Using a New Criterion to Identify Sites for Mean Soil Water Storage Evaluation

Establishing a few sites in which measurements of soil water storage (SWS) are time stable significantly reduces the efforts involved in determining average values of SWS. This study aimed to apply a new criterion the mean absolute bias error (MABE)-to identify temporally stable sites for mean SWS evaluation. The performance of MABE was compared with that of the commonly used criterion, the standard deviation of relative difference (SDRD). From October 2004 to October 2008, SWS of four soil layers (0-1.0, 1.0-2.0,2.0-3.0, and 3.0-4.0 m) was measured, using a neutron probe, at 28 sites on a hillslope of the Loess Plateau, China. A total of 37 SWS data sets taken over time were divided into two subsets, the first consisting of 22 dates collected during the calibration period from October 2004 to September 2006, and the second with 15 dates collected during the validation period from October 2006 to October 2008. The results showed that if a critical value of 5% for MABE was defined, more than half the sites were temporally stable for both periods, and the number of temporally stable sires generally increased with soil depth. Compared with SDRD, MABE was more suitable for the identification of time-stable sites for mean SS prediction. Since the absolute prediction error of drier sites is more sensitive to changes in relative difference in terms of mean SWS prediction, the sites of wet sectors should be preferable for mean SWS prediction for the same changes in relative difference.

Selection bias in gene extraction on the basis of microarray gene-expression data

In the context of cancer diagnosis and treatment, we consider the problem of constructing an accurate prediction rule on the basis of a relatively small number of tumor tissue samples of known type containing the expression data on very many (possibly thousands) genes. Recently, results have been presented in the literature suggesting that it is possible to construct a prediction rule from only a few genes such that it has a negligible prediction error rate. However, in these results the test error or the leave-one-out cross-validated error is calculated without allowance for the selection bias. There is no allowance because the rule is either tested on tissue samples that were used in the first instance to select the genes being used in the rule or because the cross-validation of the rule is not external to the selection process; that is, gene selection is not performed in training the rule at each stage of the cross-validation process. We describe how in practice the selection bias can be assessed and corrected for by either performing a cross-validation or applying the bootstrap external to the selection process. We recommend using 10-fold rather than leave-one-out cross-validation, and concerning the bootstrap, we suggest using the so-called. 632+ bootstrap error estimate designed to handle overfitted prediction rules. Using two published data sets, we demonstrate that when correction is made for the selection bias, the cross-validated error is no longer zero for a subset of only a few genes.

Reference distribution of the bioelectrical impedance vector in healthy term newborns

Bioelectrical impedance vector analysis (BIVA) is a new method that is used for the routine monitoring of the variation in body fluids and nutritional status with assumptions regarding body composition values. The aim of the present study was to determine bivariate tolerance intervals of the whole-body impedance vector and to describe phase angle (PA) values for healthy term newborns aged 7-28 d. This descriptive cross-sectional study was conducted on healthy term neonates born at a low-risk public maternity. General and anthropometric neonatal data and bioelectrical impedance data (800 mu A-50 kHz) were obtained. Bivariate vector analysis was conducted with the resistance-reactance (RXc) graph method. The BIVA software was used to construct the graphs. The study was conducted on 109 neonates (52.3% females) who were born at term, adequate for gestational age, exclusively breast-fed and aged 13 (SD 3.6) d. We constructed one standard, reference, RXc-score graph and RXc-tolerance ellipses (50, 75 and 95 %) that can be used with any analyser. Mean PA was 3.14 (SD 0.43)degrees (3.12 (SD 0.39)degrees for males and 3.17 (SD 0.48)degrees for females). Considering the overlapping of ellipses of males and females with the general distribution, a graph for newborns aged 7-28 d with the same reference tolerance ellipse was defined for boys and girls. The results differ from those reported in the literature probably, in part, due to the ethnic differences in body composition. BIVA and PA permit an assessment without the need to know body weight and the prediction error of conventional impedance formulas.

Hierarchical Reinforcement Learning in Behavior and the Brain

Dissertation presented to obtain the Ph.D degree in Biology, Neuroscience

Evaluation of geostatistical techniques to estimate the spatial distribution of aboveground biomass in the Amazon rainforest using high-resolution remote sensing data

ABSTRACT The spatial distribution of forest biomass in the Amazon is heterogeneous with a temporal and spatial variation, especially in relation to the different vegetation types of this biome. Biomass estimated in this region varies significantly depending on the applied approach and the data set used for modeling it. In this context, this study aimed to evaluate three different geostatistical techniques to estimate the spatial distribution of aboveground biomass (AGB). The selected techniques were: 1) ordinary least-squares regression (OLS), 2) geographically weighted regression (GWR) and, 3) geographically weighted regression - kriging (GWR-K). These techniques were applied to the same field dataset, using the same environmental variables derived from cartographic information and high-resolution remote sensing data (RapidEye). This study was developed in the Amazon rainforest from Sucumbíos - Ecuador. The results of this study showed that the GWR-K, a hybrid technique, provided statistically satisfactory estimates with the lowest prediction error compared to the other two techniques. Furthermore, we observed that 75% of the AGB was explained by the combination of remote sensing data and environmental variables, where the forest types are the most important variable for estimating AGB. It should be noted that while the use of high-resolution images significantly improves the estimation of the spatial distribution of AGB, the processing of this information requires high computational demand.

The feasibility of forecasting influenza epidemics in Cuba

A large influenza epidemic took place in Havana during the winter of 1988. The epidemiologic surveillance unit of the Pedro Kouri Institute of Tropical Medicine detected the begining of the epidemic wave. The Rvachev-Baroyan mathematical model of the geographic spread of an epidemic was used to forecast this epidemic under routine conditions of the public health system. The expected number of individuals who would attend outpatient services, because of influenza-like illness, was calculated and communicated to the health authorities within enough time to permit the introduction of available control measures. The approximate date of the epidemic peak, the daily expected number of individuals attending medical services, and the approximate time of the end of the epidemic wave were estimated. The prediction error was 12%. The model was sufficienty accurate to warrant its use as a pratical forecasting tool in the Cuban public health system.

Evaluation of Different Methods to Calculate Heavy-Truck VMT, Final Report, December 2004

Reliable estimates of heavy-truck volumes are important in a number of transportation applications. Estimates of truck volumes are necessary for pavement design and pavement management. Truck volumes are important in traffic safety. The number of trucks on the road also influences roadway capacity and traffic operations. Additionally, heavy vehicles pollute at higher rates than passenger vehicles. Consequently, reliable estimates of heavy-truck vehicle miles traveled (VMT) are important in creating accurate inventories of on-road emissions. This research evaluated three different methods to calculate heavy-truck annual average daily traffic (AADT) which can subsequently be used to estimate vehicle miles traveled (VMT). Traffic data from continuous count stations provided by the Iowa DOT were used to estimate AADT for two different truck groups (single-unit and multi-unit) using the three methods. The first method developed monthly and daily expansion factors for each truck group. The second and third methods created general expansion factors for all vehicles. Accuracy of the three methods was compared using n-fold cross-validation. In n-fold cross-validation, data are split into n partitions, and data from the nth partition are used to validate the remaining data. A comparison of the accuracy of the three methods was made using the estimates of prediction error obtained from cross-validation. The prediction error was determined by averaging the squared error between the estimated AADT and the actual AADT. Overall, the prediction error was the lowest for the method that developed expansion factors separately for the different truck groups for both single- and multi-unit trucks. This indicates that use of expansion factors specific to heavy trucks results in better estimates of AADT, and, subsequently, VMT, than using aggregate expansion factors and applying a percentage of trucks. Monthly, daily, and weekly traffic patterns were also evaluated. Significant variation exists in the temporal and seasonal patterns of heavy trucks as compared to passenger vehicles. This suggests that the use of aggregate expansion factors fails to adequately describe truck travel patterns.

Estudi de l'estat de Salut autopercebut: Modelització de l'índex d'utilitat EQ-5D mitjançant un model tobit

Objective: Health status measures usually have an asymmetric distribution and present a highpercentage of respondents with the best possible score (ceiling effect), specially when they areassessed in the overall population. Different methods to model this type of variables have beenproposed that take into account the ceiling effect: the tobit models, the Censored Least AbsoluteDeviations (CLAD) models or the two-part models, among others. The objective of this workwas to describe the tobit model, and compare it with the Ordinary Least Squares (OLS) model,that ignores the ceiling effect.Methods: Two different data sets have been used in order to compare both models: a) real datacomming from the European Study of Mental Disorders (ESEMeD), in order to model theEQ5D index, one of the measures of utilities most commonly used for the evaluation of healthstatus; and b) data obtained from simulation. Cross-validation was used to compare thepredicted values of the tobit model and the OLS models. The following estimators werecompared: the percentage of absolute error (R1), the percentage of squared error (R2), the MeanSquared Error (MSE) and the Mean Absolute Prediction Error (MAPE). Different datasets werecreated for different values of the error variance and different percentages of individuals withceiling effect. The estimations of the coefficients, the percentage of explained variance and theplots of residuals versus predicted values obtained under each model were compared.Results: With regard to the results of the ESEMeD study, the predicted values obtained with theOLS model and those obtained with the tobit models were very similar. The regressioncoefficients of the linear model were consistently smaller than those from the tobit model. In thesimulation study, we observed that when the error variance was small (s=1), the tobit modelpresented unbiased estimations of the coefficients and accurate predicted values, specially whenthe percentage of individuals wiht the highest possible score was small. However, when theerrror variance was greater (s=10 or s=20), the percentage of explained variance for the tobitmodel and the predicted values were more similar to those obtained with an OLS model.Conclusions: The proportion of variability accounted for the models and the percentage ofindividuals with the highest possible score have an important effect in the performance of thetobit model in comparison with the linear model.

Stochastic inversion of tracer test and electrical geophysical data to estimate hydraulic conductivities.

Quantifying the spatial configuration of hydraulic conductivity (K) in heterogeneous geological environments is essential for accurate predictions of contaminant transport, but is difficult because of the inherent limitations in resolution and coverage associated with traditional hydrological measurements. To address this issue, we consider crosshole and surface-based electrical resistivity geophysical measurements, collected in time during a saline tracer experiment. We use a Bayesian Markov-chain-Monte-Carlo (McMC) methodology to jointly invert the dynamic resistivity data, together with borehole tracer concentration data, to generate multiple posterior realizations of K that are consistent with all available information. We do this within a coupled inversion framework, whereby the geophysical and hydrological forward models are linked through an uncertain relationship between electrical resistivity and concentration. To minimize computational expense, a facies-based subsurface parameterization is developed. The Bayesian-McMC methodology allows us to explore the potential benefits of including the geophysical data into the inverse problem by examining their effect on our ability to identify fast flowpaths in the subsurface, and their impact on hydrological prediction uncertainty. Using a complex, geostatistically generated, two-dimensional numerical example representative of a fluvial environment, we demonstrate that flow model calibration is improved and prediction error is decreased when the electrical resistivity data are included. The worth of the geophysical data is found to be greatest for long spatial correlation lengths of subsurface heterogeneity with respect to wellbore separation, where flow and transport are largely controlled by highly connected flowpaths.

Lossless Image Data Embedding in Plain Areas

This letter presents a lossless data hiding scheme for digital images which uses an edge detector to locate plain areas for embedding. The proposed method takes advantage of the well-known gradient adjacent prediction utilized in image coding. In the suggested scheme, prediction errors and edge values are first computed and then, excluding the edge pixels, prediction error values are slightly modified through shifting the prediction errors to embed data. The aim of proposed scheme is to decrease the amount of modified pixels to improve transparency by keeping edge pixel values of the image. The experimental results have demonstrated that the proposed method is capable of hiding more secret data than the known techniques at the same PSNR, thus proving that using edge detector to locate plain areas for lossless data embedding can enhance the performance in terms of data embedding rate versus the PSNR of marked images with respect to original image.

Cognitive anatomy of the temporal lobe: Effect of personality in population with mild cognitive impairment & Functional specialization for memory systems in healthy individuals.

Résumé: L'impact de la maladie d'Alzheimer (MA) est dévastateur pour la vie quotidienne de la personne affectée, avec perte progressive de la mémoire et d'autres facultés cognitives jusqu'à la démence. Il n'existe toujours pas de traitement contre cette maladie et il y a aussi une grande incertitude sur le diagnostic des premiers stades de la MA. La signature anatomique de la MA, en particulier l'atrophie du lobe temporal moyen (LTM) mesurée avec la neuroimagerie, peut être utilisée comme un biomarqueur précoce, in vivo, des premiers stades de la MA. Toutefois, malgré le rôle évident du LMT dans les processus de la mémoire, nous savons que les modèles anatomiques prédictifs de la MA basés seulement sur des mesures d'atrophie du LTM n'expliquent pas tous les cas cliniques. Au cours de ma thèse, j'ai conduit trois projets pour comprendre l'anatomie et le fonctionnement du LMT dans (1) les processus de la maladie et dans (2) les processus de mémoire ainsi que (3) ceux de l'apprentissage. Je me suis intéressée à une population avec déficit cognitif léger (« Mild Cognitive Impairment », MCI), à risque pour la MA. Le but du premier projet était de tester l'hypothèse que des facteurs, autres que ceux cognitifs, tels que les traits de personnalité peuvent expliquer les différences interindividuelles dans le LTM. De plus, la diversité phénotypique des manifestations précliniques de la MA provient aussi d'une connaissance limitée des processus de mémoire et d'apprentissage dans le cerveau sain. L'objectif du deuxième projet porte sur l'investigation des sous-régions du LTM, et plus particulièrement de leur contribution dans différentes composantes de la mémoire de reconnaissance chez le sujet sain. Pour étudier cela, j'ai utilisé une nouvelle méthode multivariée ainsi que l'IRM à haute résolution pour tester la contribution de ces sous-régions dans les processus de familiarité (« ou Know ») et de remémoration (ou « Recollection »). Finalement, l'objectif du troisième projet était de tester la contribution du LTM en tant que système de mémoire dans l'apprentissage et l'interaction dynamique entre différents systèmes de mémoire durant l'apprentissage. Les résultats du premier projet montrent que, en plus du déficit cognitif observé dans une population avec MCI, les traits de personnalité peuvent expliquer les différences interindividuelles du LTM ; notamment avec une plus grande contribution du neuroticisme liée à une vulnérabilité au stress et à la dépression. Mon étude a permis d'identifier un pattern d'anormalité anatomique dans le LTM associé à la personnalité avec des mesures de volume et de diffusion moyenne du tissu. Ce pattern est caractérisé par une asymétrie droite-gauche du LTM et un gradient antéro-postérieur dans le LTM. J'ai interprété ce résultat par des propriétés tissulaires et neurochimiques différemment sensibles au stress. Les résultats de mon deuxième projet ont contribué au débat actuel sur la contribution des sous-régions du LTM dans les processus de familiarité et de remémoration. Utilisant une nouvelle méthode multivariée, les résultats supportent premièrement une dissociation des sous-régions associées aux différentes composantes de la mémoire. L'hippocampe est le plus associé à la mémoire de type remémoration et le cortex parahippocampique, à la mémoire de type familiarité. Deuxièmement, l'activation correspondant à la trace mnésique pour chaque type de mémoire est caractérisée par une distribution spatiale distincte. La représentation neuronale spécifique, « sparse-distributed», associée à la mémoire de remémoration dans l'hippocampe serait la meilleure manière d'encoder rapidement des souvenirs détaillés sans interférer les souvenirs précédemment stockés. Dans mon troisième projet, j'ai mis en place une tâche d'apprentissage en IRM fonctionnelle pour étudier les processus d'apprentissage d'associations probabilistes basé sur le feedback/récompense. Cette étude m'a permis de mettre en évidence le rôle du LTM dans l'apprentissage et l'interaction entre différents systèmes de mémoire comme la mémoire procédurale, perceptuelle ou d'amorçage et la mémoire de travail. Nous avons trouvé des activations dans le LTM correspondant à un processus de mémoire épisodique; les ganglions de la base (GB), à la mémoire procédurale et la récompense; le cortex occipito-temporal (OT), à la mémoire de représentation perceptive ou l'amorçage et le cortex préfrontal, à la mémoire de travail. Nous avons également observé que ces régions peuvent interagir; le type de relation entre le LTM et les GB a été interprété comme une compétition, ce qui a déjà été reporté dans des études récentes. De plus, avec un modèle dynamique causal, j'ai démontré l'existence d'une connectivité effective entre des régions. Elle se caractérise par une influence causale de type « top-down » venant de régions corticales associées avec des processus de plus haut niveau venant du cortex préfrontal sur des régions corticales plus primaires comme le OT cortex. Cette influence diminue au cours du de l'apprentissage; cela pourrait correspondre à un mécanisme de diminution de l'erreur de prédiction. Mon interprétation est que cela est à l'origine de la connaissance sémantique. J'ai également montré que les choix du sujet et l'activation cérébrale associée sont influencés par les traits de personnalité et des états affectifs négatifs. Les résultats de cette thèse m'ont amenée à proposer (1) un modèle expliquant les mécanismes possibles liés à l'influence de la personnalité sur le LTM dans une population avec MCI, (2) une dissociation des sous-régions du LTM dans différents types de mémoire et une représentation neuronale spécifique à ces régions. Cela pourrait être une piste pour résoudre les débats actuels sur la mémoire de reconnaissance. Finalement, (3) le LTM est aussi un système de mémoire impliqué dans l'apprentissage et qui peut interagir avec les GB par une compétition. Nous avons aussi mis en évidence une interaction dynamique de type « top -down » et « bottom-up » entre le cortex préfrontal et le cortex OT. En conclusion, les résultats peuvent donner des indices afin de mieux comprendre certains dysfonctionnements de la mémoire liés à l'âge et la maladie d'Alzheimer ainsi qu'à améliorer le développement de traitement. Abstract: The impact of Alzheimer's disease is devastating for the daily life of the affected patients, with progressive loss of memory and other cognitive skills until dementia. We still lack disease modifying treatment and there is also a great amount of uncertainty regarding the accuracy of diagnostic classification in the early stages of AD. The anatomical signature of AD, in particular the medial temporal lobe (MTL) atrophy measured with neuroimaging, can be used as an early in vivo biomarker in early stages of AD. However, despite the evident role of MTL in memory, we know that the derived predictive anatomical model based only on measures of brain atrophy in MTL does not explain all clinical cases. Throughout my thesis, I have conducted three projects to understand the anatomy and the functioning of MTL on (1) disease's progression, (2) memory process and (3) learning process. I was interested in a population with mild cognitive impairment (MCI), at risk for AD. The objective of the first project was to test the hypothesis that factors, other than the cognitive ones, such as the personality traits, can explain inter-individual differences in the MTL. Moreover, the phenotypic diversity in the manifestations of preclinical AD arises also from the limited knowledge of memory and learning processes in healthy brain. The objective of the second project concerns the investigation of sub-regions of the MTL, and more particularly their contributions in the different components of recognition memory in healthy subjects. To study that, I have used a new multivariate method as well as MRI at high resolution to test the contribution of those sub-regions in the processes of familiarity and recollection. Finally, the objective of the third project was to test the contribution of the MTL as a memory system in learning and the dynamic interaction between memory systems during learning. The results of the first project show that, beyond cognitive state of impairment observed in the population with MCI, the personality traits can explain the inter-individual differences in the MTL; notably with a higher contribution of neuroticism linked to proneness to stress and depression. My study has allowed identifying a pattern of anatomical abnormality in the MTL related to personality with measures of volume and mean diffusion of the tissue. That pattern is characterized by right-left asymmetry in MTL and an anterior to posterior gradient within MTL. I have interpreted that result by tissue and neurochemical properties differently sensitive to stress. Results of my second project have contributed to the actual debate on the contribution of MTL sub-regions in the processes of familiarity and recollection. Using a new multivariate method, the results support firstly a dissociation of the subregions associated with different memory components. The hippocampus was mostly associated with recollection and the surrounding parahippocampal cortex, with familiarity type of memory. Secondly, the activation corresponding to the mensic trace for each type of memory is characterized by a distinct spatial distribution. The specific neuronal representation, "sparse-distributed", associated with recollection in the hippocampus would be the best way to rapidly encode detailed memories without overwriting previously stored memories. In the third project, I have created a learning task with functional MRI to sudy the processes of learning of probabilistic associations based on feedback/reward. That study allowed me to highlight the role of the MTL in learning and the interaction between different memory systems such as the procedural memory, the perceptual memory or priming and the working memory. We have found activations in the MTL corresponding to a process of episodic memory; the basal ganglia (BG), to a procedural memory and reward; the occipito-temporal (OT) cortex, to a perceptive memory or priming and the prefrontal cortex, to working memory. We have also observed that those regions can interact; the relation type between the MTL and the BG has been interpreted as a competition. In addition, with a dynamic causal model, I have demonstrated a "top-down" influence from cortical regions associated with high level cortical area such as the prefrontal cortex on lower level cortical regions such as the OT cortex. That influence decreases during learning; that could correspond to a mechanism linked to a diminution of prediction error. My interpretation is that this is at the origin of the semantic knowledge. I have also shown that the subject's choice and the associated brain activation are influenced by personality traits and negative affects. Overall results of this thesis have brought me to propose (1) a model explaining the possible mechanism linked to the influence of personality on the MTL in a population with MCI, (2) a dissociation of MTL sub-regions in different memory types and a neuronal representation specific to each region. This could be a cue to resolve the actual debates on recognition memory. Finally, (3) the MTL is also a system involved in learning and that can interact with the BG by a competition. We have also shown a dynamic interaction of « top -down » and « bottom-up » types between the pre-frontal cortex and the OT cortex. In conclusion, the results could give cues to better understand some memory dysfunctions in aging and Alzheimer's disease and to improve development of treatment.

Description of epidemics of coffee rust with neural networks

The objective of this paper was to evaluate the potential of neural networks (NN) as an alternative method to the basic epidemiological approach to describe epidemics of coffee rust. The NN was developed from the intensities of coffee (Coffea arabica) rust along with the climatic variables collected in Lavras-MG between 13 February 1998 and 20 April 2001. The NN was built with climatic variables that were either selected in a stepwise regression analysis or by the Braincel® system, software for NN building. Fifty-nine networks and 26 regression models were tested. The best models were selected based on small values of the mean square deviation (MSD) and of the mean prediction error (MPE). For the regression models, the highest coefficients of determination (R²) were used. The best model developed with neural networks had an MSD of 4.36 and an MPE of 2.43%. This model used the variables of minimum temperature, production, relative humidity of the air, and irradiance 30 days before the evaluation of disease. The best regression model was developed from 29 selected climatic variables in the network. The summary statistics for this model were: MPE=6.58%, MSE=4.36, and R²=0.80. The elaborated neural networks from a time series also were evaluated to describe the epidemic. The incidence of coffee rust at four previous fortnights resulted in a model with MPE=4.72% and an MSD=3.95.

Top-down and Bottom-up influences on ACC activation: Evaluation of a proposed model of the feedback-related negativity

The Feedback-Related Negativity (FRN) is thought to reflect the dopaminergic prediction error signal from the subcortical areas to the ACC (i.e., a bottom-up signal). Two studies were conducted in order to test a new model of FRN generation, which includes direct modulating influences of medial PFC (i.e., top-down signals) on the ACC at the time of the FRN. Study 1 examined the effects of one’s sense of control (top-down) and of informative cues (bottom-up) on the FRN measures. In Study 2, sense of control and instruction-based (top-down) and probability-based expectations (bottom-up) were manipulated to test the proposed model. The results suggest that any influences of medial PFC on the activity of the ACC that occur in the context of incentive tasks are not direct. The FRN was shown to be sensitive to salient stimulus characteristics. The results of this dissertation partially support the reinforcement learning theory, in that the FRN is a marker for prediction error signal from subcortical areas. However, the pattern of results outlined here suggests that prediction errors are based on salient stimulus characteristics and are not reward specific. A second goal of this dissertation was to examine whether ACC activity, measured through the FRN, is altered in individuals at-risk for problem-gambling behaviour (PG). Individuals in this group were more sensitive to the valence of the outcome in a gambling task compared to not at-risk individuals, suggesting that gambling contexts increase the sensitivity of the reward system to valence of the outcome in individuals at risk for PG. Furthermore, at-risk participants showed an increased sensitivity to reward characteristics and a decreased response to loss outcomes. This contrasts with those not at risk whose FRNs were sensitive to losses. As the results did not replicate previous research showing attenuated FRNs in pathological gamblers, it is likely that the size and time of the FRN does not change gradually with increasing risk of maladaptive behaviour. Instead, changes in ACC activity reflected by the FRN in general can be observed only after behaviour becomes clinically maladaptive or through comparison between different types of gain/loss outcomes.

Sélection de modèle d'imputation à partir de modèles bayésiens hiérarchiques linéaires multivariés

Les logiciels utilisés sont Splus et R.

Modèle informatique du coapprentissage des ganglions de la base et du cortex : l'apprentissage par renforcement et le développement de représentations

Tout au long de la vie, le cerveau développe des représentations de son environnement permettant à l’individu d’en tirer meilleur profit. Comment ces représentations se développent-elles pendant la quête de récompenses demeure un mystère. Il est raisonnable de penser que le cortex est le siège de ces représentations et que les ganglions de la base jouent un rôle important dans la maximisation des récompenses. En particulier, les neurones dopaminergiques semblent coder un signal d’erreur de prédiction de récompense. Cette thèse étudie le problème en construisant, à l’aide de l’apprentissage machine, un modèle informatique intégrant de nombreuses évidences neurologiques. Après une introduction au cadre mathématique et à quelques algorithmes de l’apprentissage machine, un survol de l’apprentissage en psychologie et en neuroscience et une revue des modèles de l’apprentissage dans les ganglions de la base, la thèse comporte trois articles. Le premier montre qu’il est possible d’apprendre à maximiser ses récompenses tout en développant de meilleures représentations des entrées. Le second article porte sur l'important problème toujours non résolu de la représentation du temps. Il démontre qu’une représentation du temps peut être acquise automatiquement dans un réseau de neurones artificiels faisant office de mémoire de travail. La représentation développée par le modèle ressemble beaucoup à l’activité de neurones corticaux dans des tâches similaires. De plus, le modèle montre que l’utilisation du signal d’erreur de récompense peut accélérer la construction de ces représentations temporelles. Finalement, il montre qu’une telle représentation acquise automatiquement dans le cortex peut fournir l’information nécessaire aux ganglions de la base pour expliquer le signal dopaminergique. Enfin, le troisième article évalue le pouvoir explicatif et prédictif du modèle sur différentes situations comme la présence ou l’absence d’un stimulus (conditionnement classique ou de trace) pendant l’attente de la récompense. En plus de faire des prédictions très intéressantes en lien avec la littérature sur les intervalles de temps, l’article révèle certaines lacunes du modèle qui devront être améliorées. Bref, cette thèse étend les modèles actuels de l’apprentissage des ganglions de la base et du système dopaminergique au développement concurrent de représentations temporelles dans le cortex et aux interactions de ces deux structures.