Effects of short-term high-fructose diets in humans: modulation by environmental factors

It is currently suspected that sugar overconsumption, and more specifically fructose, may promote the development of obesity and of several cardio-metabolic disorders. However, environmental factors, such as fish oil and dietary proteins, may prevent some deleterious effects of fructose. The aim of this thesis was to identify potential environmental factors that may modulate the metabolic effects of fructose. The first study was designed to evaluate the impact of endurance exercise in healthy young men fed a high-fructose, isocaloric diet. Fructose-induced effects on lipid profile were totally prevented by endurance exercise and may be explained by an enhanced clearance of TRL-TG and the inhibition of de novo lipogenesis. As energy intake was adjusted to energy requirement, we can conclude that exercise acts on fructose metabolism independently of energy imbalance. The second study aimed at determining whether coffee and more specifically chlorogenic acid consumption may prevent fructose-induced intrahepatic lipids accumulation, hypertriglyceridemia and hepatic insulin resistance, through a stimulation of lipid oxidation. Coffee did not prevent the fructose-induced increase in IHCL or plasma TG. Interestingly, the three coffees tested prevented the decrease in hepatic insulin sensitivity, independently of their content in caffeine or chlorogenic acid. Finally, in the third study, we evaluated the effect of essential amino acid supplementation on the increase of hepatic lipids induced by a high-fructose diet. This intervention slightly decreased IHCL concentration. The exact mechanisms remain unidentified but may involve an increased secretion of VLDL-TG. In conclusion, the environmental factors evaluated allow to prevent some of the deleterious effects of fructose and suggest that recommendations on fructose consumption should also take into account environmental factors.

Greenhouse gases and climatic states of minimum entropy production

The hypothesis of minimum entropy production is applied to a simple one-dimensional energy balance model and is analysed for different values of the radiative forcing due to greenhouse gases. The extremum principle is used to determine the planetary “conductivity” and to avoid the “diffusive” approximation, which is commonly assumed in this type of model. For present conditions the result at minimum radiative entropy production is similar to that obtained by applying the classical model. Other climatic scenarios show visible differences, with better behaviour for the extremal case

Protein turnover, ureagenesis and gluconeogenesis.

The major processes discussed below are protein turnover (degradation and synthesis), degradation into urea, or conversion into glucose (gluconeogenesis, Figure 1). Daily protein turnover is a dynamic process characterized by a double flux of amino acids: the amino acids released by endogenous (body) protein breakdown can be reutilized and reconverted to protein synthesis, with very little loss. Daily rates of protein turnover in humans (300 to 400 g per day) are largely in excess of the level of protein intake (50 to 80 g per day). A fast growing rate, as in premature babies or in children recovering from malnutrition, leads to a high protein turnover rate and a high protein and energy requirement. Protein metabolism (synthesis and breakdown) is an energy-requiring process, dependent upon endogenous ATP supply. The contribution made by whole-body protein turnover to the resting metabolic rate is important: it represents about 20 % in adults and more in growing children. Metabolism of proteins cannot be disconnected from that of energy since energy balance influences net protein utilization, and since protein intake has an important effect on postprandial thermogenesis - more important than that of fats or carbohydrates. The metabolic need for amino acids is essentially to maintain stores of endogenous tissue proteins within an appropriate range, allowing protein homeostasis to be maintained. Thanks to a dynamic, free amino acid pool, this demand for amino acids can be continuously supplied. The size of the free amino acid pool remains limited and is regulated within narrow limits. The supply of amino acids to cover physiological needs can be derived from 3 sources: 1. Exogenous proteins that release amino acids after digestion and absorption 2. Tissue protein breakdown during protein turnover 3. De novo synthesis, including amino acids (as well as ammonia) derived from the process of urea salvage, following hydrolysis and microflora metabolism in the hind gut. When protein intake surpasses the physiological needs of amino acids, the excess amino acids are disposed of by three major processes: 1. Increased oxidation, with terminal end products such as CO₂ and ammonia 2. Enhanced ureagenesis i. e. synthesis of urea linked to protein oxidation eliminates the nitrogen radical 3. Gluconeogenesis, i. e. de novo synthesis of glucose. Most of the amino groups of the excess amino acids are converted into urea through the urea cycle, whereas their carbon skeletons are transformed into other intermediates, mostly glucose. This is one of the mechanisms, essential for life, developed by the body to maintain blood glucose within a narrow range, (i. e. glucose homeostasis). It includes the process of gluconeogenesis, i. e. de novo synthesis of glucose from non-glycogenic precursors; in particular certain specific amino acids (for example, alanine), as well as glycerol (derived from fat breakdown) and lactate (derived from muscles). The gluconeogenetic pathway progressively takes over when the supply of glucose from exogenous or endogenous sources (glycogenolysis) becomes insufficient. This process becomes vital during periods of metabolic stress, such as starvation.

Decreased thermogenic response to an oral glucose load in older subjects.

The thermogenic response to a 100 g oral glucose load was studied by indirect calorimetry in 13 older persons (age range, 38-68 years) and compared with that of 16 young matched controls of similar body weight (age range, 19-30 years). The glucose-induced thermogenesis measured over 180 min and expressed as a per cent of the energy content of the glucose load was found to be reduced in the older subjects, i.e., 5.8 +/- 0.3 per cent vs 8.6 +/- 0.7 per cent, P less than 0.002). This was also accompanied by a significant decrease in the glucose oxidation rate when averaged over the same three-hour period following the glucose load, i.e., 153 mg/min vs 213 mg/min in the control subjects (P less than 0.001) despite a similar time course of glycemia. This study suggests that the thermogenic response to an oral glucose load is blunted in older people, and this may represent an additional factor that contributes to the decreased energy requirement with age and therefore to the increased propensity to obesity if energy intake is not adjusted.

Unaltered glucose-induced thermogenesis in Graves' disease.

The thermogenic response to a 100-g oral glucose challenge was studied in 12 patients with Graves' disease using continuous indirect calorimetry. Seven hyperthyroid patients were reinvestigated under the same experimental conditions after medical therapy. The mean net increase in energy expenditure (delta EE) following the glucose challenge was the same in hyperthyroid patients as compared to a control group (delta EE = +0.15 +/- 0.02 and 0.15 +/- 0.01 kcal/min, respectively) and the treated patients (delta EE = +0.11 +/- 0.03 kcal/min ns). When expressed as a percentage of the energy content of the glucose challenge, the mean glucose induced thermogenesis was similar in all three groups: 7.0 +/- 1.0%, 7.4 +/- 0.5%, and 5.5 +/- 1.3% in hyperthyroid, control subjects, and treated patients, respectively. It is concluded that the high energy requirement of hyperthyroid patients is due primarily to an elevated resting energy expenditure. The postprandial thermogenesis in itself does not contribute to the elevated fuel utilization in Graves' disease.

Total and exogenous carbohydrate oxidation in obese prepubertal children.

The aim was to explore whether the origin of carbohydrate oxidation (exogenous compared with endogenous carbohydrate) after consumption of a mixed meal was influenced by obesity in children. Ten obese prepubertal children 8 y of age (44.2 +/- 3.6 kg) were studied over 9.5 h and compared with eight normal-weight, matched control children (28.5 +/- 1.6 kg). They were fed a mixed meal containing naturally enriched [13C]carbohydrate (cane sugar and popcorn) providing 55% of the daily energy requirement as measured by 24-h resting metabolic rate. Total carbohydrate oxidation was calculated by indirect calorimetry (hood system) whereas exogenous carbohydrate oxidation was estimated from carbon dioxide production (VCO2), the isotopic enrichment of breath 13CO2, and the abundance of [13C]carbohydrate in the meal ingested. The time course of 13CO2 in breath-measured over 570 min-followed a similar pattern in both groups. Although total carbohydrate oxidation was not significantly different among the two groups, exogenous carbohydrate utilization was significantly greater (P < 0.03) and endogenous carbohydrate oxidation was significantly lower (P < 0.05) in obese compared with control children. In addition, the rate of exogenous carbohydrate oxidation expressed as a proportion of total carbohydrate oxidation was positively related to the body fat of the children (r = 0.68, P < 0.01). The study suggests that in the postprandial phase, a smaller proportion of carbohydrate oxidation is accounted for by glycogen breakdown in obese children. The sparing of endogenous glycogen may result from decreased glycogen turnover already present at an early age.

Quantum transport phenomena and shallow impurity states in CdSb

This work is dedicated to investigation of the energy spectrum of one of the most anisotropic narrow-gap semiconductors, CdSb. At the beginning of the present studies even the model of its energy band structure was not clear. Measurements of galvanomagnetic effects in wide temperature range (1.6 - 300 K) and in magnetic fields up to 30 T were chosen for clarifying of the energy spectrum in the intentionally undoped CdSb single crystals and doped with shallow impurities (In, Ag). Detection of the Shubnikov - de Haas oscillations allowed estimating the fundamental energy spectrum parameters. The shapes of the Fermi surfaces of electrons (sphere) and holes (ellipsoid), the number of the equivalent extremums for valence band (2) and their positions in the Brillouin zone were determined for the first time in this work. Also anisotropy coefficients, components of the tensor of effective masses of carriers, effective masses of density of states, nonparabolicity of the conduction and valence bands, g-factor and its anisotropy for n- and p-CdSb were estimated for the first time during these studies. All the results obtained are compared with the cyclotron resonance data and the corresponding theoretical calculations for p-CdSb. This is basic information for the analyses of the complex transport properties of CdSb and for working out the energy spectrum model of the shallow energy levels of defects and impurities in this semiconductor. It was found out existence of different mechanisms of hopping conductivity in the presence of metal - insulator transition induced by magnetic field in n- and p-CdSb. Quite unusual feature opened in CdSb is that different types of hopping conductivity may take place in the same crystal depending on temperature, magnetic field or even orientation of crystal in magnetic field. Transport properties of undoped p-CdSb samples show that the anisotropy of the resistivity in weak and strong magnetic fields is determined completely by the anisotropy of the effective mass of the holes. Temperature and magnetic field dependence of the Hall coefficient and magnetoresistance is attributed to presence of two groups of holes with different concentrations and mobilities. The analysis demonstrates that below Tcr ~ 20 K and down to ~ 6 - 7 K the low-mobile carriers are itinerant holes with energy E2 ≈ 6 meV. The high-mobile carriers, at all temperatures T < Tcr, are holes activated thermally from a deeper acceptor band to itinerant states of a shallower acceptor band with energy E1 ≈ 3 meV. Analysis of temperature dependences of mobilities confirms the existence of the heavy-hole band or a non-equivalent maximum and two equivalent maxima of the light-hole valence band. Galvanomagnetic effects in n-CdSb reveal the existence of two groups of carriers. These are the electrons of a single minimum in isotropic conduction band and the itinerant electrons of the narrow impurity band, having at low temperatures the energies above the bottom of the conduction band. It is found that above this impurity band exists second impurity band of only localized states and the energy of both impurity bands depend on temperature so that they sink into the band gap when temperature is increased. The bands are splitted by the spin, and in strong magnetic fields the energy difference between them decreases and redistribution of the electrons between the two impurity bands takes place. Mobility of the conduction band carriers demonstrates that scattering in n-CdSb at low temperatures is strongly anisotropic. This is because of domination from scattering on the neutral impurity centers and increasing of the contribution to mobility from scattering by acoustic phonons when temperature increases. Metallic conductivity in zero or weak magnetic field is changed to activated conductivity with increasing of magnetic field. This exhibits a metal-insulator transition (MIT) induced by the magnetic field due to shift of the Fermi level from the interval of extended states to that of the localized states of the electron spectrum near the edge of the conduction band. The Mott variablerange hopping conductivity is observed in the low- and high-field intervals on the insulating side of the MIT. The results yield information about the density of states, the localization radius of the resonant impurity band with completely localized states and about the donor band. In high magnetic fields this band is separated from the conduction band and lies below the resonant impurity bands.

Hybridilämmityksen vaikutus sähkön- ja lämmöntuotantoon

Tämän tutkimuksen tarkoituksena on selvittää kaukolämmityksen rinnalla käytettävän toisen lämmitysmuodon, tässä tapauksessa sähkölämmityksen, vaikutusta sähkön- ja lämmöntuotantoon. Tämä tutkimus liittyy ¿Kehittyvä kaukolämpö -hankkeen pilottiosaan. Hankkeen pilottiosassa tutkitaan hybridilämmityksen kannattavuutta ja vaikutuksia sekä kuluttajan että yhdyskunnan kannalta. Tämätutkimus jatkaa jo aikaisemmin tehtyä tutkimusta 'Hybridilämmityksen kustannusvaikutukset', jossa tutkittiin kaukolämmityksen taloudellisuutta kuluttajan kannalta elinkaarianalyysin avulla. Tämän tutkimuksen tarkoituksena on määrittää hybridilämmityksen niin taloudelliset kuin ympäristölliset vaikutukset yhdyskunnan kannalta. Yhdyskunnan osalta vaikutuksia tarkasteltiin referenssikaupungin avulla. Referenssikaupungin alkuarvot perustuvat jo aiemmin tähän pilottiosaan tehtyyn tutkimukseen 'Hybridilämmityksen kustannusvaikutukset'. Näitä arvoja hyväksi käyttäen referenssikaupungille perustettiin kaksi energian tuotantorakennemallia ja molemmille malleille kaksi eri skenaariota hybridilämmityksen kasvamisesta. Skenaarioissa otettiin huomioon myös päästökaupan vaikutukset. Molemmat skenaariot osoittivat päästökaupan vaikutukset mukaan luettuna, ettei sähkölämmityksen käyttäminen kaukolämmityksen ohella tuo ainoastaan yhdyskunnalle lisää tuotantokustannuksia, vaan se lisää myös päästöjä. Tulevaisuuden epävarmuutta analysoitiin herkkyysanalyysin avulla. Tutkimusta varten laadittiin tuontienergian ja kotimaisten polttoaineiden hinnoille kaksi skenaariota, joilla laskettiin vuositason tuotantokustannukset. Jokainen skenaario toi huomattavan lisän niin tuotantokustannuksiin kuin päästöihin. Eri skenaarioilla oli vaikutus kaukolämmön pysyvyyskäyrän muotoon ja näin myös voimalaitoksien käyttötunteihin. Laitoksien huipun käyttötunnit pienenevät ja tuotantokustannukset tuotettua energiayksikköä kohden kasvavat. Tapauksessa, jossa on edullista käyttää peruskuormalaitoksia mahdollisimman paljonvuoden aikana, hybridilämmityksen käyttäminen siirsi tilannetta päinvastaiseen suuntaan. Tämä suunta tarkoittaa sitä, että halpaa peruskuormatuotantoapitää korvata kalliimmalla ja enemmän päästöjä aiheuttavalla erillistuotannolla. Tämän tutkimuksen tulokset osoittavat, että mikäli sähkölämmityksen yleistyminen kaukolämmityksen rinnalla lisääntyy, aiheuttaa se yhdyskunnalle huomattavia lisäkustannuksia ja päästöjä. Yksi sähkölämmityksen käyttöä lisäävä tekijä on kuluttajien mielikuva. Kuluttajien mielikuva sähkölämmityksestä on, että se on asennuskustannuksiltaan edullinen ja helppo asentaa. Todellisuudessa sähkölämmityksen käyttäminen tuo odottamattomia lisäkustannuksia kuluttajille energiantuotantolaitoksien omien lisääntyvien tuotantokustannusten kautta. Nämä kustannukset voivat realisoitua esimerkiksi kohonneiden sähkön ja kaukolämmön energiamaksujen tai sähkön siirtomaksujen muodossa. Ainoastaan kuluttajien mielikuvien muuttamisella voidaan päästä yhdyskunnan kannalta taloudellisempaan ja ympäristöystävällisempään energiantuotantomalliin.

Preparação de eletrodos opticamente transparentes

A simple experiment for the preparation of transparent conducting glass electrodes by deposition of pure and fluorine doped SnO2 films is described. This procedure was tested in the undergraduate inorganic course at IQ-UNICAMP. The success in achieving a conducting layer was easily checked using the standard probes of a volt-ohm meter. The optical transmittance and thickness were studied by UV-vis spectrophotometry. To discuss the experimental results we place significant emphasis on molecular orbital and energy band model theories. The undergraduate students can also discuss the concepts related to the electronic properties of solids and to interesting new materials, such as transparent conducting films, which are the subject of significant current research and technological applications.

On the Rule of Mixtures for Predicting Stress-Softening and Residual Strain Effects in Biological Tissues and Biocompatible Materials

In this work, we use the rule of mixtures to develop an equivalent material model in which the total strain energy density is split into the isotropic part related to the matrix component and the anisotropic energy contribution related to the fiber effects. For the isotropic energy part, we select the amended non-Gaussian strain energy density model, while the energy fiber effects are added by considering the equivalent anisotropic volumetric fraction contribution, as well as the isotropized representation form of the eight-chain energy model that accounts for the material anisotropic effects. Furthermore, our proposed material model uses a phenomenological non-monotonous softening function that predicts stress softening effects and has an energy term, derived from the pseudo-elasticity theory, that accounts for residual strain deformations. The model’s theoretical predictions are compared with experimental data collected from human vaginal tissues, mice skin, poly(glycolide-co-caprolactone) (PGC25 3-0) and polypropylene suture materials and tracheal and brain human tissues. In all cases examined here, our equivalent material model closely follows stress-softening and residual strain effects exhibited by experimental data

Adsorção de alfa-lactalbumina do soro de leite em hidroxiapatita: efeito do pH e da temperatura e análise termodinâmica

The nutritional and functional benefits offered by whey protein α-lactalbumin justify the great interest in its manufacture in large quantities at a high purity level. Hydroxyapatite is a calcium phosphate material able to adsorb proteins and can be synthesized at low production cost. Therefore, this work evaluated the adsorption of α-lactalbumin on hydroxyapatite using solid-liquid phase equilibrium data reported as adsorption isotherms. Van't Hoff's thermodynamics analysis showed that the adsorption process is entropically driven.

Alueellinen energiatasemalli

Kiinnostus energia-asioiden käsittelemiseen alueellisesti yksittäisen rakennuksen asioiden käsittelyn sijaan on lisääntynyt voimakkaasti viime vuosina. Keskustelu alueellisesta ener-gia-asioista on kuitenkin vasta alussa ja alueellisiin energiatarkasteluihin tarvitaan työväli-neitä. Tämän työn päätavoitteena oli luoda alueellinen energiatasemalli, jolla voidaan laskea alueen rakennuskannan energiankulutusta ja -tuotantoa sekä energiantuotannon hiilidi-oksidipäästöjä helposti ja nopeasti. Työssä perehdyttiin alueellisten energiatarkasteluiden tämän hetken tasoon, työvälineisiin ja ohjeistukseen. Myös Suomen rakentamismääräyskokoelman rakennusten energiatehok-kuusmääräyksiin tutustuttiin. Alueellisen energiatasemallin perustana käytettiin uusien rakentamismääräysten E-lukulaskennan ohjeistusta. Mallin toimivuutta testattiin Helsingin Salmenkallioon rakennettavaan uudisrakennusalueeseen. Testattaessa mallilla laskettiin kyseisen alueen energiankulutus sekä erilaisia energiantuo-tantovaihtoehtoja ja todettiin, että erityisesti energiankulutuksen laskenta on hyvin helppoa, kun alueen rakennusten kerrosalat ovat tiedossa. Tuotantopuolen uusiutuvan energian laskeminen luotettavasti vaatii hiukan enemmän taustatyötä, mutta varsinainen mallin käyt-täminen on helppoa ja tulokset saadaan suoraan selkeinä kuvaajina. Työssä onnistuttiin luomaan tavoitteiden mukainen malli, jota voidaan soveltaa erilaisissa alueellisissa energiatarkasteluissa. Tarkasteltaessa energia-asioita alueellisesti, saadaan esimerkiksi alueen huipputehon tarvetta pienennettyä ja energiantuotantoa optimoitua. Malli ei kuitenkaan ole vielä valmis laskentatyökalu, vaan vaatii jatkokehitystä, mikäli sitä halutaan hyödyntää laajemmin.

Production of Mg(OH)2 from Mg-silicate rock for CO2 mineral sequestration

Sequestration of carbon dioxide in mineral rocks, also known as CO2 Capture and Mineralization (CCM), is considered to have a huge potential in stabilizing anthropogenic CO2 emissions. One of the CCM routes is the ex situ indirect gas/sold carbonation of reactive materials, such as Mg(OH)2, produced from abundantly available Mg-silicate rocks. The gas/solid carbonation method is intensively researched at Åbo Akademi University (ÅAU ), Finland because it is energetically attractive and utilizes the exothermic chemistry of Mg(OH)2 carbonation. In this thesis, a method for producing Mg(OH)2 from Mg-silicate rocks for CCM was investigated, and the process efficiency, energy and environmental impact assessed. The Mg(OH)2 process studied here was first proposed in 2008 in a Master’s Thesis by the author. At that time the process was applied to only one Mg-silicate rock (Finnish serpentinite from the Hitura nickel mine site of Finn Nickel) and the optimum process conversions, energy and environmental performance were not known. Producing Mg(OH)2 from Mg-silicate rocks involves a two-staged process of Mg extraction and Mg(OH)2 precipitation. The first stage extracts Mg and other cations by reacting pulverized serpentinite or olivine rocks with ammonium sulfate (AS) salt at 400 - 550 oC (preferably < 450 oC). In the second stage, ammonia solution reacts with the cations (extracted from the first stage after they are leached in water) to form mainly FeOOH, high purity Mg(OH)2 and aqueous (dissolved) AS. The Mg(OH)2 process described here is closed loop in nature; gaseous ammonia and water vapour are produced from the extraction stage, recovered and used as reagent for the precipitation stage. The AS reagent is thereafter recovered after the precipitation stage. The Mg extraction stage, being the conversion-determining and the most energy-intensive step of the entire CCM process chain, received a prominent attention in this study. The extraction behavior and reactivity of different rocks types (serpentinite and olivine rocks) from different locations worldwide (Australia, Finland, Lithuania, Norway and Portugal) was tested. Also, parametric evaluation was carried out to determine the optimal reaction temperature, time and chemical reagent (AS). Effects of reactor types and configuration, mixing and scale-up possibilities were also studied. The Mg(OH)2 produced can be used to convert CO2 to thermodynamically stable and environmentally benign magnesium carbonate. Therefore, the process energy and life cycle environmental performance of the ÅAU CCM technique that first produces Mg(OH)2 and the carbonates in a pressurized fluidized bed (FB) were assessed. The life cycle energy and environmental assessment approach applied in this thesis is motivated by the fact that the CCM technology should in itself offer a solution to what is both an energy and environmental problem. Results obtained in this study show that different Mg-silicate rocks react differently; olivine rocks being far less reactive than serpentinite rocks. In summary, the reactivity of Mg-silicate rocks is a function of both the chemical and physical properties of rocks. Reaction temperature and time remain important parameters to consider in process design and operation. Heat transfer properties of the reactor determine the temperature at which maximum Mg extraction is obtained. Also, an increase in reaction temperature leads to an increase in the extent of extraction, reaching a maximum yield at different temperatures depending on the reaction time. Process energy requirement for producing Mg(OH)2 from a hypothetical case of an iron-free serpentine rock is 3.62 GJ/t-CO2. This value can increase by 16 - 68% depending on the type of iron compound (FeO, Fe2O3 or Fe3O4) in the mineral. This suggests that the benefit from the potential use of FeOOH as an iron ore feedstock in iron and steelmaking should be determined by considering the energy, cost and emissions associated with the FeOOH by-product. AS recovery through crystallization is the second most energy intensive unit operation after the extraction reaction. However, the choice of mechanical vapor recompression (MVR) over the “simple evaporation” crystallization method has a potential energy savings of 15.2 GJ/t-CO2 (84 % savings). Integrating the Mg(OH)2 production method and the gas/solid carbonation process could provide up to an 25% energy offset to the CCM process energy requirements. Life cycle inventory assessment (LCIA) results show that for every ton of CO2 mineralized, the ÅAU CCM process avoids 430 - 480 kg CO2. The Mg(OH)2 process studied in this thesis has many promising features. Even at the current high energy and environmental burden, producing Mg(OH)2 from Mg-silicates can play a significant role in advancing CCM processes. However, dedicated future research and development (R&D) have potential to significantly improve the Mg(OH)2 process performance.

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.

Apprentissage des réseaux de neurones profonds et applications en traitement automatique de la langue naturelle

En apprentissage automatique, domaine qui consiste à utiliser des données pour apprendre une solution aux problèmes que nous voulons confier à la machine, le modèle des Réseaux de Neurones Artificiels (ANN) est un outil précieux. Il a été inventé voilà maintenant près de soixante ans, et pourtant, il est encore de nos jours le sujet d'une recherche active. Récemment, avec l'apprentissage profond, il a en effet permis d'améliorer l'état de l'art dans de nombreux champs d'applications comme la vision par ordinateur, le traitement de la parole et le traitement des langues naturelles. La quantité toujours grandissante de données disponibles et les améliorations du matériel informatique ont permis de faciliter l'apprentissage de modèles à haute capacité comme les ANNs profonds. Cependant, des difficultés inhérentes à l'entraînement de tels modèles, comme les minima locaux, ont encore un impact important. L'apprentissage profond vise donc à trouver des solutions, en régularisant ou en facilitant l'optimisation. Le pré-entraînnement non-supervisé, ou la technique du ``Dropout'', en sont des exemples. Les deux premiers travaux présentés dans cette thèse suivent cette ligne de recherche. Le premier étudie les problèmes de gradients diminuants/explosants dans les architectures profondes. Il montre que des choix simples, comme la fonction d'activation ou l'initialisation des poids du réseaux, ont une grande influence. Nous proposons l'initialisation normalisée pour faciliter l'apprentissage. Le second se focalise sur le choix de la fonction d'activation et présente le rectifieur, ou unité rectificatrice linéaire. Cette étude a été la première à mettre l'accent sur les fonctions d'activations linéaires par morceaux pour les réseaux de neurones profonds en apprentissage supervisé. Aujourd'hui, ce type de fonction d'activation est une composante essentielle des réseaux de neurones profonds. Les deux derniers travaux présentés se concentrent sur les applications des ANNs en traitement des langues naturelles. Le premier aborde le sujet de l'adaptation de domaine pour l'analyse de sentiment, en utilisant des Auto-Encodeurs Débruitants. Celui-ci est encore l'état de l'art de nos jours. Le second traite de l'apprentissage de données multi-relationnelles avec un modèle à base d'énergie, pouvant être utilisé pour la tâche de désambiguation de sens.