Emotionally negative stimuli can overcome attentional deficits in patients with visuo-spatial hemineglect.

Left unilateral spatial neglect resulting from right brain damage is characterized by loss of awareness for stimuli in the contralesional side of space, despite intact visual pathways. We examined using fMRI whether patients with neglect are more likely to consciously detect in the neglected hemifield, emotionally negative complex scenes rather than visually similar neutral pictures and if so, what neural mechanisms mediate this effect. Photographs of emotional and neutral scenes taken from the IAPS were presented in a divided visual field paradigm. As expected, the detection rate for emotional stimuli presented in the neglected field was higher than for neutral ones. Successful detection of emotional scenes as opposed to neutral stimuli in the left visual field (LVF) produced activations in the parahippocampal and anterior cingulate areas in the right hemisphere. Detection of emotional stimuli presented in the intact right visual field (RVF) activated a distributed network of structures in the left hemisphere, including anterior and posterior cingulate cortex, insula, as well as visual striate and extrastriate areas. LVF-RVF contrasts for emotional stimuli revealed activations in right and left attention related prefrontal areas whereas RVF-LVF comparison showed activations in the posterior cingulate and extrastriate visual cortex in the left hemisphere. An additional analysis contrasting detected vs. undetected emotional LVF stimuli showed involvement of left anterior cingulate, right frontal and extrastriate areas. We hypothesize that beneficial role of emotion in overcoming neglect is achieved by activation of frontal and limbic lobe networks, which provide a privileged access of emotional stimuli to attention by top-down modulation of processing in the higher-order extrastriate visual areas. Our results point to the importance of top-down regulatory role of the frontal attentional systems, which might enhance visual activations and lead to greater salience of emotional stimuli for perceptual awareness.

Topographic organization of V1 projections through the corpus callosum in humans.

The visual cortex in each hemisphere is linked to the opposite hemisphere by axonal projections that pass through the splenium of the corpus callosum. Visual-callosal connections in humans and macaques are found along the V1/V2 border where the vertical meridian is represented. Here we identify the topography of V1 vertical midline projections through the splenium within six human subjects with normal vision using diffusion-weighted MR imaging and probabilistic diffusion tractography. Tractography seed points within the splenium were classified according to their estimated connectivity profiles to topographic subregions of V1, as defined by functional retinotopic mapping. First, we report a ventral-dorsal mapping within the splenium with fibers from ventral V1 (representing the upper visual field) projecting to the inferior-anterior corner of the splenium and fibers from dorsal V1 (representing the lower visual field) projecting to the superior-posterior end. Second, we also report an eccentricity gradient of projections from foveal-to-peripheral V1 subregions running in the anterior-superior to posterior-inferior direction, orthogonal to the dorsal-ventral mapping. These results confirm and add to a previous diffusion MRI study (Dougherty et al., 2005) which identified a dorsal/ventral mapping of human splenial fibers. These findings yield a more detailed view of the structural organization of the splenium than previously reported and offer new opportunities to study structural plasticity in the visual system.

Sample size and statistical power in the hierarchical analysis of variance: applications in morphometry of the nervous system.

Analysis of variance is commonly used in morphometry in order to ascertain differences in parameters between several populations. Failure to detect significant differences between populations (type II error) may be due to suboptimal sampling and lead to erroneous conclusions; the concept of statistical power allows one to avoid such failures by means of an adequate sampling. Several examples are given in the morphometry of the nervous system, showing the use of the power of a hierarchical analysis of variance test for the choice of appropriate sample and subsample sizes. In the first case chosen, neuronal densities in the human visual cortex, we find the number of observations to be of little effect. For dendritic spine densities in the visual cortex of mice and humans, the effect is somewhat larger. A substantial effect is shown in our last example, dendritic segmental lengths in monkey lateral geniculate nucleus. It is in the nature of the hierarchical model that sample size is always more important than subsample size. The relative weight to be attributed to subsample size thus depends on the relative magnitude of the between observations variance compared to the between individuals variance.

Visuo-motor pathways in humans revealed by event-related fMRI.

Whether different brain networks are involved in generating unimanual responses to a simple visual stimulus presented in the ipsilateral versus contralateral hemifield remains a controversial issue. Visuo-motor routing was investigated with event-related functional magnetic resonance imaging (fMRI) using the Poffenberger reaction time task. A 2 hemifield x 2 response hand design generated the "crossed" and "uncrossed" conditions, describing the spatial relation between these factors. Both conditions, with responses executed by the left or right hand, showed a similar spatial pattern of activated areas, including striate and extrastriate areas bilaterally, SMA, and M1 contralateral to the responding hand. These results demonstrated that visual information is processed bilaterally in striate and extrastriate visual areas, even in the "uncrossed" condition. Additional analyses based on sorting data according to subjects' reaction times revealed differential crossed versus uncrossed activity only for the slowest trials, with response strength in infero-temporal cortices significantly correlating with crossed-uncrossed differences (CUD) in reaction times. Collectively, the data favor a parallel, distributed model of brain activation. The presence of interhemispheric interactions and its consequent bilateral activity is not determined by the crossed anatomic projections of the primary visual and motor pathways. Distinct visuo-motor networks need not be engaged to mediate behavioral responses for the crossed visual field/response hand condition. While anatomical connectivity heavily influences the spatial pattern of activated visuo-motor pathways, behavioral and functional parameters appear to also affect the strength and dynamics of responses within these pathways.

Multisensory anatomical pathways.

In order to interact with the multisensory world that surrounds us, we must integrate various sources of sensory information (vision, hearing, touch...). A fundamental question is thus how the brain integrates the separate elements of an object defined by several sensory components to form a unified percept. The superior colliculus was the main model for studying multisensory integration. At the cortical level, until recently, multisensory integration appeared to be a characteristic attributed to high-level association regions. First, we describe recently observed direct cortico-cortical connections between different sensory cortical areas in the non-human primate and discuss the potential role of these connections. Then, we show that the projections between different sensory and motor cortical areas and the thalamus enabled us to highlight the existence of thalamic nuclei that, by their connections, may represent an alternative pathway for information transfer between different sensory and/or motor cortical areas. The thalamus is in position to allow a faster transfer and even an integration of information across modalities. Finally, we discuss the role of these non-specific connections regarding behavioral evidence in the monkey and recent electrophysiological evidence in the primary cortical sensory areas.

Contribution à l'étude neuroanatomique de systèmes enzymatiques impliqués dans le métabolisme énergétique cérébral

RESUME Il a longtemps été admis que le glucose était le principal, sinon le seul substrat du métabolisme énergétique cérébral. Néanmoins, des études récentes indiquent que dans des situations particulières, d'autres substrats peuvent être employés. C'est le cas des monocarboxylates (lactate et pyruvate principalement). Bien que la barrière hématoencéphalique soit peu perméable à ces molécules, elles deviennent néanmoins des substrats possibles si elles sont produites localement. Les deux systèmes enzymatiques pivots des voies glycolytiques et oxydatives sont la lactate déshydrogénase (LDH, EC 1.1.1.27) qui catalyse l'interconversion du pyruvate et du lactate et le complexe pyruvate déshydrogénase qui catalyse la conversion irréversible du pyruvate en acétyl-CoA qui entre dans la respiration mitochondriale. Nous avons étudié la localisation, tant régionale que cellulaire, des isoformes LDH-1, LDH-5 et PDHEla dans le cerveau du chat et dé l'homme au moyen de diverses techniques histologiques. Dans un premier temps, des investigations par hybridation in situ au moyen d'oligosondes marquées au 33P sur de coupes de cerveau de chat ont permis de montrer une différence de l'expression des enzymes à vocation oxydative (LDH-1 et PDHA1, le gène codant pour la protéine PDHEIa) par rapport à LDH-5, isoforme qui catalyse préférentiellement la formation de lactate. LDH-1 et PDHA 1 ont des distributions similaires et sont enrichies dans de nombreuses structures cérébrales, comme l'hippocampe, de nombreux noyaux thalamiques et des structures pontiques. Le cortex cérébral exhibe également une expression importante de LDH-1 et PDH. LDH-5 a par contre une expression largement plus diffuse à travers le cerveau, bien que l'on trouve néanmoins un enrichissement plus important dans l'hippocampe. Ces résultats sont en accord avec les observations que nous avons précédemment publiées chez le rongeur pour LDH-1 et LDH-5 (Laughton et collaborateurs, 2000). Des analyses par PCR en temps réel ont confirmé que dans certaines régions, LDH-1 est exprimée de façon nettement plus importante que LDH-5. Dans un deuxième temps, nous avons appliqué sur des coupes histologiques d'hippocampe et de cortex occipital humain post-mortem des anticorps monoclonaux spécifiques de l'isoforme LDH-5 et la sous-unité PDHela du complexe pyruvate déshydrogénase. Là aussi, les immunoréactions révèlent une ségrégation régionale mais aussi cellulaire des deux enzymes. Dans les deux régions étudiées, LDH-5 est localisée exclusivement dans les astrocytes. Dans le cortex occipital, la matière blanche et également la couche I corticale sont immunopositives pour LDH-5. Dans l'hippocampe, le CA4 et l'alveus exhibe l'immunomarquage le plus intense pour LDH-5. Seuls des neurones (à de rares exceptions quelques astrocytes) sont immunopositifs à l'anticorps monoclonal dirigé contre PDHela. La couche IV du cortex occipital présente la plus forte immunoréaction. Dans l'hippocampe, une immunoréactivité est observée dans le stratum granulosum et à travers la région CA1 jusqu'à la région CA3. L'ensemble de ces résultats montre une hétérogénéité métabolique dans le cerveau et étaye l'hypothèse "astrocyte-neurone lactate shuttle" (ANL5) (Bittar et collaborateurs, 1996; Magistretti et Pellerin, 1999) qui propose que les astrocytes fournissent aux neurones activés du lactate comme substrat alternatif de leur métabolisme énergétique. ABSTRACT For a long time now, glucose has been thought to be the main, if not the sole substrate for brain energy metabolism. Recent data nevertheless suggest that other molecules, such as monocarboxylates (lactate and pyruvate mainly) could be suitable substrates. Although monocarboxylates poorly cross the blood brain barrier (BBB), such substrates could replace glucose if produced locally. The two key enzymatic systems required for the use and production of these substats are lactate dehydrogenase (LDH; EC 1.1.1.27) that catalyses the interconversion of lactate and pyruvate and the pyruvate dehydrogenase complex that irreversibly funnels pyruvate towards the mitochondrial TCA cycle and oxydative phosphorylation. Our study consisted in localizing these different systems with various histochemical procedures in the cat brain and two regions, i.e. hippocampus and primary visual cortex, of the human brain. First, by means of in situ hybridization with 33P labeled oligoprobes, we have demonstrated that the more oxidative enzymes (LDH-1 and PDHA1, the gene coding for PDHEla) are highly expressed in a variety of feline brain structures. These structures include the hippocampus, various thalamic nuclei and the pons. The cerebral cortex exhibits also a high LDH-1 and PDHAl expression. On the other hand, LDH-5 expression is poorer and more diffuse, although the hippocampus does seem to have a higher expression. These fmdings are consistent with our previous observation of the expression of LDH1 and LDH-5 in the rodent brain (Laughton et al, 2000). Real-time PCR (TagMan tm) revealed that, in various regions, LDH-1 is effectively more highly expressed than LDH-5. In a second set of experiments, monoclonal antibodies to LDH-5 and PDHeIa were applied to cryostat sections of post-mortem human hippocampus and occipital cortex. These procedures revealed not only that the two enzymes have different regional distributions, but also distinct cellular localisation. LDH-5 immunoreactivity is solely observed in astrocytes. In the occipital cortex, the white matter and layer I are immunopositive. In the hippocampus, the alveus and CA4 show LDH-5 immunoréactivity. PDHeIa has been detected, with few exceptions, only in neurons. Layer IV of the occipital cortex was most immmunoreactive. In the hippocampus, PDHela immunoreactivity is noticed in the stratum granulosum and through CA 1 to CA3 areas. The overall observations made in this study show that there is a metabolic heterogeneity in the brain and our findings support the hypothesis of an astrocyte-neuron lactate shuttle (ANLS)(Bittar et al., 1996; Magistretti & Pellerin, 1999) where astrocytes export to active neurons lactate to fuel their energy demands.

The brain tracks the energetic value in food images.

Do our brains implicitly track the energetic content of the foods we see? Using electrical neuroimaging of visual evoked potentials (VEPs) we show that the human brain can rapidly discern food's energetic value, vis à vis its fat content, solely from its visual presentation. Responses to images of high-energy and low-energy food differed over two distinct time periods. The first period, starting at approximately 165 ms post-stimulus onset, followed from modulations in VEP topography and by extension in the configuration of the underlying brain network. Statistical comparison of source estimations identified differences distributed across a wide network including both posterior occipital regions and temporo-parietal cortices typically associated with object processing, and also inferior frontal cortices typically associated with decision-making. During a successive processing stage (starting at approximately 300 ms), responses differed both topographically and in terms of strength, with source estimations differing predominantly within prefrontal cortical regions implicated in reward assessment and decision-making. These effects occur orthogonally to the task that is actually being performed and suggest that reward properties such as a food's energetic content are treated rapidly and in parallel by a distributed network of brain regions involved in object categorization, reward assessment, and decision-making.

Changes of MAP2 phosphorylation during brain development.

The microtubule-associated protein MAP2 is essential for development of early neuronal morphology and maintenance of adult neuronal morphology. Several splice variants exist, MAP2a-d, with a lack of MAP2a in cat brain. MAP2 is widely used as a neuronal marker. In this study we compared five monoclonal antibodies (MAbs) against MAP2. They show differences in the immunocytochemical distribution of MAP2 isoforms during development of the visual cortex and cerebellum of the cat. Local and temporal differences were seen with MAb AP18, an antibody directed against a phosphorylation-dependent epitope near the N-terminal end. In large pyramidal dendrites in visual cortex, the AP18 epitope remained in parts immunoreactive after treatment with alkaline phosphatase. Three MAbs, AP14, MT-01, and MT-02, recognized the central region of the MAP2b molecule, which is not present in MAP2c and 2d, and reacted with phosphorylation-independent epitopes. During the first postnatal week the immunostaining in cerebellum differed between antibodies in that some cellular elements in external and internal granular layers and Purkinje cells were stained to various degrees, whereas at later stages staining patterns were similar. At early stages, antibody MT-02 stained cell bodies and dendrites in cerebral cortex and cerebellum. With progressing maturation, immunoreactivity became restricted to distal parts of apical dendrites of pyramidal cells and was absent from perikarya and finer proximal dendrites in cortex. MT-02 did not stain MAP2 in cerebellum of adult animals. This study demonstrates that the immunocytochemical detection of MAP2 depends on modifications such as phosphorylation and conformational changes of the molecule, and that MAP2 staining patterns differ between MAbs. Phosphorylation and specific conformations in the molecule may be essential for modulating function and molecular stability of MAP2, and monoclonal antibodies against such sites may provide tools for studying the functional role of modifications.

Gamma oscillations in V1 are correlated with GABAA receptor density: A multi-modal MEG and Flumazenil-PET study.

High-frequency oscillations in the gamma-band reflect rhythmic synchronization of spike timing in active neural networks. The modulation of gamma oscillations is a widely established mechanism in a variety of neurobiological processes, yet its neurochemical basis is not fully understood. Modeling, in-vitro and in-vivo animal studies suggest that gamma oscillation properties depend on GABAergic inhibition. In humans, search for evidence linking total GABA concentration to gamma oscillations has led to promising -but also to partly diverging- observations. Here, we provide the first evidence of a direct relationship between the density of GABAA receptors and gamma oscillatory gamma responses in human primary visual cortex (V1). By combining Flumazenil-PET (to measure resting-levels of GABAA receptor density) and MEG (to measure visually-induced gamma oscillations), we found that GABAA receptor densities correlated positively with the frequency and negatively with amplitude of visually-induced gamma oscillations in V1. Our findings demonstrate that gamma-band response profiles of primary visual cortex across healthy individuals are shaped by GABAA-receptor-mediated inhibitory neurotransmission. These results bridge the gap with in-vitro and animal studies and may have future clinical implications given that altered GABAergic function, including dysregulation of GABAA receptors, has been related to psychiatric disorders including schizophrenia and depression.

The natural hallucinogen 5-MeO-DMT, component of Ayahuasca, disrupts cortical function in rats: reversal by antipsychotic drugs

5-Methoxy-N,N-dimethyltryptamine (5-MeO-DMT) is a natural hallucinogen component of Ayahuasca, an Amazonian beverage traditionally used for ritual, religious and healing purposes that is being increasingly used for recreational purposes in US and Europe. 5MeO-DMT is of potential interest for schizophrenia research owing to its hallucinogenic properties. Two other psychotomimetic agents, phencyclidine and 2,5-dimethoxy-4-iodo-phenylisopropylamine (DOI), markedly disrupt neuronal activity and reduce the power of low frequency cortical oscillations (<4 Hz, LFCO) in rodent medial prefrontal cortex (mPFC). Here we examined the effect of 5-MeO-DMT on cortical function and its potential reversal by antipsychotic drugs. Moreover, regional brain activity was assessed by blood-oxygen level dependent (BOLD) functional magnetic resonance imaging (fMRI). 5-MeO-DMT disrupted mPFC activity, increasing and decreasing the discharge of 51 and 35% of the recorded pyramidal neurons, and reducing (−31%) the power of LFCO. The latter effect depended on 5-HT1A and 5-HT2A receptor activation and was reversed by haloperidol, clozapine, risperidone, and the mGlu2/3 agonist LY379268. Likewise, 5-MeO-DMT decreased BOLD responses in visual cortex (V1) and mPFC. The disruption of cortical activity induced by 5-MeO-DMT resembles that produced by phencyclidine and DOI. This, together with the reversal by antipsychotic drugs, suggests that the observed cortical alterations are related to the psychotomimetic action of 5-MeO-DMT. Overall, the present model may help to understand the neurobiological basis of hallucinations and to identify new targets in antipsychotic drug development.

Neural mechanisms underlying conscious and unconscious vision. Evidence from event-related potentials and transcranial magnetic stimulation

Vision affords us with the ability to consciously see, and use this information in our behavior. While research has produced a detailed account of the function of the visual system, the neural processes that underlie conscious vision are still debated. One of the aims of the present thesis was to examine the time-course of the neuroelectrical processes that correlate with conscious vision. The second aim was to study the neural basis of unconscious vision, that is, situations where a stimulus that is not consciously perceived nevertheless influences behavior. According to current prevalent models of conscious vision, the activation of visual cortical areas is not, as such, sufficient for consciousness to emerge, although it might be sufficient for unconscious vision. Conscious vision is assumed to require reciprocal communication between cortical areas, but views differ substantially on the extent of this recurrent communication. Visual consciousness has been proposed to emerge from recurrent neural interactions within the visual system, while other models claim that more widespread cortical activation is needed for consciousness. Studies I-III compared models of conscious vision by studying event-related potentials (ERP). ERPs represent the brain’s average electrical response to stimulation. The results support the model that associates conscious vision with activity localized in the ventral visual cortex. The timing of this activity corresponds to an intermediate stage in visual processing. Earlier stages of visual processing may influence what becomes conscious, although these processes do not directly enable visual consciousness. Late processing stages, when more widespread cortical areas are activated, reflect the access to and manipulation of contents of consciousness. Studies IV and V concentrated on unconscious vision. By using transcranial magnetic stimulation (TMS) we show that when early visual cortical processing is disturbed so that subjects fail to consciously perceive visual stimuli, they may nevertheless guess (above chance-level) the location where the visual stimuli were presented. However, the results also suggest that in a similar situation, early visual cortex is necessary for both conscious and unconscious perception of chromatic information (i.e. color). Chromatic information that remains unconscious may influence behavioral responses when activity in visual cortex is not disturbed by TMS. Our results support the view that early stimulus-driven (feedforward) activation may be sufficient for unconscious processing. In conclusion, the results of this thesis support the view that conscious vision is enabled by a series of processing stages. The processes that most closely correlate with conscious vision take place in the ventral visual cortex ~200 ms after stimulus presentation, although preceding time-periods and contributions from other cortical areas such as the parietal cortex are also indispensable. Unconscious vision relies on intact early visual activation, although the location of visual stimulus may be unconsciously resolved even when activity in the early visual cortex is interfered with.

Horizontal projections of area 17 in Cebus monkeys: metric features, and modular and laminar distribution

Metric features and modular and laminar distributions of intrinsic projections of area 17 were studied in Cebus apella. Anterogradely and retrogradely labeled cell appendages were obtained using both saturated pellets and iontophoretic injections of biocytin into the operculum. Laminar and modular distributions of the labeled processes were analyzed using Nissl counterstaining, and/or cytochrome oxidase and/or NADPH-diaphorase histochemistry. We distinguished three labeled cell types: pyramidal, star pyramidal and stellate cells located in supragranular cortical layers (principally in layers IIIa, IIIb a, IIIb ß and IIIc). Three distinct axon terminal morphologies were found, i.e., Ia, Ib and II located in granular and supragranular layers. Both complete and partial segregation of group I axon terminals relative to the limits of the blobs of V1 were found. The results are compatible with recent evidence of incomplete segregation of visual information flow in V1 of Old and New World primates

GABA-induced inactivation of Cebus apella V2 neurons: effects on orientation tuning and direction selectivity

We investigated the GABA-induced inactivation of V2 neurons and terminals on the receptive field properties of this area in an anesthetized and paralyzedCebus apella monkey. Extracellular single-unit activity was recorded using tungsten microelectrodes in a monkey before and after pressure-injection of a 0.25 or 0.5 M GABA solution. The visual stimulus consisted of a bar moving in 8 possible directions. In total, 24 V2 neurons were studied before and after blocker injections in 4 experimental sessions following GABA injection into area V2. A group of 10 neurons were studied over a short period. An additional 6 neurons were investigated over a long period after the GABA injection. A third group of 8 neurons were studied over a very long period. Overall, these 24 neurons displayed an early (1-20 min) significant general decrease in excitability with concomitant changes in orientation or direction selectivity. GABA inactivation in area V2 produced robust inhibition in 80% and a significant change in directional selectivity in 60% of the neurons examined. These GABA projections are capable of modulating not only levels of spontaneous and driven activity of V2 neurons but also receptive field properties such as direction selectivity.

Effets de l’alcoolisme fœtal sur le développement du corps genouillé latéral du singe

L’exposition du fœtus à l’éthanol est reconnue comme étant la principale cause de maladies évitables lors du développement. Une forte exposition à l’alcool durant la gestation peut occasionner des dysmorphies cranio-faciales et des retards mentaux, ainsi que des troubles d’apprentissages et du comportement. Le développement du système visuel est également perturbé chez une grande majorité d’enfants qui ont été exposés à l’alcool. Lorsque les doses prises sont élevées, le système visuel peut présenter une panoplie de symptômes comme une augmentation de la tortuosité des vaisseaux rétiniens, de la myopie, de l’hypermétropie, du strabisme et une hypoplasie du nerf optique. Cependant, très peu d’études se sont penchées sur les effets de plus faibles doses sur le développement du système visuel du primate. Le singe est un excellent modèle pour étudier le système visuel car il possède plusieurs similitudes avec l’humain tant au niveau développemental qu’au niveau structurel. De plus, le singe utilisé, le Chlrocebus aethiops sabeus, possède l’avantage que des individus de cette espèce ont une consommation naturelle et volontaire à l’alcool. Une étude (Clarren et al., 1990) a suggéré qu’une faible exposition à l’alcool du fœtus du primate non humain occasionnait une diminution du nombre de cellules ganglionnaires de la rétine (CGRs). Étant donné que le corps genouillé latéral dorsal (CGLd) reçoit la plupart de ses intrants de la rétine, il est raisonnable d’assumer que les couches rétino-récipientes du CGLd devraient être aussi affectées. Nous avons alors émis l’hypothèse que le CGLd devrait également subir une diminution du nombre de neurones. Pour la première fois, nous avons utilisé une méthode stéréologique pour quantifier le nombre de cellules dans les couches parvo- (P) et magnocellulaires (M) du CGLd. Contrairement à notre hypothèse de départ, nous n’avons pas observé de diminution dans le nombre global de neurones dans le CGLd des animaux exposés à l’alcool par rapport à des sujets contrôles, ni une diminution de son volume. Nous avons toutefois observé une diminution de la taille du corps cellulaire seulement dans la population M du CGLd. Ces résultats suggèrent que le système visuel est affecté par une faible exposition à l’alcool durant son développement qui devrait se traduire sur le comportement par des déficits dans les fonctions de la voie M.

Apprentissage de représentations sur-complètes par entraînement d’auto-encodeurs

Les avancés dans le domaine de l’intelligence artificielle, permettent à des systèmes informatiques de résoudre des tâches de plus en plus complexes liées par exemple à la vision, à la compréhension de signaux sonores ou au traitement de la langue. Parmi les modèles existants, on retrouve les Réseaux de Neurones Artificiels (RNA), dont la popularité a fait un grand bond en avant avec la découverte de Hinton et al. [22], soit l’utilisation de Machines de Boltzmann Restreintes (RBM) pour un pré-entraînement non-supervisé couche après couche, facilitant grandement l’entraînement supervisé du réseau à plusieurs couches cachées (DBN), entraînement qui s’avérait jusqu’alors très difficile à réussir. Depuis cette découverte, des chercheurs ont étudié l’efficacité de nouvelles stratégies de pré-entraînement, telles que l’empilement d’auto-encodeurs traditionnels(SAE) [5, 38], et l’empilement d’auto-encodeur débruiteur (SDAE) [44]. C’est dans ce contexte qu’a débuté la présente étude. Après un bref passage en revue des notions de base du domaine de l’apprentissage machine et des méthodes de pré-entraînement employées jusqu’à présent avec les modules RBM, AE et DAE, nous avons approfondi notre compréhension du pré-entraînement de type SDAE, exploré ses différentes propriétés et étudié des variantes de SDAE comme stratégie d’initialisation d’architecture profonde. Nous avons ainsi pu, entre autres choses, mettre en lumière l’influence du niveau de bruit, du nombre de couches et du nombre d’unités cachées sur l’erreur de généralisation du SDAE. Nous avons constaté une amélioration de la performance sur la tâche supervisée avec l’utilisation des bruits poivre et sel (PS) et gaussien (GS), bruits s’avérant mieux justifiés que celui utilisé jusqu’à présent, soit le masque à zéro (MN). De plus, nous avons démontré que la performance profitait d’une emphase imposée sur la reconstruction des données corrompues durant l’entraînement des différents DAE. Nos travaux ont aussi permis de révéler que le DAE était en mesure d’apprendre, sur des images naturelles, des filtres semblables à ceux retrouvés dans les cellules V1 du cortex visuel, soit des filtres détecteurs de bordures. Nous aurons par ailleurs pu montrer que les représentations apprises du SDAE, composées des caractéristiques ainsi extraites, s’avéraient fort utiles à l’apprentissage d’une machine à vecteurs de support (SVM) linéaire ou à noyau gaussien, améliorant grandement sa performance de généralisation. Aussi, nous aurons observé que similairement au DBN, et contrairement au SAE, le SDAE possédait une bonne capacité en tant que modèle générateur. Nous avons également ouvert la porte à de nouvelles stratégies de pré-entraînement et découvert le potentiel de l’une d’entre elles, soit l’empilement d’auto-encodeurs rebruiteurs (SRAE).