in my investigation of the four novels. Nevertheless, I also uncover the appropriation of a higher percentage of IndE features to foreground the masculinity of specific characters and to heighten the quintessentially traditional values of the older Brahmin generation, which justifies a contesting ideology about IndE that elevates it as the prestigious variety, not an aberration. Using an approach which combines literary criticism with linguistic analysis, I map and recommend a multidisciplinary methodology, which allows for a reevaluation of fictionalized IndE speech that goes beyond impressionistic analyses.

The work in this thesis is separated into two parts. The first part deals with the load planning and sequencing problem for double-stack intermodal railcars, an operational problem found at many rail container terminals. In this problem, containers must be assigned to a platform on which the container will be loaded, and the loading order must be determined. These decisions are made with the objective of minimizing the costs associated with handling the containers, as well as minimizing the cost of containers left behind. The deterministic version of the problem can be cast as a shortest path problem on an ordered graph. This problem is challenging to solve because of the large size of the graph. We propose a two-stage heuristic based on the Iterative Deepening A* algorithm to compute solutions to the load planning and sequencing problem within a five-minute time budget. Next, we also illustrate how a Deep Q-learning algorithm can be used to heuristically solve the same problem.The second part of this thesis considers sequential models in deep learning. A recent strategy to circumvent the exploding and vanishing gradient problem in recurrent neural networks (RNNs) is to enforce recurrent weight matrices to be orthogonal or unitary. While this ensures stable dynamics during training, it comes at the cost of reduced expressivity due to the limited variety of orthogonal transformations. We propose a parameterization of RNNs, based on the Schur decomposition, that mitigates the exploding and vanishing gradient problem, while allowing for non-orthogonal recurrent weight matrices in the model. / Le travail de cette thèse est divisé en deux parties. La première partie traite du problème de planification et de séquencement des chargements de conteneurs sur des wagons, un problème opérationnel rencontré dans de nombreux terminaux ferroviaires intermodaux. Dans ce problème, les conteneurs doivent être affectés à une plate-forme sur laquelle un ou deux conteneurs seront chargés et l'ordre de chargement doit être déterminé. Ces décisions sont prises dans le but de minimiser les coûts associés à la manutention des conteneurs, ainsi que de minimiser le coût des conteneurs non chargés. La version déterministe du problème peut être formulé comme un problème de plus court chemin sur un graphe ordonné. Ce problème est difficile à résoudre en raison de la grande taille du graphe. Nous proposons une heuristique en deux étapes basée sur l'algorithme Iterative Deepening A* pour calculer des solutions au problème de planification et de séquencement de la charge dans un budget de cinq minutes. Ensuite, nous illustrons également comment un algorithme d'apprentissage Deep Q peut être utilisé pour résoudre heuristiquement le même problème. La deuxième partie de cette thèse examine les modèles séquentiels en apprentissage profond. Une stratégie récente pour contourner le problème de gradient qui explose et disparaît dans les réseaux de neurones récurrents (RNN) consiste à imposer des matrices de poids récurrentes orthogonales ou unitaires. Bien que cela assure une dynamique stable pendant l'entraînement, cela se fait au prix d'une expressivité réduite en raison de la variété limitée des transformations orthogonales. Nous proposons une paramétrisation des RNN, basée sur la décomposition de Schur, qui atténue les problèmes de gradient, tout en permettant des matrices de poids récurrentes non orthogonales dans le modèle.