The Role of Surfactant Properties of Extractants in Hydrometallurgical LiquidLiquid Extraction Processes

The amphiphilic nature of metal extractants causes the formation of micelles and other microscopic aggregates when in contact with water and an organic diluent. These phenomena and their effects on metal extraction were studied using carboxylic acid (Versatic 10) and organophosphorus acid (Cyanex 272) based extractants. Special emphasis was laid on the study of phase behaviour in a pre neutralisation stage when the extractant is transformed to a sodium or ammonium salt form. The pre neutralised extractants were used to extract nickel and to separate cobalt and nickel. Phase diagrams corresponding to the pre neutralisation stage in a metal extraction process were determined. The maximal solubilisation of the components in the system water(NH3)/extractant/isooctane takes place when the molar ratio between the ammonia salt form and the free form of the extractant is 0.5 for the carboxylic acid and 1 for the organophosphorus acid extractant. These values correspond to the complex stoichiometry of NH4A•HA and NIi4A, respectively. When such a solution is contacted with water a microemulsion is formed. If the aqueous phase contains also metal ions (e.g. Ni²+), complexation will take place on the microscopic interface of the micellar aggregates. Experimental evidence showing that the initial stage of nickel extraction with pre neutralised Versatic 10 is a fast pseudohomogeneous reaction was obtained. About 90% of the metal were extracted in the first 15 s after the initial contact. For nickel extraction with pre neutralised Versatic 10 it was found that the highest metal loading and the lowest residual ammonia and water contents in the organic phase are achieved when the feeds are balanced so that the stoichiometry is 2NH4+(org) = Nit2+(aq). In the case of Co/Ni separation using pre neutralised Cyanex 272 the highest separation is achieved when the Co/extractant molar ratio in the feeds is 1 : 4 and at the same time the optimal degree of neutralisation of the Cyanex 272 is about 50%. The adsorption of the extractants on solid surfaces may cause accumulation of solid fine particles at the interface between the aqueous and organic phases in metal extraction processes. Copper extraction processes are known to suffer of this problem. Experiments were carried out using model silica and mica particles. It was found that high copper loading, aromacity of the diluent, modification agents and the presence of aqueous phase decrease the adsorption of the hydroxyoxime on silica surfaces.

Analysis of patterns in mining time series using Qlucore

Raw measurement data does not always immediately convey useful information, but applying mathematical statistical analysis tools into measurement data can improve the situation. Data analysis can offer benefits like acquiring meaningful insight from the dataset, basing critical decisions on the findings, and ruling out human bias through proper statistical treatment. In this thesis we analyze data from an industrial mineral processing plant with the aim of studying the possibility of forecasting the quality of the final product, given by one variable, with a model based on the other variables. For the study mathematical tools like Qlucore Omics Explorer (QOE) and Sparse Bayesian regression (SB) are used. Later on, linear regression is used to build a model based on a subset of variables that seem to have most significant weights in the SB model. The results obtained from QOE show that the variable representing the desired final product does not correlate with other variables. For SB and linear regression, the results show that both SB and linear regression models built on 1-day averaged data seriously underestimate the variance of true data, whereas the two models built on 1-month averaged data are reliable and able to explain a larger proportion of variability in the available data, making them suitable for prediction purposes. However, it is concluded that no single model can fit well the whole available dataset and therefore, it is proposed for future work to make piecewise non linear regression models if the same available dataset is used, or the plant to provide another dataset that should be collected in a more systematic fashion than the present data for further analysis.

Cat's claw oxindole alkaloid isomerization induced by common extraction methods

Cat's claw oxindole alkaloids are prone to isomerization in aqueous solution. However, studies on their behavior in extraction processes are scarce. This paper addressed the issue by considering five commonly used extraction processes. Unlike dynamic maceration (DM) and ultrasound-assisted extraction, substantial isomerization was induced by static maceration, turbo-extraction and reflux extraction. After heating under reflux in DM, the kinetic order of isomerization was established and equations were fitted successfully using a four-parameter Weibull model (R² > 0.999). Different isomerization rates and equilibrium constants were verified, revealing a possible matrix effect on alkaloid isomerization.

POPULATION STRUCTURE AND PRODUCTION OF COPAIBA OLEORESIN BETWEEN VALLEYS AND HILLSIDES OF THE MINING AREA OFTROMBETAS RIVER - PARÁ1

ABSTRACT We aimed in this work to study natural populations of copaiba (Copaifera multijuga Hayne) on the Monte Branco mountain at Porto Trombetas-PA, in order to support sustainable management and the exploitation of oleoresin from copaiba. We studied the population structure of copaiba on hillsides and valleys of the south face of Monte Branco, within Saracá Taquera National Forest, where bauxite ore was extracted in the biennium 2013-2014 by Mineração Rio do Norte (MRN). We produced a 100% forest inventory of the specie and of oleoresin extraction in order to quantify the potential production of the remaining area. The density of copaiba individuals with DBH > 30 cm was 0.33 individuals per hectare in the hillside and 0.25 individuals per hectare in the valley. Both environments presented a density of 0.28 individuals per hectare. The average copaiba oleoresin yield was 0.661±0.334 liters in the hillside and 0.765±0.280 liters in the valley. The average value of both environments together (hillside and valley) was 0.714±0.218 liters. From all individuals with DBH over 30 cm, 38 (58%) produced some amount of oleoresin, averaging 1.113±0.562 liters in the hillside, 1.329±0.448 liters in the valley and 1.190±0.355 liters in both environments together. The results show the need for planning the use of the surroundings of the study area in order to reach the required volume of copaiba to make feasible the sustainable management of oleoresin extraction in the region.

Efficiency of distinct data mining algorithms for classifying stress level in piglets from their vocalization

Among the challenges of pig farming in today's competitive market, there is factor of the product traceability that ensures, among many points, animal welfare. Vocalization is a valuable tool to identify situations of stress in pigs, and it can be used in welfare records for traceability. The objective of this work was to identify stress in piglets using vocalization, calling this stress on three levels: no stress, moderate stress, and acute stress. An experiment was conducted on a commercial farm in the municipality of Holambra, São Paulo State , where vocalizations of twenty piglets were recorded during the castration procedure, and separated into two groups: without anesthesia and local anesthesia with lidocaine base. For the recording of acoustic signals, a unidirectional microphone was connected to a digital recorder, in which signals were digitized at a frequency of 44,100 Hz. For evaluation of sound signals, Praat® software was used, and different data mining algorithms were applied using Weka® software. The selection of attributes improved model accuracy, and the best attribute selection was used by applying Wrapper method, while the best classification algorithms were the k-NN and Naive Bayes. According to the results, it was possible to classify the level of stress in pigs through their vocalization.

Extraction of polymeric galactoglucomannans from spruce wood by pressurized hot water

The major type of non-cellulosic polysaccharides (hemicelluloses) in softwoods, the partly acetylated galactoglucomannans (GGMs), which comprise about 15% of spruce wood, have attracted growing interest because of their potential to become high-value products with applications in many areas. The main objective of this work was to explore the possibilities to extract galactoglucomannans in native, polymeric form in high yield from spruce wood with pressurised hot-water, and to obtain a deeper understanding of the process chemistry involved. Spruce (Picea abies) chips and ground wood particles were extracted using an accelerated solvent extractor (ASE) in the temperature range 160 – 180°C. Detailed chemical analyses were done on both the water extracts and the wood residues. As much as 80 – 90% of the GGMs in spruce wood, i.e. about 13% based on the original wood, could be extracted from ground spruce wood with pure water at 170 – 180°C with an extraction time of 60 min. GGMs comprised about 75% of the extracted carbohydrates and about 60% of the total dissolved solids. Other substances in the water extracts were xylans, arabinogalactans, pectins, lignin and acetic acid. The yields from chips were only about 60% of that from ground wood. Both the GGMs and other non-cellulosic polysaccharides were extensively hydrolysed at severe extraction conditions when pH dropped to the level of 3.5. Addition of sodium bicarbonate increased the yields of polymeric GGMs at low additions, 2.5 – 5 mM, where the end pH remained around 3.9. However, at higher addition levels the yields decreased, mainly because the acetyl groups in GGMs were split off, leading to a low solubility of GGMs. Extraction with buffered water in the pH range 3.8 – 4.4 gave similar yields as with plain water, but gave a higher yield of polymeric GGMs. Moreover, at these pH levels the hydrolysis of acetyl groups in GGMs was significantly inhibited. It was concluded that hot-water extraction of polymeric GGMs in good yields (up to 8% of wood) demands appropriate control of pH, in a narrow range about 4. These results were supported by a study of hydrolysis of GGM at constant pH in the range of 3.8 – 4.2 where a kinetic model for degradation of GGM was developed. The influence of wood particle size on hot-water extraction was studied with particles in the range of 0.1 – 2 mm. The smallest particles (< 0.1 mm) gave 20 – 40% higher total yield than the coarsest particles (1.25 – 2 mm). The difference was greatest at short extraction times. The results indicated that extraction of GGMs and other polysaccharides is limited mainly by the mass transfer in the fibre wall, and for coarse wood particles also in the wood matrix. Spruce sapwood, heartwood and thermomechnical pulp were also compared, but only small differences in yields and composition of extracts were found. Two methods for isolation and purification of polymeric GGMs, i.e. membrane filtration and precipitation in ethanol-water, were compared. Filtration through a series of membranes with different pore sizes separated GGMs of different molar masses, from polymers to oligomers. Polysaccharides with molar mass higher than 4 kDa were precipitated in ethanol-water. GGMs comprised about 80% of the precipitated polysaccharides. Other polysaccharides were mainly arabinoglucuronoxylans and pectins. The ethanol-precipitated GGMs were by 13C NMR spectroscopy verified to be very similar to GGMs extracted from spruce wood in low yield at a much lower temperature, 90°C. The obtained large body of experimental data could be utilised for further kinetic and economic calculations to optimise technical hot-water extractionof softwoods.

Biomedical Event Extraction with Machine Learning

Biomedical natural language processing (BioNLP) is a subfield of natural language processing, an area of computational linguistics concerned with developing programs that work with natural language: written texts and speech. Biomedical relation extraction concerns the detection of semantic relations such as protein-protein interactions (PPI) from scientific texts. The aim is to enhance information retrieval by detecting relations between concepts, not just individual concepts as with a keyword search. In recent years, events have been proposed as a more detailed alternative for simple pairwise PPI relations. Events provide a systematic, structural representation for annotating the content of natural language texts. Events are characterized by annotated trigger words, directed and typed arguments and the ability to nest other events. For example, the sentence “Protein A causes protein B to bind protein C” can be annotated with the nested event structure CAUSE(A, BIND(B, C)). Converted to such formal representations, the information of natural language texts can be used by computational applications. Biomedical event annotations were introduced by the BioInfer and GENIA corpora, and event extraction was popularized by the BioNLP'09 Shared Task on Event Extraction. In this thesis we present a method for automated event extraction, implemented as the Turku Event Extraction System (TEES). A unified graph format is defined for representing event annotations and the problem of extracting complex event structures is decomposed into a number of independent classification tasks. These classification tasks are solved using SVM and RLS classifiers, utilizing rich feature representations built from full dependency parsing. Building on earlier work on pairwise relation extraction and using a generalized graph representation, the resulting TEES system is capable of detecting binary relations as well as complex event structures. We show that this event extraction system has good performance, reaching the first place in the BioNLP'09 Shared Task on Event Extraction. Subsequently, TEES has achieved several first ranks in the BioNLP'11 and BioNLP'13 Shared Tasks, as well as shown competitive performance in the binary relation Drug-Drug Interaction Extraction 2011 and 2013 shared tasks. The Turku Event Extraction System is published as a freely available open-source project, documenting the research in detail as well as making the method available for practical applications. In particular, in this thesis we describe the application of the event extraction method to PubMed-scale text mining, showing how the developed approach not only shows good performance, but is generalizable and applicable to large-scale real-world text mining projects. Finally, we discuss related literature, summarize the contributions of the work and present some thoughts on future directions for biomedical event extraction. This thesis includes and builds on six original research publications. The first of these introduces the analysis of dependency parses that leads to development of TEES. The entries in the three BioNLP Shared Tasks, as well as in the DDIExtraction 2011 task are covered in four publications, and the sixth one demonstrates the application of the system to PubMed-scale text mining.

Optimization of the extraction of carrageenan from Kappaphycus alvarezii using response surface methodology

This study aims to optimize an alternative method of extraction of carrageenan without previous alkaline treatment and ethanol precipitation using Response Surface Methodology (RSM). In order to introduce an innovation in the isolation step, atomization drying was used reducing the time for obtaining dry carrageenan powder. The effects of extraction time and temperature on yield, gel strength, and viscosity were evaluated. Furthermore, the extracted material was submitted to structural analysis, by infrared spectroscopy and nuclear magnetic resonance spectroscopy (¹H-NMR), and chemical composition analysis. Results showed that the generated regression models adequately explained the data variation. Carrageenan yield and gel viscosity were influenced only by the extraction temperature. However, gel strength was influenced by both, extraction time and extraction temperature. Optimal extraction conditions were 74 ºC and 4 hours. In these conditions, the carrageenan extract properties determined by the polynomial model were 31.17%, 158.27 g.cm-2, and 29.5 cP for yield, gel strength, and viscosity, respectively, while under the experimental conditions they were 35.8 ± 4.68%, 112.50 ± 4.96 g.cm-2, and 16.01 ± 1.03 cP, respectively. The chemical composition, nuclear magnetic resonance spectroscopy, and infrared spectroscopy analyses showed that the crude carrageenan extracted is composed mainly of κ-carrageenan.

Heat Demand Forecasting Models’ Development: Use of Data Mining Tools in SQL Server Analysis Services

This thesis introduces heat demand forecasting models which are generated by using data mining algorithms. The forecast spans one full day and this forecast can be used in regulating heat consumption of buildings. For training the data mining models, two years of heat consumption data from a case building and weather measurement data from Finnish Meteorological Institute are used. The thesis utilizes Microsoft SQL Server Analysis Services data mining tools in generating the data mining models and CRISP-DM process framework to implement the research. Results show that the built models can predict heat demand at best with mean average percentage errors of 3.8% for 24-h profile and 5.9% for full day. A deployment model for integrating the generated data mining models into an existing building energy management system is also discussed.

Simultaneous extraction of order parameters and orientational distribution fuctions from 31[superscript]P NMR spectra of magnetically partially oriented phospholipid bilayers

Order parameter profiles extracted from the NMR spectra of model membranes are a valuable source of information about their structure and molecular motions. To al1alyze powder spectra the de-Pake-ing (numerical deconvolution) ~echnique can be used, but it assumes a random (spherical) dist.ribution of orientations in the sample. Multilamellar vesicles are known to deform and orient in the strong magnetic fields of NMR magnets, producing non-spherical orientation distributions. A recently developed technique for simultaneously extracting the anisotropies of the system as well as the orientation distributions is applied to the analysis of partially magnetically oriented 31p NMR spectra of phospholipids. A mixture of synthetic lipids, POPE and POPG, is analyzed to measure distortion of multilamellar vesicles in a magnetic field. In the analysis three models describing the shape of the distorted vesicles are examined. Ellipsoids of rotation with a semiaxis ratio of about 1.14 are found to provide a good approximation of the shape of the distorted vesicles. This is in reasonable agreement with published experimental work. All three models yield clearly non-spherical orientational distributions, as well as a precise measure of the anisotropy of the chemical shift. Noise in the experimental data prevented the analysis from concluding which of the three models is the best approximation. A discretization scheme for finding stability in the algorithm is outlined

Developing a numerical inverse-theory-based extraction of orientation-dependent relaxation rates from partially- relaxed spectra

Second-rank tensor interactions, such as quadrupolar interactions between the spin- 1 deuterium nuclei and the electric field gradients created by chemical bonds, are affected by rapid random molecular motions that modulate the orientation of the molecule with respect to the external magnetic field. In biological and model membrane systems, where a distribution of dynamically averaged anisotropies (quadrupolar splittings, chemical shift anisotropies, etc.) is present and where, in addition, various parts of the sample may undergo a partial magnetic alignment, the numerical analysis of the resulting Nuclear Magnetic Resonance (NMR) spectra is a mathematically ill-posed problem. However, numerical methods (de-Pakeing, Tikhonov regularization) exist that allow for a simultaneous determination of both the anisotropy and orientational distributions. An additional complication arises when relaxation is taken into account. This work presents a method of obtaining the orientation dependence of the relaxation rates that can be used for the analysis of the molecular motions on a broad range of time scales. An arbitrary set of exponential decay rates is described by a three-term truncated Legendre polynomial expansion in the orientation dependence, as appropriate for a second-rank tensor interaction, and a linear approximation to the individual decay rates is made. Thus a severe numerical instability caused by the presence of noise in the experimental data is avoided. At the same time, enough flexibility in the inversion algorithm is retained to achieve a meaningful mapping from raw experimental data to a set of intermediate, model-free

Investigating the potential of thecamoebians (testate amoebae) as bio-indicators of impact of oil sands mining operations on freshwater environments in Northeastern Alberta, Canada

Thecamoebian (testate amoeba) species diversity and assemblages in reclamation wetlands and lakes in northeastern Alberta respond to chemical and physical parameters associated with oil sands extraction. Ecosystems more impacted by OSPM (oil sands process-affected material) contain sparse, low-diversity populations dominated by centropyxid taxa and Arcella vulgaris. More abundant and diverse thecamoebian populations rich in difflugiid species characterize environments with lower OSPM concentrations. These shelled protists respond quickly to environmental change, allowing year-to-year variations in OSPM impact to be recorded. Their fossil record thus provides corporations with interests in the Athabasca Oil Sands with a potential means of measuring the progression of highlyimpacted aquatic environments to more natural wetlands. Development of this metric required investigation of controls on their fossil assemblage (e.g. seasonal variability, fossilization potential) and their biogeographic distribution, not only in the constructed lakes and wetlands on the oil sands leases, but also in natural environments across Alberta.

Urban Detection From Hyperspectral Images Using Dimension-Reduction Model and Fusion of Multiple Segmentations Based on Stuctural and Textural Features

Ce mémoire de maîtrise présente une nouvelle approche non supervisée pour détecter et segmenter les régions urbaines dans les images hyperspectrales. La méthode proposée n ́ecessite trois étapes. Tout d’abord, afin de réduire le coût calculatoire de notre algorithme, une image couleur du contenu spectral est estimée. A cette fin, une étape de réduction de dimensionalité non-linéaire, basée sur deux critères complémentaires mais contradictoires de bonne visualisation; à savoir la précision et le contraste, est réalisée pour l’affichage couleur de chaque image hyperspectrale. Ensuite, pour discriminer les régions urbaines des régions non urbaines, la seconde étape consiste à extraire quelques caractéristiques discriminantes (et complémentaires) sur cette image hyperspectrale couleur. A cette fin, nous avons extrait une série de paramètres discriminants pour décrire les caractéristiques d’une zone urbaine, principalement composée d’objets manufacturés de formes simples g ́eométriques et régulières. Nous avons utilisé des caractéristiques texturales basées sur les niveaux de gris, la magnitude du gradient ou des paramètres issus de la matrice de co-occurrence combinés avec des caractéristiques structurelles basées sur l’orientation locale du gradient de l’image et la détection locale de segments de droites. Afin de réduire encore la complexité de calcul de notre approche et éviter le problème de la ”malédiction de la dimensionnalité” quand on décide de regrouper des données de dimensions élevées, nous avons décidé de classifier individuellement, dans la dernière étape, chaque caractéristique texturale ou structurelle avec une simple procédure de K-moyennes et ensuite de combiner ces segmentations grossières, obtenues à faible coût, avec un modèle efficace de fusion de cartes de segmentations. Les expérimentations données dans ce rapport montrent que cette stratégie est efficace visuellement et se compare favorablement aux autres méthodes de détection et segmentation de zones urbaines à partir d’images hyperspectrales.

Extraction automatique et visualisation des thèmes abordés dans des résumés de mémoires et de thèses en anthropologie au Québec, de 1985 à 2009

S’insérant dans les domaines de la Lecture et de l’Analyse de Textes Assistées par Ordinateur (LATAO), de la Gestion Électronique des Documents (GÉD), de la visualisation de l’information et, en partie, de l’anthropologie, cette recherche exploratoire propose l’expérimentation d’une méthodologie descriptive en fouille de textes afin de cartographier thématiquement un corpus de textes anthropologiques. Plus précisément, nous souhaitons éprouver la méthode de classification hiérarchique ascendante (CHA) pour extraire et analyser les thèmes issus de résumés de mémoires et de thèses octroyés de 1985 à 2009 (1240 résumés), par les départements d’anthropologie de l’Université de Montréal et de l’Université Laval, ainsi que le département d’histoire de l’Université Laval (pour les résumés archéologiques et ethnologiques). En première partie de mémoire, nous présentons notre cadre théorique, c'est-à-dire que nous expliquons ce qu’est la fouille de textes, ses origines, ses applications, les étapes méthodologiques puis, nous complétons avec une revue des principales publications. La deuxième partie est consacrée au cadre méthodologique et ainsi, nous abordons les différentes étapes par lesquelles ce projet fut conduit; la collecte des données, le filtrage linguistique, la classification automatique, pour en nommer que quelques-unes. Finalement, en dernière partie, nous présentons les résultats de notre recherche, en nous attardant plus particulièrement sur deux expérimentations. Nous abordons également la navigation thématique et les approches conceptuelles en thématisation, par exemple, en anthropologie, la dichotomie culture ̸ biologie. Nous terminons avec les limites de ce projet et les pistes d’intérêts pour de futures recherches.

Improving sampling, optimization and feature extraction in Boltzmann machines

L’apprentissage supervisé de réseaux hiérarchiques à grande échelle connaît présentement un succès fulgurant. Malgré cette effervescence, l’apprentissage non-supervisé représente toujours, selon plusieurs chercheurs, un élément clé de l’Intelligence Artificielle, où les agents doivent apprendre à partir d’un nombre potentiellement limité de données. Cette thèse s’inscrit dans cette pensée et aborde divers sujets de recherche liés au problème d’estimation de densité par l’entremise des machines de Boltzmann (BM), modèles graphiques probabilistes au coeur de l’apprentissage profond. Nos contributions touchent les domaines de l’échantillonnage, l’estimation de fonctions de partition, l’optimisation ainsi que l’apprentissage de représentations invariantes. Cette thèse débute par l’exposition d’un nouvel algorithme d'échantillonnage adaptatif, qui ajuste (de fa ̧con automatique) la température des chaînes de Markov sous simulation, afin de maintenir une vitesse de convergence élevée tout au long de l’apprentissage. Lorsqu’utilisé dans le contexte de l’apprentissage par maximum de vraisemblance stochastique (SML), notre algorithme engendre une robustesse accrue face à la sélection du taux d’apprentissage, ainsi qu’une meilleure vitesse de convergence. Nos résultats sont présent ́es dans le domaine des BMs, mais la méthode est générale et applicable à l’apprentissage de tout modèle probabiliste exploitant l’échantillonnage par chaînes de Markov. Tandis que le gradient du maximum de vraisemblance peut-être approximé par échantillonnage, l’évaluation de la log-vraisemblance nécessite un estimé de la fonction de partition. Contrairement aux approches traditionnelles qui considèrent un modèle donné comme une boîte noire, nous proposons plutôt d’exploiter la dynamique de l’apprentissage en estimant les changements successifs de log-partition encourus à chaque mise à jour des paramètres. Le problème d’estimation est reformulé comme un problème d’inférence similaire au filtre de Kalman, mais sur un graphe bi-dimensionnel, où les dimensions correspondent aux axes du temps et au paramètre de température. Sur le thème de l’optimisation, nous présentons également un algorithme permettant d’appliquer, de manière efficace, le gradient naturel à des machines de Boltzmann comportant des milliers d’unités. Jusqu’à présent, son adoption était limitée par son haut coût computationel ainsi que sa demande en mémoire. Notre algorithme, Metric-Free Natural Gradient (MFNG), permet d’éviter le calcul explicite de la matrice d’information de Fisher (et son inverse) en exploitant un solveur linéaire combiné à un produit matrice-vecteur efficace. L’algorithme est prometteur: en terme du nombre d’évaluations de fonctions, MFNG converge plus rapidement que SML. Son implémentation demeure malheureusement inefficace en temps de calcul. Ces travaux explorent également les mécanismes sous-jacents à l’apprentissage de représentations invariantes. À cette fin, nous utilisons la famille de machines de Boltzmann restreintes “spike & slab” (ssRBM), que nous modifions afin de pouvoir modéliser des distributions binaires et parcimonieuses. Les variables latentes binaires de la ssRBM peuvent être rendues invariantes à un sous-espace vectoriel, en associant à chacune d’elles, un vecteur de variables latentes continues (dénommées “slabs”). Ceci se traduit par une invariance accrue au niveau de la représentation et un meilleur taux de classification lorsque peu de données étiquetées sont disponibles. Nous terminons cette thèse sur un sujet ambitieux: l’apprentissage de représentations pouvant séparer les facteurs de variations présents dans le signal d’entrée. Nous proposons une solution à base de ssRBM bilinéaire (avec deux groupes de facteurs latents) et formulons le problème comme l’un de “pooling” dans des sous-espaces vectoriels complémentaires.