Graduation date: 2006 / Presented on 2005-12-02 / An empirical statistical model is developed that relates the non-tidal motion of the ocean surface currents off the Oregon coast to forecasts of the coastal winds. The empirical statistical model is then used to produce predictions of the surface currents that are evaluated for their agreement with measured currents. Measurements of the ocean surface currents were made at 6 km resolution using Long-Range CODAR SeaSonde high-frequency (HF) surface current mappers and wind forecasts were provided at 12 km resolution by the North American Mesoscale (NAM) model. First, the response of the surface currents to wind-forcing measured by five coastal National Data Buoy Center (NDBC) stations was evaluated using empirical orthogonal function (EOF) analysis. A significant correlation of approximately 0.8 was found between the majority of the variability in the seasonal anomalies of the low-pass filtered surface currents and the seasonal anomalies of the low-pass filtered wind stress measurements. The U and the V components of the measured surface currents were both shown to be forced by the zonal and meridional components of the wind-stress at the NDBC stations. Next, the NAM wind forecasts were tested for agreement with the measurements of the wind at the NDBC stations. Significant correlations of around 0.8 for meridional wind stress and 0.6 for zonal wind stress were found between the seasonal anomalies of the low-pass filtered wind stress measured by the NDBC stations and the seasonal anomalies of the low-pass filtered wind stress forecast by the NAM model. Given the amount of the variance in the winds captured by the NAM model and the response of the ocean surface currents to both components of the wind, bilinear regressions were formed relating the seasonal anomalies of the low-pass filtered NAM forecasts to the seasonal anomalies of the low-pass filtered surface currents. The regressions turned NAM wind forecasts into predictions of the seasonal anomalies of the low-pass filtered surface currents. Calculations of the seasonal cycle in the surface currents, added to these predicted seasonal anomalies, produced a non-tidal estimation of the surface currents that allowed a residual difference to be calculated from recent surface current measurements. The sum of the seasonal anomalies, the seasonal cycle, and the residual formed a prediction of the non-tidal surface currents. The average error in this prediction of the surface currents off the Oregon coast remained less than 4 cm/s out through 48 hours into the future.

The main goal of this thesis is to develop practical digital controller architectures for multi-phase dc-dc converters utilized in low power (up to few hundred watts) and cost-sensitive applications. The proposed controllers are suitable for on-chip integration while being capable of providing advanced features, such as dynamic efficiency optimization, inductor current estimation, converter component identification, as well as combined dynamic current sharing and fast transient response. The first part of this thesis addresses challenges related to the practical implementation of digital controllers for low-power multi-phase dc-dc converters. As a possible solution, a multi-use high-frequency digital PWM controller IC that can regulate up to four switching converters (either interleaved or standalone) is presented. Due to its configurability, low current consumption (90.25 μA/MHz per phase), fault-tolerant work, and ability to operate at high switching frequencies (programmable, up to 10 MHz), the IC is suitable to control various dc-dc converters. The applications range from dc-dc converters used in miniature battery-powered electronic devices consuming a fraction of watt to multi-phase dedicated supplies for communication systems, consuming hundreds of watts. A controller for multi-phase converters with unequal current sharing is introduced and an efficiency optimization method based on logarithmic current sharing is proposed in the second part. By forcing converters to operate at their peak efficiencies and dynamically adjusting the number of active converter phases based on the output load current, a significant improvement in efficiency over the full range of operation is obtained (up to 25%). The stability and inductor current transition problems related to this mode of operation are also resolved. At last, two reconfigurable digital controller architectures with multi-parameter estimation are introduced. Both controllers eliminate the need for external analog current/temperature sensing circuits by accurately estimating phase inductor currents and identifying critical phase parameters such as equivalent resistances, inductances and output capacitance. A sensorless non-linear, average current-mode controller is introduced to provide fast transient response (under 5 μs), small voltage deviation and dynamic current sharing with multi-phase converters. To equalize the thermal stress of phase components, a conduction loss-based current sharing scheme is proposed and implemented.

En la presente tesis se muestra una nueva metodología de extracción "inteligente" de modelos circuitales lineales y no lineales para dispositivos MESFET y HEMT, además de efectuar numerosas aportaciones en el campo de las medidas radioeléctricas de dichos dispositivos mediante diseño del hardware y del software necesario para la automatización de las mismas. Por otro lado se presenta un novedoso modelo de Gran Señal para dispositivos HEMT de potencia que da cuenta del fenómeno de la compresión de la transconductancia y es fácilmente implementable en simuladores no lineales comerciales del tipo de MDS, LIBRA, HARMONICA, etc. Además se ha aumentado el rango de validez frecuencial de los modelos de pequeña señal mediante la obtención de las expresiones "exactas" de los modelos usuales de pequeña señal Vendelin-Dambrine, Vickes, Berroth & Bosch, etc. Otra novedad aportada por este trabajo de tesis ha sido aplicar estos modelos lineales a los transistores HEMT, evitando la obtención valores carentes de significado físico como ocurría hasta ahora. Como validación del modelo no lineal de HEMT se han llevado a cabo numerosas simulaciones del mismo en MDS que han sido comparadas con las medidas experimentales realizadas en nuestro laboratorio (Scattering, DC, Pulsadas y Pin/Pout) poniendo de manifiesto la exactitud del modelo. Para validar los modelos de pequeña señal se han efectuado simulaciones con el simulador lineal MMICAD utilizando transistores de diferentes tamaños procedentes de distintas foundries con objeto de visualizar el comportamiento del dispositivo independientemente del origen del mismo. / In this thesis a new methodology for the "intelligent" parameter extraction of linear and non-linear model for GaAs MESFET and HEMT devices is shown, besides numerous contributions in the field of Scattering and DC measurements of this kind of devices by means of hardware design and necessary software for the automation of the same have been done. On the other hand a novel Great Signal model for HEMT devices is presented. This model is capable to model the transconductance compression phenomenon and it is easily to built in commercial non-linear simulators like MDS, LIBRA, Microwave HARMONICA, etc. This work has also increased the frequency range for the usual small-signal models by means of calculate "exact" expressions of them. Another novelty contribution of this thesis is to apply for first time these linear models to HEMT transistors, avoiding the lacking of physical meaning values like it occurred up to now. To make possible the validation of non-linear HEMT model, simulations with MDS software and comparisons with experimental measurements made in our laboratory (Scattering, DC, Pulsed and Pin/ Pout) have been carried out and there was very good agreement between measured and simulated data. To validate small-signal models referred before, simulations with MMICAD software and comparisons between simulated and experimental scattering measurements using transistors of different sizes from several foundries and technological processes have been made.

La tesi desenvolupa el tema de la predicció de la volatilitat financera en el context de l’ús de dades d’alta freqüència, i se centra en una línia de recerca doble: proposar models alternatius que millorarien la predicció de la volatilitat i classificar els models de volatilitat ja existents com els que es proposen en aquesta tesi. Els objectius es poden classificar en tres categories. El primer consisteix en la proposta d’un nou mètode de predicció de la volatilitat que segueix una línia de recerca desenvolupada recentment, la qual apunta al fet de mesurar la volatilitat intradia, com també la nocturna. Es proposa una categoria de models realized GARCH bivariants. El segon objectiu consisteix en la proposta d’una metodologia per predir la volatilitat diària multivariant amb models autoregressius que utilitzen estimacions de volatilitat diària (i nocturna, en el cas dels bivariants), a més d’informació d’alta freqüència, quan se’n disposava. S’aplica l’anàlisi de components principals (ACP) a un conjunt de models de tipus realized GARCH univariants i bivariants. El mètode representa una extensió d’un model ja existent (PC-GARCH) que estimava un model GARCH multivariant a partir de l’estimació de models GARCH univariants dels components principals de les variables inicials. El tercer objectiu de la tesi és classificar el rendiment dels models de predicció de la volatilitat ja existents o dels nous, a més de la precisió de les mesures intradia que s’utilitzaven en les estimacions dels models. En relació amb els resultats, s’observa que els models EGARCHX, realized EGARCH i realized GARCH(2,2) obtenen una millor valoració, mentre que els models GARCH i no realized EGARCH obtenen uns resultats inferiors en gairebé totes les proves. Això permet concloure que el fet d’incorporar mesures de volatilitat intradia millora el problema de la modelització. Quant a la classificació dels models realized bivariants, s’observa que tant els models realized GARCH bivariant (en versions completes i parcials) com el model realized EGARCH bivariant obtenen millors resultats; els segueixen els models realized GARCH(2,2) bivariant, EGARCH bivariant I EGARCHX bivariant. En comparar les versions bivariants amb les univariants, amb l’objectiu d’investigar si l’ús de mesures de volatilitat nocturna a les equacions dels models millora l’estimació de la volatilitat, es mostra que els models bivariants superen els univariants. Els resultats proven que els models bivariants no són totalment inferiors als seus homòlegs univariants, sinó que resulten ser bones alternatives per utilitzar-los en la predicció, juntament amb els models univariants, per tal d’obtenir unes estimacions més fiables. / La tesis desarrolla el tema de la predicción de la volatilidad financiera en el contexto del uso de datos de alta frecuencia, y se centra en una doble línea de investigación: la de proponer modelos alternativos que mejorarían la predicción de la volatilidad y la de clasificar modelos de volatilidad ya existentes como los propuestos en esta tesis. Los objetivos se pueden clasificar en tres categorías. El primero consiste en la propuesta de un nuevo método de predicción de la volatilidad que sigue una línea de investigación recientemente desarrollada, la cual apunta al hecho de medir la volatilidad intradía, así como la nocturna. Se propone una categoría de modelos realized GARCH bivariantes. El segundo objetivo consiste en proponer una metodología para predecir la volatilidad diaria multivariante con modelos autorregresivos que utilizaran estimaciones de volatilidad diaria (y nocturna, en el caso de los bivariantes), además de información de alta frecuencia, si la había disponible. Se aplica el análisis de componentes principales (ACP) a un conjunto de modelos de tipo realized GARCH univariantes y bivariantes. El método representa una extensión de un modelo ya existente (PCGARCH) que calculaba un modelo GARCH multivariante a partir de la estimación de modelos GARCH univariantes de los componentes principales de las variables iniciales. El tercer objetivo de la tesis es clasificar el rendimiento de los modelos de predicción de la volatilidad ya existentes o de los nuevos, así como la precisión de medidas intradía utilizadas en las estimaciones de los modelos. En relación con los resultados, se observa que los modelos EGARCHX, realized EGARCH y GARCH(2,2) obtienen una mejor valoración, mientras que los modelos GARCH y no realized EGARCH obtienen unos resultados inferiores en casi todas las pruebas. Esto permite concluir que el hecho de incorporar medidas de volatilidad intradía mejora el problema de la modelización. En cuanto a la clasificación de modelos realized bivariantes, se observa que tanto los modelos realized GARCH bivariante (en versiones completas y parciales) como realized EGARCH bivariante obtienen mejores resultados; les siguen los modelos realized GARCH(2,2) bivariante, EGARCH bivariante y EGARCHX bivariante. Al comparar las versiones bivariantes con las univariantes, con el objetivo de investigar si el uso de medidas de volatilidad nocturna en las ecuaciones de los modelos mejora la estimación de la volatilidad, se muestra que los modelos bivariantes superan los univariantes. Los resultados prueban que los modelos bivariantes no son totalmente inferiores a sus homólogos univariantes, sino que resultan ser buenas alternativas para utilizarlos en la predicción, junto con los modelos univariantes, para lograr unas estimaciones más fiables. / The thesis develops the topic of financial volatility forecasting in the context of the usage of high frequency data, and focuses on a twofold line of research: that of proposing alternative models that would enhance volatility forecasting and that of ranking existing or newly proposed volatility models. The objectives may be disseminated in three categories. The first scope constitutes of the proposal of a new method of volatility forecasting that follows a recently developed research line that pointed to using measures of intraday volatility and also of measures of night volatility, the need for new models being given by the question whether adding measures of night volatility improves day volatility estimations. As a result, a class of bivariate realized GARCH models was proposed. The second scope was to propose a methodology to forecast multivariate day volatility with autoregressive models that used day (and night for bivariate) volatility estimates, as well as high frequency information when that was available. For this, the Principal Component algorithm (PCA) was applied to a class of univariate and bivariate realized GARCH-type of models. The method represents an extension of one existing model (PC GARCH) that estimated a multivariate GARCH model by estimating univariate GARCH models of the principal components of the initial variables. The third goal of the thesis was to rank the performance of existing or newly proposed volatility forecasting models, as well as the accuracy of the intraday measures used in the realized models estimations. With regards to the univariate realized models’ rankings, it was found that EGARCHX, Realized EGARCH and Realized GARCH(2,2) models persistently ranked better, while the non-realized GARCH and EGARCH models performed poor in each stance almost. This allowed us to conclude that incorporating measures of intraday volatility enhances the modeling problem. With respect to the bivariate realized models’ ranking, it was found that Bivariate Realized GARCH (partial and complete versions) and Bivariate Realized EGARCH models performed the best, followed by the Bivariate Realized GARCH(2,2), Bivariate EGARCH and Bivariate EGARCHX models. When the bivariate versions were compared to the univariate ones in order to investigate whether using night volatility measurements in the models’ equations improves volatility estimation, it was found that the bivariate models surpassed the univariate ones when specific methodology, ranking criteria and stocks were used. The results were mixed, allowing us to conclude that the bivariate models did not prove totally inferior to their univariate counterparts, proving as good alternative options to be used in the forecasting exercise, together with the univariate models, for more reliable estimates. Finally, the PC realized models and PC bivariate realized models were estimated and their performances were ranked; improvements the PC methodology brought in high frequency multivariate modeling of stock returns were also discussed. PC models were found to be highly effective in estimating multivariate volatility of highly correlated stock assets and suggestions on how investors could use them for portfolio selection were made.

No description available.

Ma thèse est composée de trois chapitres reliés à l'estimation des modèles espace-état et volatilité stochastique. Dans le première article, nous développons une procédure de lissage de l'état, avec efficacité computationnelle, dans un modèle espace-état linéaire et gaussien. Nous montrons comment exploiter la structure particulière des modèles espace-état pour tirer les états latents efficacement. Nous analysons l'efficacité computationnelle des méthodes basées sur le filtre de Kalman, l'algorithme facteur de Cholesky et notre nouvelle méthode utilisant le compte d'opérations et d'expériences de calcul. Nous montrons que pour de nombreux cas importants, notre méthode est plus efficace. Les gains sont particulièrement grands pour les cas où la dimension des variables observées est grande ou dans les cas où il faut faire des tirages répétés des états pour les mêmes valeurs de paramètres. Comme application, on considère un modèle multivarié de Poisson avec le temps des intensités variables, lequel est utilisé pour analyser le compte de données des transactions sur les marchés financières. Dans le deuxième chapitre, nous proposons une nouvelle technique pour analyser des modèles multivariés à volatilité stochastique. La méthode proposée est basée sur le tirage efficace de la volatilité de son densité conditionnelle sachant les paramètres et les données. Notre méthodologie s'applique aux modèles avec plusieurs types de dépendance dans la coupe transversale. Nous pouvons modeler des matrices de corrélation conditionnelles variant dans le temps en incorporant des facteurs dans l'équation de rendements, où les facteurs sont des processus de volatilité stochastique indépendants. Nous pouvons incorporer des copules pour permettre la dépendance conditionnelle des rendements sachant la volatilité, permettant avoir différent lois marginaux de Student avec des degrés de liberté spécifiques pour capturer l'hétérogénéité des rendements. On tire la volatilité comme un bloc dans la dimension du temps et un à la fois dans la dimension de la coupe transversale. Nous appliquons la méthode introduite par McCausland (2012) pour obtenir une bonne approximation de la distribution conditionnelle à posteriori de la volatilité d'un rendement sachant les volatilités d'autres rendements, les paramètres et les corrélations dynamiques. Le modèle est évalué en utilisant des données réelles pour dix taux de change. Nous rapportons des résultats pour des modèles univariés de volatilité stochastique et deux modèles multivariés. Dans le troisième chapitre, nous évaluons l'information contribuée par des variations de volatilite réalisée à l'évaluation et prévision de la volatilité quand des prix sont mesurés avec et sans erreur. Nous utilisons de modèles de volatilité stochastique. Nous considérons le point de vue d'un investisseur pour qui la volatilité est une variable latent inconnu et la volatilité réalisée est une quantité d'échantillon qui contient des informations sur lui. Nous employons des méthodes bayésiennes de Monte Carlo par chaîne de Markov pour estimer les modèles, qui permettent la formulation, non seulement des densités a posteriori de la volatilité, mais aussi les densités prédictives de la volatilité future. Nous comparons les prévisions de volatilité et les taux de succès des prévisions qui emploient et n'emploient pas l'information contenue dans la volatilité réalisée. Cette approche se distingue de celles existantes dans la littérature empirique en ce sens que ces dernières se limitent le plus souvent à documenter la capacité de la volatilité réalisée à se prévoir à elle-même. Nous présentons des applications empiriques en utilisant les rendements journaliers des indices et de taux de change. Les différents modèles concurrents sont appliqués à la seconde moitié de 2008, une période marquante dans la récente crise financière. / My thesis consists of three chapters related to the estimation of state space models and stochastic volatility models. In the first chapter we develop a computationally efficient procedure for state smoothing in Gaussian linear state space models. We show how to exploit the special structure of state-space models to draw latent states efficiently. We analyze the computational efficiency of Kalman-filter-based methods, the Cholesky Factor Algorithm, and our new method using counts of operations and computational experiments. We show that for many important cases, our method is most efficient. Gains are particularly large for cases where the dimension of observed variables is large or where one makes repeated draws of states for the same parameter values. We apply our method to a multivariate Poisson model with time-varying intensities, which we use to analyze financial market transaction count data. In the second chapter, we propose a new technique for the analysis of multivariate stochastic volatility models, based on efficient draws of volatility from its conditional posterior distribution. It applies to models with several kinds of cross-sectional dependence. Full VAR coefficient and covariance matrices give cross-sectional volatility dependence. Mean factor structure allows conditional correlations, given states, to vary in time. The conditional return distribution features Student's t marginals, with asset-specific degrees of freedom, and copulas describing cross-sectional dependence. We draw volatility as a block in the time dimension and one-at-a-time in the cross-section. Following McCausland(2012), we use close approximations of the conditional posterior distributions of volatility blocks as Metropolis-Hastings proposal distributions. We illustrate using daily return data for ten currencies. We report results for univariate stochastic volatility models and two multivariate models. In the third chapter, we evaluate the information contributed by (variations of) realized volatility to the estimation and forecasting of volatility when prices are measured with and without error using a stochastic volatility model. We consider the viewpoint of an investor for whom volatility is an unknown latent variable and realized volatility is a sample quantity which contains information about it. We use Bayesian Markov Chain Monte Carlo (MCMC) methods to estimate the models, which allow the formulation of the posterior densities of in-sample volatilities, and the predictive densities of future volatilities. We then compare the volatility forecasts and hit rates from predictions that use and do not use the information contained in realized volatility. This approach is in contrast with most of the empirical realized volatility literature which most often documents the ability of realized volatility to forecast itself. Our empirical applications use daily index returns and foreign exchange during the 2008-2009 financial crisis.

在這份研究中，我們提出一個類分散式神經網路框架，此框架為高頻交易系統研究下之子專案。在系統中，我們透過資料探勘程序發掘財務時間序列中的模式，其中所採用的資料探勘演算法之一即為類神經網路。我們實作一個在分散式平台上訓練類神經網路的框架。我們採用Apache Spark來建立底層的運算叢集，因為它提供高效能的記憶體內運算（in-memory computing）。我們分析一些分散式後向傳導演算法（特別是用來預測財務時間序列的），加以調整，並將其用於我們的框架。我們提供了許多細部的選項，讓使用者在進行類神經網路建模時有很高的彈性。 / In this research, we introduce a distributed framework of artificial neural network (ANN) as a subproject under the research of a high-frequency trading (HFT) system. In the system, ANNs are used in the data mining process for identifying patterns in financial time series. We implement a framework for training ANNs on a distributed computing platform. We adopt Apache Spark to build the base computing cluster because it is capable of high performance in-memory computing. We investigate a number of distributed backpropagation algorithms and techniques, especially ones for time series prediction, and incorporate them into our framework with some modifications. With various options for the details, we provide the user with flexibility in neural network modeling.

The main goal of this thesis is to develop practical digital controller architectures for multi-phase dc-dc converters utilized in low power (up to few hundred watts) and cost-sensitive applications. The proposed controllers are suitable for on-chip integration while being capable of providing advanced features, such as dynamic efficiency optimization, inductor current estimation, converter component identification, as well as combined dynamic current sharing and fast transient response. The first part of this thesis addresses challenges related to the practical implementation of digital controllers for low-power multi-phase dc-dc converters. As a possible solution, a multi-use high-frequency digital PWM controller IC that can regulate up to four switching converters (either interleaved or standalone) is presented. Due to its configurability, low current consumption (90.25 μA/MHz per phase), fault-tolerant work, and ability to operate at high switching frequencies (programmable, up to 10 MHz), the IC is suitable to control various dc-dc converters. The applications range from dc-dc converters used in miniature battery-powered electronic devices consuming a fraction of watt to multi-phase dedicated supplies for communication systems, consuming hundreds of watts. A controller for multi-phase converters with unequal current sharing is introduced and an efficiency optimization method based on logarithmic current sharing is proposed in the second part. By forcing converters to operate at their peak efficiencies and dynamically adjusting the number of active converter phases based on the output load current, a significant improvement in efficiency over the full range of operation is obtained (up to 25%). The stability and inductor current transition problems related to this mode of operation are also resolved. At last, two reconfigurable digital controller architectures with multi-parameter estimation are introduced. Both controllers eliminate the need for external analog current/temperature sensing circuits by accurately estimating phase inductor currents and identifying critical phase parameters such as equivalent resistances, inductances and output capacitance. A sensorless non-linear, average current-mode controller is introduced to provide fast transient response (under 5 μs), small voltage deviation and dynamic current sharing with multi-phase converters. To equalize the thermal stress of phase components, a conduction loss-based current sharing scheme is proposed and implemented.

No description available.

Made available in DSpace on 2014-12-17T15:28:53Z (GMT). No. of bitstreams: 1 AndersonBS_DISSERT.pdf: 4240084 bytes, checksum: 0331343a1aab5e54d0d9cb6baeccb72d (MD5) Previous issue date: 2013-12-13 / Epilepsies are neurological disorders characterized by recurrent and spontaneous seizures due to an abnormal electric activity in a brain network. The mesial temporal lobe epilepsy (MTLE) is the most prevalent type of epilepsy in adulthood, and it occurs frequently in association with hippocampal sclerosis. Unfortunately, not all patients benefit from pharmacological treatment (drug-resistant patients), and therefore become candidates for surgery, a procedure of high complexity and cost. Nowadays, the most common surgery is the anterior temporal lobectomy with selective amygdalohippocampectomy, a procedure standardized by anatomical markers. However, part of patients still present seizure after the procedure. Then, to increase the efficiency of this kind of procedure, it is fundamental to know the epileptic human brain in order to create new tools for auxiliary an individualized surgery procedure. The aim of this work was to identify and quantify the occurrence of epilepticform activity -such as interictal spikes (IS) and high frequency oscillations (HFO) - in electrocorticographic (ECoG) signals acutely recorded during the surgery procedure in drug-resistant patients with MTLE. The ECoG recording (32 channels at sample rate of 1 kHz) was performed in the surface of temporal lobe in three moments: without any cortical resection, after anterior temporal lobectomy and after amygdalohippocampectomy (mean duration of each record: 10 min; N = 17 patients; ethic approval #1038/03 in Research Ethic Committee of Federal University of S?o Paulo). The occurrence of IS and HFO was quantified automatically by MATLAB routines and validated manually. The events rate (number of events/channels) in each recording time was correlated with seizure control outcome. In 8 hours and 40 minutes of record, we identified 36,858 IS and 1.756 HFO. We observed that seizure-free outcome patients had more HFO rate before the resection than non-seizure free, however do not differentiate in relation of frequency, morphology and distribution of IS. The HFO rate in the first record was better than IS rate on prediction of seizure-free patients (IS: AUC = 57%, Sens = 70%, Spec = 71% vs HFO: AUC = 77%, Sens = 100%, Spec = 70%). We observed the same for the difference of the rate of pre and post-resection (IS: AUC = 54%, Sens = 60%, Spec = 71%; vs HFO: AUC = 84%, Sens = 100%, Spec = 80%). In this case, the algorithm identifies all seizure-free patients (N = 7) with two false positives. To conclude, we observed that the IS and HFO can be found in intra-operative ECoG record, despite the anesthesia and the short time of record. The possibility to classify the patients before any cortical resection suggest that ECoG can be important to decide the use of adjuvant pharmacological treatment or to change for tailored resection procedure. The mechanism responsible for this effect is still unknown, thus more studies are necessary to clarify the processes related to it / As epilepsias s?o dist?rbios neurol?gicos caracterizados por crises espont?neas e recorrentes, resultantes de uma atividade el?trica anormal de uma rede neural. Dentre os diferentes tipos de epilepsia, a epilepsia mesial do lobo temporal (EMLT) ? a mais observada em adultos, sendo frequentemente associada ? esclerose hipocampal. Infelizmente, nem todos os pacientes s?o beneficiados pelo tratamento farmacol?gico (pacientes f?rmaco-resistentes). Para estes sujeitos, uma alternativa ? a realiza??o de cirurgia, um procedimento de alta complexidade e elevado custo. Atualmente, o procedimento mais realizado ? a lobectomia temporal anterior com amigdalo-hipocampectomia seletiva, uma cirurgia padronizada por marcos anat?micos. Entretanto, uma parcela dos pacientes continua a apresentar crises incapacitantes ap?s o tratamento cir?rgico. Desta forma, para aumentar a efici?ncia deste tipo de tratamento, ? fundamental a compreens?o do enc?falo humano epil?ptico com vistas a se criar ferramentas que auxiliem na realiza??o de procedimentos individualizados. O objetivo do presente trabalho foi identificar e quantificar a ocorr?ncia de atividade epileptiforme - esp?culas interictais (EI) e oscila??es de alta frequ?ncia (OAF) - em registros eletrocorticogr?ficos (ECoG) realizados durante procedimento cir?rgico em pacientes com EMLT refrat?ria ao tratamento farmacol?gico. Registros ECoG (32 canais a uma taxa de amostragem de 1 kHz) foram realizados na superf?cie do lobo temporal em 3 momentos cir?rgicos: no c?rtex intacto, ap?s lobectomia temporal anterior e ap?s amigdalo-hipocampectomia (dura??o m?dia de cada um desses registros: 10 min; N=17 pacientes). A ocorr?ncia de EI e OAF foi quantificada automatica-mente, por meio de rotinas em MATLAB, e validadas manualmente. A taxa de ocorr?ncia em cada um dos tempos cir?rgicos foi correlacionada com o resultado cir?rgico quanto ao controle das crises, num seguimento de 2 anos. De um total de 8 h e 40 min de registro, identificamos 36.858 EI e 1.756 OAF. Observamos que os pacientes que ficaram livres de crises no p?s-operat?rio apresentaram maior quanti-dade de OAF antes da cirurgia do que aqueles que continuaram a ter crises; por?m, n?o diferiram quanto a frequ?ncia, morfologia e distribui??o de EI. A ocorr?ncia de OAF no registro basal apresentou melhor desempenho que as EI na previs?o do controle total das crises no p?s-operat?rio (EI: AUC = 57%, S = 71% , E = 70% vs OAF: AUC = 77%, S = 100%, E=70%). O mesmo foi observado com a varia??o da ocorr?ncia entre os momentos pr?- e p?s-ressec??o (EI: AUC = 54%, S = 71%, E = 60% vs OAF: AUC = 84%, S = 100%, E = 80%). Nesse caso, o classificador foi capaz de identificar todos os pacientes livres de crises (N = 7) , apresentando apenas dois falsos positivos. Desta forma, podemos concluir que as OAF, juntamente com as EI, podem ser encontradas no registro ECoG intra-operat?rio, mesmo na presen?a de anest?sicos e em uma curta sess?o de registro. Al?m disso, a observa??o de que a ocorr?ncia desses eventos no in?cio da cirurgia permite classificar o paciente quanto ao progn?stico cir?rgico abre caminho para aplicar o ECoG intra-operat?rio, por exemplo, na decis?o sobre o uso de tratamento farmacol?gico adjuvante ou da convers?o para ressec??es individualizadas. No entanto, o mecanismo respons?vel por esse efeito ainda ? desconhecido, logo novos estudos s?o necess?rios para melhor esclarec?-lo