Modeling Vegetable Fractals by means of Fractional-order Equations

This paper characterizes four ‘fractal vegetables’: (i) cauliflower (brassica oleracea var. Botrytis); (ii) broccoli (brassica oleracea var. italica); (iii) round cabbage (brassica oleracea var. capitata) and (iv) Brussels sprout (brassica oleracea var. gemmifera), by means of electrical impedance spectroscopy and fractional calculus tools. Experimental data is approximated using fractional-order models and the corresponding parameters are determined with a genetic algorithm. The Havriliak-Negami five-parameter model fits well into the data, demonstrating that classical formulae can constitute simple and reliable models to characterize biological structures.

Modelling HIV/AIDS length of stay in Portugal

Tese apresentada como requisito parcial para obtenção do grau de Doutor em Estatística e Gestão de Informação pelo Instituto Superior de Estatística e Gestão de Informação da Universidade Nova de Lisboa

Optimization of the electrodialytic phosphorus recovery from sewage sludge ash

Phosphorus (P) is becoming a scarce element due to the decreasing availability of primary sources. Therefore, recover P from secondary sources, e.g. waste streams, have become extremely important. Sewage sludge ash (SSA) is a reliable secondary source of P. The use of SSAs as a direct fertilizer has very restricted legislation due to the presence of inorganic contaminants. Furthermore, the P present in SSAs is not in a plant-available form. The electrodialytic (ED) process is one of the methods under development to recover P and simultaneously remove heavy metals. The present work aimed to optimize the P recovery through a 2 compartment electrodialytic cell. The research was divided in three independent phases. In the first phase, ED experiments were carried out for two SSAs from different seasons, varying the duration of the ED process (2, 4, 6 and 9 days). During the ED treatment the SSA was suspended in distilled water in the anolyte, which was separated from the catholyte by a cation exchange membrane. From both ashes 90% of P was successfully extracted after 6 days of treatment. Regarding the heavy metals removal, one of the SSAs had a better removal than the other. Therefore, it was possible to conclude that SSAs from different seasons can be submitted to ED process under the same parameters. In the second phase, the two SSAs were exposed to humidity and air prior to ED, in order to carbonate them. Although this procedure was not successful, ED experiments were carried out varying the duration of the treatment (2 and 6 days) and the period of air exposure that SSAs were submitted to (7, 14 and 30 days). After 6 days of treatment and 30 days of air exposure, 90% of phosphorus was successfully extracted from both ashes. No differences were identified between carbonated and non-carbonated SSAs. Thus, SSAs that were exposed to the air and humidity, e.g. SSAs stored for 30 days in an open deposit, can be treated under the same parameters as the SSAs directly collected from the incineration process. In the third phase, ED experiments were carried out during 6 days varying the stirring time (0, 1, 2 and 4 h/day) in order to investigate if energy can be saved on the stirring process. After 6 days of treatment and 4 h/day stirring, 80% and 90% of P was successfully extracted from SSA-A and SSA-B, respectively. This value is very similar to the one obtained for 6 days of treatment stirring 24 h/day.

Gestão de crises de âmbito policial

The issues concerning Crisis Situations under the scope of police activity, raised after incidents considered critical, has emerged with greater intensity during the most recent decades, posing a major challenge for police forces around the world. These are situations or events of crucial importance, involving hostage taken or barricaded individuals, in which inevitably human lives are at risk, requiring from law enforcement agencies a specific response capability, i.e., a type of intervention not framed under the parameters considered as routine, in order to obtain solutions to minimize the possibility of casualties. Because this is about impacting situations of extreme gravity, where the preservation of human lives is concerned and, in many cases, the very Rule of Law as well, we understand the need for police forces to adapt to new procedures and working methods. Such procedures are an enormously complex task that requires the coordination and articulation of several components, including not infrequently the performance of different police forces, as well as organizations and entities with varied powers and duties, which implies the need for effective management. This explains the emergence of Crisis Management Structures, imposing to determine which are their fundamental components, their importance, how they interconnect, and their major goal. The intrinsic features will also be analyzed in the aspect that we consider to be the fundamental groundwork of a Crisis Management Structure, i.e., Negotiation itself, considering it as a kind of police intervention, where a wide range of procedures feeds a channel of dialogue, aiming at minimizing the damage resultant from an extreme action, in particular, to prevent the death of any of those involved. This is in essence the path we have chosen to develop this study, trying to find out an answer to the fundamental question: What model of Crisis Management Structure should be adopted to manage a critical event involving hostage negotiation?

pH como un trazador de la variabilidad biogeoquímica en el Sistema de Humboldt

Se analizó información sobre parámetros biogeoquímicos como pH, CO2 parcialmente disuelto, clorofila-a, oxígeno, salinidad y temperatura, obtenidos en el Sistema de Humboldt frente a Perú en los años 2007-2009, a fin de comprender la importancia del pH como una variable clave para trazar la variabilidad biogeoquímica del Sistema de Afloramiento de Humboldt. Las Aguas Costeras Frías (ACF) ricas en nutrientes evidencian una relación inversa con el pH, en ese sentido valores de 7,6 a 8,0 se encontraron asociados a la ocurrencia de eventos de afloramiento; en condiciones frías como La Niña 2007 el pH predominante se encuentra por debajo de 8,0, y en el caso de las Aguas Subtropicales Superficiales presentan un pH entorno a 8,3. La productividad en los veranos se incrementa debido a factores como la intensidad de luz generándose núcleos de clorofila-a superiores a 10,0 μg.L-1; como constituyentes de la reacción bioquímica de fotosíntesis se tiene en cuenta la forma inmediata del compuesto CO2 y el consumo de iones hidronio; en situaciones extremas de fertilización, se tiene elevado pH entre 8,2 e inclusive 9,0, en donde los contenidos de oxígeno disuelto superan los 6,0 mL.L-1.

A comparative study of ballpoint ink ageing parameters using GC/MS.

For more than a decade scientists tried to develop methods capable of dating ink by monitoring the loss of phenoxyethanol (PE) over time. While many methods were proposed in the literature, few were really used to solve practical cases and they still raise much concern within the scientific community. In fact, due to the complexity of ink drying processes it is particularly difficult to find a reliable ageing parameter to reproducibly follow ink ageing. Moreover, systematic experiments are required in order to evaluate how different factors actually influence the results over time. Therefore, this work aimed at evaluating the capacity of four different ageing parameters to reliably follow ink ageing over time: (1) the quantity of solvent PE in an ink line, (2) the relative peak area (RPA) normalising the PE results using stable volatile compounds present in the ink formulation, (3) the solvent loss ratio (R%) calculated from PE results obtained by the analyses of naturally and artificially aged samples, (4) a modified solvent loss ratio version (R%*) calculated from RPA results. After the determination of the limits of reliable measurements of the analytical method, the repeatability of the different ageing parameters was evaluated over time, as well as the influence of ink composition, writing pressure and storage conditions on the results. Surprisingly, our results showed that R% was not the most reliable parameter, as it showed the highest standard deviation. Discussion of the results in an ink dating perspective suggests that other proposed parameters, such as RPA values, may be more adequate to follow ink ageing over time.

Modeling reaction kinetics and mass transfer in ozonation in water solutions

This dissertation is based on four articles dealing with modeling of ozonation. The literature part of this considers some models for hydrodynamics in bubble column simulation. A literature review of methods for obtaining mass transfer coefficients is presented. The methods presented to obtain mass transfer are general models and can be applied to any gas-liquid system. Ozonation reaction models and methods for obtaining stoichiometric coefficients and reaction rate coefficients for ozonation reactions are discussed in the final section of the literature part. In the first article, ozone gas-liquid mass transfer into water in a bubble column was investigated for different pH values. A more general method for estimation of mass transfer and Henry’s coefficient was developed from the Beltrán method. The ozone volumetric mass transfer coefficient and the Henry’s coefficient were determined simultaneously by parameter estimation using a nonlinear optimization method. A minor dependence of the Henry’s law constant on pH was detected at the pH range 4 - 9. In the second article, a new method using the axial dispersion model for estimation of ozone self-decomposition kinetics in a semi-batch bubble column reactor was developed. The reaction rate coefficients for literature equations of ozone decomposition and the gas phase dispersion coefficient were estimated and compared with the literature data. The reaction order in the pH range 7-10 with respect to ozone 1.12 and 0.51 the hydroxyl ion were obtained, which is in good agreement with literature. The model parameters were determined by parameter estimation using a nonlinear optimization method. Sensitivity analysis was conducted using object function method to obtain information about the reliability and identifiability of the estimated parameters. In the third article, the reaction rate coefficients and the stoichiometric coefficients in the reaction of ozone with the model component p-nitrophenol were estimated at low pH of water using nonlinear optimization. A novel method for estimation of multireaction model parameters in ozonation was developed. In this method the concentration of unknown intermediate compounds is presented as a residual COD (chemical oxygen demand) calculated from the measured COD and the theoretical COD for the known species. The decomposition rate of p-nitrophenol on the pathway producing hydroquinone was found to be about two times faster than the p-nitrophenol decomposition rate on the pathway producing 4- nitrocatechol. In the fourth article, the reaction kinetics of p-nitrophenol ozonation was studied in a bubble column at pH 2. Using the new reaction kinetic model presented in the previous article, the reaction kinetic parameters, rate coefficients, and stoichiometric coefficients as well as the mass transfer coefficient were estimated with nonlinear estimation. The decomposition rate of pnitrophenol was found to be equal both on the pathway producing hydroquinone and on the path way producing 4-nitrocathecol. Comparison of the rate coefficients with the case at initial pH 5 indicates that the p-nitrophenol degradation producing 4- nitrocathecol is more selective towards molecular ozone than the reaction producing hydroquinone. The identifiability and reliability of the estimated parameters were analyzed with the Marcov chain Monte Carlo (MCMC) method. @All rights reserved. No part of the publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior permission of the author.

The shifting terrain of feminist theory and activism : university-based women's centres and third wave feminism

This thesis, based on the results of an organizational ethnography of a university-based feminist organization in Southern Ontario (the Centre), traces how third wave feminism is being constituted in the goals, initiatives, mandate, organizational structure, and overall culture of university-based feminist organizations. I argue that, from its inception, the meanings and goals of the Centre have been contested through internal critique, reflection, and discussion inspired by significant shifts in feminist theory that challenge the fundamental principles of second wave feminism. I identify a major shift in the development and direction of the Centre that occurs in two distinct phases. The first phase of the shift occurs with the emergence of an antioppression framework, which broadens the Centre's mandate beyond gender and sexism to consider multiple axes of identity and oppression that affect women's lives. The second phase of this shift is characterized by a focus on (trans) inclusion and accessibility and has involved changing the Centre's name so that it is no longer identified as a women's centre in order to reflect more accurately its focus on mUltiple axes of identity and oppression. Along with identifying two phases of a major shift in the direction of the Centre, I trace two discourses about its development. The dominant discourse of the Centre's development is one of progress and evolution. The dominant discourse characterizes the Centre as a dynamic feminist organization that consistently strives to be more inclusive and diverse. The reverse discourse undermines the dominant discourse by emphasizing that, despite the Centre's official attempts to be inclusive and to build diversity, little has actually changed, leaving women of colour marginalized in the Centre's dominant culture of whiteness. This research reveals that, while many of their strategies have unintended (negative) consequences, members of the Centre are working to build an inclusive politics of resistance that avoids the mistakes of earlier feminist movements and organizations. These members, along with other activists, actively constitute third wave feminism in a process that is challenging, contradictory, and often painful. A critical analysis of this process and the strategies it involves provides an opportunity for activists to reflect on their experiences and develop new strategies in an effort to further struggles for social justice and equity.

Solute effects on the production and decay of primary species in the flash photolysis of indole

Maximum production rates ofs and decay kinetics for the hydrated electron, the indolyl neutral radical and the indole triplet state have been obtained in the microsecond, broadband (X > 260 nm) flash photolysis of helium-saturated, neutral aqueous solutions of indole, in the absence and in the presence of the solutes NaBr, BaCl2*2H20 and CdSCV Fluorescence spectra and fluorescence lifetimes have also been obtained in the absence and in the presence of the above solutes, The hydrated electron is produced monophotonically and biphotonically at an apparent maximum rate which is increased by BaCl2*2H20 and decreased by NaBr and CdSOif. The neutral indolyl radical may be produced monophotonically and biphotonically or strictly monophotonically at an apparent maximum rate which is increased by NaBr and CdSO^ and is unaffected by BaCl2*2H20. The indole triplet state is produced monophotonically at a maximum rate which is increased by all solutes. The hydrated electron decays by pseudo first order processes, the neutral indolyl radical decays by second order recombination and the indole triplet state decays by combined first and second order processes. Hydrated electrons are shown to react with H , H2O, indole, Na and Cd"*""1"". No evidence has been found for the reaction of hydrated electrons with Ba . The specific rate of second order neutral indolyl radical recombination is unaffected by NaBr and BaCl2*2H20, and is increased by CdSO^. Specific rates for both first and second order triplet state decay processes are increased by all solutes. While NaBr greatly reduced the fluorescence lifetime and emission band intensity, BaCl2*2H20 and CdSO^ had no effect on these parameters. It is suggested that in solute-free solutions and in those containing BaCl2*2H20 and CdSO^, direct excitation occurs to CTTS states as well as to first excited singlet states. It is further suggested that in solutions containing NaBr, direct excitation to first excited singlet states predominates. This difference serves to explain increased indole triplet state production (by ISC from CTTS states) and unchanged fluorescence lifetimes and emission band intensities in the presence of BaCl2*2H20 and CdSOt^., and increased indole triplet state production (by ISC from S^ states) and decreased fluorescence lifetime and emission band intensity in the presence of NaBr. Evidence is presented for (a) very rapid (tx ^ 1 us) processes involving reactions of the hydrated electron with Na and Cd which compete with the reformation of indole by hydrated electron-indole radical cation recombination, and (b) first and second order indole triplet decay processes involving the conversion of first excited triplet states to vibrationally excited ground singlet states.

The determination of sinigrin in Brassica, and investigations into the use of allyl-isothiocyanate as a nematicide

The goal of this thesis was to study factors related to the development of Brassica juncea as a sustainable nematicide. Brassica juncea is characterized by the glycoside (glucosinolate) sinigrin. Various methods were developed for the determination of sinigrin in Brassica juncea tissue extracts. Sinigrin concentrations in plant tissues at various stages of growth were monitored. Sinigrin enzymatically breaks down into allylisothiocyanate (AITC). AITC is unstable in aqueous solution and degradation was studied in water and in soil. Finally, the toxicity of AITC against the root-lesion nematode (Pratylenchus penetrans) was determined. A method was developed to extract sinigrin from whole Brassica j uncea tissues. The optimal time of extraction wi th boiling phosphate buffer (0.7mM, pH=6.38) and methanol/water (70:30 v/v) solutions were both 25 minutes. Methanol/water extracted 13% greater amount of sinigrin than phosphate buffer solution. Degradation of sinigrin in boiling phosphate buffer solution (0.13%/minute) was similar to the loss of sinigrin during the extraction procedure. The loss of sinigrin from boiling methanol/water was estimated to be O.Ol%/minute. Brassica juncea extract clean up was accomplished by an ion-pair solid phase extraction (SPE) method. The recovery of sinigrin was 92.6% and coextractive impurities were not detected in the cleaned up extract. Several high performance liquid chromatography (HPLC) methods were developed for the determination of sinigrin. All the developed methods employed an isocratic mobile phase system wi th a low concentration of phosphate buffer solution, ammonium acetate solution or an ion-pair reagent solution. A step gradient system was also developed. The method involved preconditioning the analytical column with phosphate buffer solution and then switching the mobile phase to 100% water after sample injection.Sinigrin and benzyl-glucosinolate were both studied by HPLC particle beam negative chemical ionization mass spectrometry (HPLCPB- NCI-MS). Comparison of the mass spectra revealed the presence of fragments arising from the ~hioglucose moiety and glucosinolate side-chain. Variation in the slnlgrin concentration within Brassica juncea plants was studied (Domo and Cutlass cuItivars). The sinigrin concentration in the top three leaves was studied during growth of each cultivar. For Cutlass, the minimum (200~100~g/g) and maximum (1300~200~g/g) concentrations were observed at the third and seventh week after planting, respectively. For Domo, the minimum (190~70~g/g) and maximum (1100~400~g/g) concentrations were observed at the fourth and eighth week after planting, respectively. The highest sinigrin concentration was observed in flower tissues 2050±90~g/g and 2300±100~g/g for Cutlass and Domo cultivars, respectively. Physical properties of AITC were studied. The solubility of AITC in water was determined to be approximately 1290~g/ml at 24°C. An HPLC method was developed for the separation of degradation compounds from aqueous AITC sample solutions. Some of the degradation compounds identified have not been reported in the literature: allyl-thiourea, allyl-thiocyanate and diallyl-sulfide. In water, AITC degradation to' diallyl-thiourea was favored at basic pH (9.07) and degradation to diallyl-sulfide was favored at acidic pH (4 . 97). It wap necessary to amend the aqueous AITC sample solution with acetonitrile ?efore injection into the HPLC system. The acetonitrile amendment considerably improved AITC recovery and the reproducibility of the results. The half-life of aqueous AITC degradation at room temperature did not follow first-order kinetics. Beginning with a 1084~g/ml solution, the half-life was 633 hours. Wi th an ini tial AITC concentration of 335~g/ml the half-life was 865 hours. At 35°C the half-life AITC was 76+4 hours essentially independent of the iiisolution pH over the range of pH=4.97 to 9.07 (1000~g/ml). AITC degradation was also studied in soil at 35°C; after 24 hours approximately 75% of the initial AITC addition was unrecoverable by water extraction. The ECso of aqueous AITC against the root-lesion nematode (Pratylenchus penetrans) was determined to be approximately 20~g/ml at one hour exposure of the nematode to the test solution. The toxicological study was also performed with a myrosinase treated Brassica juncea extract. Myrosinase treatment of the Brassica juncea extract gave nearly quantitative conversion of sinigrin into AITC. The myrosinase treated extract was of the same efficacy as an aqueous AITC solution of equivalent concentration. The work of this thesis was focused upon understanding parameters relevant to the development of Brassica juncea as a sustainable nematicide. The broad range of experiments were undertaken in support of a research priority at Agriculture and Agri-Food Canada.

Subjective, behavioural and electrophysiological measurements of the time-course and intensity of sleep inertia after a full night of sleep /

Several recent studies have described the period of impaired alertness and performance known as sleep inertia that occurs upon awakening from a full night of sleep. They report that sleep inertia dissipates in a saturating exponential manner, the exact time course being task dependent, but generally persisting for one to two hours. A number of factors, including sleep architecture, sleep depth and circadian variables are also thought to affect the duration and intensity. The present study sought to replicate their findings for subjective alertness and reaction time and also to examine electrophysiological changes through the use of event-related potentials (ERPs). Secondly, several sleep parameters were examined for potential effects on the initial intensity of sleep inertia. Ten participants spent two consecutive nights and subsequent mornings in the sleep lab. Sleep architecture was recorded for a fiiU nocturnal episode of sleep based on participants' habitual sleep patterns. Subjective alertness and performance was measured for a 90-minute period after awakening. Alertness was measured every five minutes using the Stanford Sleepiness Scale (SSS) and a visual analogue scale (VAS) of sleepiness. An auditory tone also served as the target stimulus for an oddball task designed to examine the NlOO and P300 components ofthe ERP waveform. The five-minute oddball task was presented at 15-minute intervals over the initial 90-minutes after awakening to obtain six measures of average RT and amplitude and latency for NlOO and P300. Standard polysomnographic recording were used to obtain digital EEG and describe the night of sleep. Power spectral analyses (FFT) were used to calculate slow wave activity (SWA) as a measure of sleep depth for the whole night, 90-minutes before awakening and five minutes before awakening.

An examination of the effects on learning seen in children afforded the opportunity to control the order of repetitions for three novel spatiotemporal sequences

Children were afforded the opportunity to control the order of repetitions for three novel spatiotemporal sequences. The following was predicted: a) children and adults in the self-regulated (SELF) groups would produce faster movement (MT) and reaction times (R T) and greater recall success (RS) during retention compared to the age-matched yoked (YOKE) groups; b) children would choose to switch sequences less often than adults; c) adults would produce faster MT and RT and greater RS than the children during acquisition and retention, independent of experimental group. During acquisition, no effects were seen for RS, however for MT and RT there was a main effect for age as well as block. During retention a main effect for practice condition was seen for RS and failed to reach statistical significance for MT and RT, thus partially supporting our first and second hypotheses. The third hypothesis was not supported.

Asymmetric Smiles, Leverage Effects and Structural Parameters.

In this paper, we characterize the asymmetries of the smile through multiple leverage effects in a stochastic dynamic asset pricing framework. The dependence between price movements and future volatility is introduced through a set of latent state variables. These latent variables can capture not only the volatility risk and the interest rate risk which potentially affect option prices, but also any kind of correlation risk and jump risk. The standard financial leverage effect is produced by a cross-correlation effect between the state variables which enter into the stochastic volatility process of the stock price and the stock price process itself. However, we provide a more general framework where asymmetric implied volatility curves result from any source of instantaneous correlation between the state variables and either the return on the stock or the stochastic discount factor. In order to draw the shapes of the implied volatility curves generated by a model with latent variables, we specify an equilibrium-based stochastic discount factor with time non-separable preferences. When we calibrate this model to empirically reasonable values of the parameters, we are able to reproduce the various types of implied volatility curves inferred from option market data.

Les Réseaux de la Pénombre:Typologie de l'aide reçue par les Personnes Âgées

Nous nous sommes intéressés à l’analyse et à la mise à jour d’une typologie de l’aide reçue par les personnes âgées de 65 ans et plus vivant à domicile. Cette étude secondaire s’est basée sur les données recueuillies dans deux milieux francophones, Hochelaga-Maisonneuve (HM) et Moncton (MCT). La collecte de données avait été faite par l’entremise d’un questionnaire administré par entrevue face à face. Les deux objectifs, de cette thèse sont : 1) Établir une typologie des réseaux d’aide, résultant de la combinaison des sources d’aide et des tâches accomplies ; 2) Identifier les principaux déterminants d’appartenance aux réseaux. La typologie obtenue met en relation les ressources, formelles ou informelles, utilisées par les personnes âgées et l’aide instrumentale reçue. La capacité ou l’incapacité à effectuer neuf activités de la vie quotidienne et huit de la vie domestique ont servi à évaluer l’aide reçue. Six ressources formelles et dix informelles ont été examinées selon qu’elles étaient les 1ères, 2ièmes ou 3ièmes sources d’aide utilisées par les personnes âgées. L’approche privilégiée s’est inspirée de celle des réseaux sociaux et du modèle de Pescosolido. C’est l’influence des caractéristiques sociodémographiques des personnes âgées, de leurs états de santé, de leurs habitudes de vie sur leurs réseaux qui nous ont intéressés. Les résultats sont présentés à chaque fois pour nos deux milieux séparément. Nous commençons par un descriptif des sources d’aide utilisées et des aides reçues. Puis les profils des sources d’aide utilisées et des activités accomplies sont exposés pour l’ensemble des personnes âgées. Ces profils servent de base pour obtenir notre typologie. Elle comprend cinq catégories. Ces catégories sont toutes composées de personnes âgées faisant appel à de l’aide formelle, informelle ou mixte pour accomplir des tâches uniques ou multiples. La première catégorie « Transitoire », comprend 39% (HM) et 46% (MCT) des personnes âgées qui débute un processus d’incapacité. Elles font appel à des ressources informelles pour accomplir une tâche unique. La deuxième catégorie « Personnes âgées seules » en rassemble 14% (HM et MCT), majoritairement des femmes, avec peu d’incapacités. Ces dernières utilisent de l’aide formelle pour une tâche unique. La troisième catégorie « Familiale » regroupe 12% (HM et MCT) des personnes âgées bien entourées qui ont plusieurs incapacités. Ces gens font appel à des sources d’aide informelles pour réaliser des tâches multiples. La quatrième catégorie « Très fragile » rassemble 30% (HM) et 25% (MCT) des personnes âgées peu entourées ayant beaucoup d’incapacités. Elles utilisent des ressources d’aide mixtes pour effectuer des tâches multiples. La cinquième catégorie « Pré institutionnel » comprend 4% (HM et MCT) des personnes âgées qui ont le plus d’incapacités et qui sont seules. Ces gens font appel à de l’aide formelle pour des tâches multiples. Les déterminants d’appartenance à ces catégories proviennent des blocs sociodémographiques, état de santé et réseaux sociaux de notre modèle théorique. Une des contributions importantes de cette thèse a été de pouvoir identifier cinq catégories bien distinctes composant une typologie de l’aide reçue, indépendamment du milieu, par des personnes âgées vivant à domicile. MOTS CLÉS : Typologie, réseaux sociaux, personnes âgées, services de soins, formels, informels, aides reçues, sources d’aide, incapacités, déterminants d’appartenance, fragilité

Training deep convolutional architectures for vision

Les tâches de vision artificielle telles que la reconnaissance d’objets demeurent irrésolues à ce jour. Les algorithmes d’apprentissage tels que les Réseaux de Neurones Artificiels (RNA), représentent une approche prometteuse permettant d’apprendre des caractéristiques utiles pour ces tâches. Ce processus d’optimisation est néanmoins difficile. Les réseaux profonds à base de Machine de Boltzmann Restreintes (RBM) ont récemment été proposés afin de guider l’extraction de représentations intermédiaires, grâce à un algorithme d’apprentissage non-supervisé. Ce mémoire présente, par l’entremise de trois articles, des contributions à ce domaine de recherche. Le premier article traite de la RBM convolutionelle. L’usage de champs réceptifs locaux ainsi que le regroupement d’unités cachées en couches partageant les même paramètres, réduit considérablement le nombre de paramètres à apprendre et engendre des détecteurs de caractéristiques locaux et équivariant aux translations. Ceci mène à des modèles ayant une meilleure vraisemblance, comparativement aux RBMs entraînées sur des segments d’images. Le deuxième article est motivé par des découvertes récentes en neurosciences. Il analyse l’impact d’unités quadratiques sur des tâches de classification visuelles, ainsi que celui d’une nouvelle fonction d’activation. Nous observons que les RNAs à base d’unités quadratiques utilisant la fonction softsign, donnent de meilleures performances de généralisation. Le dernière article quand à lui, offre une vision critique des algorithmes populaires d’entraînement de RBMs. Nous montrons que l’algorithme de Divergence Contrastive (CD) et la CD Persistente ne sont pas robustes : tous deux nécessitent une surface d’énergie relativement plate afin que leur chaîne négative puisse mixer. La PCD à "poids rapides" contourne ce problème en perturbant légèrement le modèle, cependant, ceci génère des échantillons bruités. L’usage de chaînes tempérées dans la phase négative est une façon robuste d’adresser ces problèmes et mène à de meilleurs modèles génératifs.