Adaptive learning in agents behaviour: a framework for electricity markets simulation

Electricity markets are complex environments with very particular characteristics. A critical issue regarding these specific characteristics concerns the constant changes they are subject to. This is a result of the electricity markets’ restructuring, which was performed so that the competitiveness could be increased, but it also had exponential implications in the increase of the complexity and unpredictability in those markets scope. The constant growth in markets unpredictability resulted in an amplified need for market intervenient entities in foreseeing market behaviour. The need for understanding the market mechanisms and how the involved players’ interaction affects the outcomes of the markets, contributed to the growth of usage of simulation tools. Multi-agent based software is particularly well fitted to analyze dynamic and adaptive systems with complex interactions among its constituents, such as electricity markets. This dissertation presents ALBidS – Adaptive Learning strategic Bidding System, a multiagent system created to provide decision support to market negotiating players. This system is integrated with the MASCEM electricity market simulator, so that its advantage in supporting a market player can be tested using cases based on real markets’ data. ALBidS considers several different methodologies based on very distinct approaches, to provide alternative suggestions of which are the best actions for the supported player to perform. The approach chosen as the players’ actual action is selected by the employment of reinforcement learning algorithms, which for each different situation, simulation circumstances and context, decides which proposed action is the one with higher possibility of achieving the most success. Some of the considered approaches are supported by a mechanism that creates profiles of competitor players. These profiles are built accordingly to their observed past actions and reactions when faced with specific situations, such as success and failure. The system’s context awareness and simulation circumstances analysis, both in terms of results performance and execution time adaptation, are complementary mechanisms, which endow ALBidS with further adaptation and learning capabilities.

Olfactory traces and spatial learning in rats.

We conducted an experiment to assess the use of olfactory traces for spatial orientation in an open environment in rats, Rattus norvegicus. We trained rats to locate a food source at a fixed location from different starting points, in the presence or absence of visual information. A single food source was hidden in an array of 19 petri dishes regularly arranged in an open-field arena. Rats were trained to locate the food source either in white light (with full access to distant visuospatial information) or in darkness (without any visual information). In both cases, the goal was in a fixed location relative to the spatial frame of reference. The results of this experiment revealed that the presence of noncontrolled olfactory traces coherent with the spatial frame of reference enables rats to locate a unique position as accurately in darkness as with full access to visuospatial information. We hypothesize that the olfactory traces complement the use of other orientation mechanisms, such as path integration or the reliance on visuospatial information. This experiment demonstrates that rats can rely on olfactory traces for accurate orientation, and raises questions about the establishment of such traces in the absence of any other orientation mechanism. Copyright 1998 The Association for the Study of Animal Behaviour.

Is sociality driven by the costs of dispersal or the benefits of philopatry? A role for kin-discrimination mechanisms.

The role of ecological constraints in promoting sociality is currently much debated. Using a direct-fitness approach, we show this role to depend on the kin-discrimination mechanisms underlying social interactions. Altruism cannot evolve under spatially based discrimination, unless ecological constraints prevent complete dispersal. Increasing constraints enhances both the proportion of philopatric (and thereby altruistic) individuals and the level of altruistic investments conceded in pairwise interactions. Familiarity-based discrimination, by contrast, allows philopatry and altruism to evolve at significant levels even in the absence of ecological constraints. Increasing constraints further enhances the proportion of philopatric (and thereby altruistic) individuals but not the level of altruism conceded. Ecological constraints are thus more likely to affect social evolution in species in which restricted cognitive abilities, large group size, and/or limited period of associative learning force investments to be made on the basis of spatial cues.

Plastic brain mechanisms for attaining auditory temporal order judgment proficiency.

Accurate perception of the order of occurrence of sensory information is critical for the building up of coherent representations of the external world from ongoing flows of sensory inputs. While some psychophysical evidence reports that performance on temporal perception can improve, the underlying neural mechanisms remain unresolved. Using electrical neuroimaging analyses of auditory evoked potentials (AEPs), we identified the brain dynamics and mechanism supporting improvements in auditory temporal order judgment (TOJ) during the course of the first vs. latter half of the experiment. Training-induced changes in brain activity were first evident 43-76 ms post stimulus onset and followed from topographic, rather than pure strength, AEP modulations. Improvements in auditory TOJ accuracy thus followed from changes in the configuration of the underlying brain networks during the initial stages of sensory processing. Source estimations revealed an increase in the lateralization of initially bilateral posterior sylvian region (PSR) responses at the beginning of the experiment to left-hemisphere dominance at its end. Further supporting the critical role of left and right PSR in auditory TOJ proficiency, as the experiment progressed, responses in the left and right PSR went from being correlated to un-correlated. These collective findings provide insights on the neurophysiologic mechanism and plasticity of temporal processing of sounds and are consistent with models based on spike timing dependent plasticity.

Evaluation of Machine Learning Classifiers for Mobile Network Intrusion Detection Systems

Mobile malwares are increasing with the growing number of Mobile users. Mobile malwares can perform several operations which lead to cybersecurity threats such as, stealing financial or personal information, installing malicious applications, sending premium SMS, creating backdoors, keylogging and crypto-ransomware attacks. Knowing the fact that there are many illegitimate Applications available on the App stores, most of the mobile users remain careless about the security of their Mobile devices and become the potential victim of these threats. Previous studies have shown that not every antivirus is capable of detecting all the threats; due to the fact that Mobile malwares use advance techniques to avoid detection. A Network-based IDS at the operator side will bring an extra layer of security to the subscribers and can detect many advanced threats by analyzing their traffic patterns. Machine Learning(ML) will provide the ability to these systems to detect unknown threats for which signatures are not yet known. This research is focused on the evaluation of Machine Learning classifiers in Network-based Intrusion detection systems for Mobile Networks. In this study, different techniques of Network-based intrusion detection with their advantages, disadvantages and state of the art in Hybrid solutions are discussed. Finally, a ML based NIDS is proposed which will work as a subsystem, to Network-based IDS deployed by Mobile Operators, that can help in detecting unknown threats and reducing false positives. In this research, several ML classifiers were implemented and evaluated. This study is focused on Android-based malwares, as Android is the most popular OS among users, hence most targeted by cyber criminals. Supervised ML algorithms based classifiers were built using the dataset which contained the labeled instances of relevant features. These features were extracted from the traffic generated by samples of several malware families and benign applications. These classifiers were able to detect malicious traffic patterns with the TPR upto 99.6% during Cross-validation test. Also, several experiments were conducted to detect unknown malware traffic and to detect false positives. These classifiers were able to detect unknown threats with the Accuracy of 97.5%. These classifiers could be integrated with current NIDS', which use signatures, statistical or knowledge-based techniques to detect malicious traffic. Technique to integrate the output from ML classifier with traditional NIDS is discussed and proposed for future work.

An exploration of the experience of implementing the second step violence prevention curriculum in one primary general learning disabilities class /

Violence has always been a part of the human experience, and therefore, a popular topic for research. It is a controversial issue, mostly because the possible sources of violent behaviour are so varied, encompassing both biological and environmental factors. However, very little disagreement is found regarding the severity of this societal problem. Most researchers agree that the number and intensity of aggressive acts among adults and children is growing. Not surprisingly, many educational policies, programs, and curricula have been developed to address this concern. The research favours programs which address the root causes of violence and seek to prevent rather than provide consequences for the undesirable behaviour. But what makes a violence prevention program effective? How should educators choose among the many curricula on the market? After reviewing the literature surrounding violence prevention programs and their effectiveness, The Second Step Violence Prevention Curriculum surfaced as unique in many ways. It was designed to address the root causes of violence in an active, student-centred way. Empathy training, anger management, interpersonal cognitive problem solving, and behavioural social skills form the basis of this program. Published in 1992, the program has been the topic of limited research, almost entirely carried out using quantitative methodologies.The purpose of this study was to understand what happens when the Second Step Violence Prevention Curriculum is implemented with a group of students and teachers. I was not seeking a statistical correlation between the frequency of violence and program delivery, as in most prior research. Rather, I wished to gain a deeper understanding of the impact ofthe program through the eyes of the participants. The Second Step Program was taught to a small, primary level, general learning disabilities class by a teacher and student teacher. Data were gathered using interviews with the teachers, personal observations, staff reports, and my own journal. Common themes across the four types of data collection emerged during the study, and these themes were isolated and explored for meaning. Findings indicate that the program does not offer a "quick fix" to this serious problem. However, several important discoveries were made. The teachers feU that the program was effective despite a lack of concrete evidence to support this claim. They used the Second Step strategies outside their actual instructional time and felt it made them better educators and disciplinarians. The students did not display a marked change in their behaviour during or after the program implementation, but they were better able to speak about their actions, the source of their aggression, and the alternatives which were available. Although they were not yet transferring their knowledge into positive action,a heightened awareness was evident. Finally, staff reports and my own journal led me to a deeper understanding ofhow perception frames reality. The perception that the program was working led everyone to feel more empowered when a violent incident occurred, and efforts were made to address the cause rather than merely to offer consequences. A general feeling that we were addressing the problem in a productive way was prevalent among the staff and students involved. The findings from this investigation have many implications for research and practice. Further study into the realm of violence prevention is greatly needed, using a balance of quantitative and qualitative methodologies. Such a serious problem can only be effectively addressed with a greater understanding of its complexities. This study also demonstrates the overall positive impact of the Second Step Violence Prevention Curriculum and, therefore, supports its continued use in our schools.

The role of learning and growth hormone-releasing factor in the control of protein intake in rats /

Both learning and basic biological mechanisms have been shown to play a role in the control of protein int^e. It has previously been shown that rats can adapt their dietary selection patterns successfully in the face of changing macronutrient requirements and availability. In particular, it has been demonstrated that when access to dietary protein is restricted for a period of time, rats selectively increase their consumption of a proteincontaining diet when it becomes available. Furthermore, it has been shown that animals are able to associate various orosensory cues with a food's nutrient content. In addition to the role that learning plays in food intake, there are also various biological mechanisms that have been shown to be involved in the control of feeding behaviour. Numerous studies have documented that various hormones and neurotransmitter substances mediate food intake. One such hormone is growth hormone-releasing factor (GRF), a peptide that induces the release of growth hormone (GH) from the anterior pituitary gland. Recent research by Vaccarino and Dickson ( 1 994) suggests that GRF may stimulate food intake by acting as a neurotransmitter in the suprachiasmatic nucleus (SCN) and the adjacent medial preoptic area (MPOA). In particular, when GRF is injected directly into the SCN/MPOA, it has been shown to selectively enhance the intake of protein in both fooddeprived and sated rats. Thus, GRF may play a role in activating protein consumption generally, and when animals have a need for protein, GRF may serve to trigger proteinseeking behaviour. Although researchers have separately examined the role of learning and the central mechanisms involved in the control of protein selection, no one has yet attempted to bring together these two lines of study. Thus, the purpose of this study is to join these two parallel lines of research in order to further our understanding of mechanisms controlling protein selection. In order to ascertain the combined effects that GRF and learning have on protein intake several hypothesis were examined. One major hypothesis was that rats would successfully alter their dietary selection patterns in response to protein restriction. It was speculated that rats kept on a nutritionally complete maintenance diet (NCMD) would consume equal amount of the intermittently presented high protein conditioning diet (HPCD) and protein-free conditioning diet (PFCD). However, it was hypothesized that rats kept on a protein-free maintenance diet (PFMD) would selectively increase their intake of the HPCD. Another hypothesis was that rats would learn to associate a distinct marker flavour with the nutritional content of the diets. If an animal is able to make the association between a marker flavour and the nutrient content of the food, then it is hypothesized that they will consume more of a mixed diet (equal portion HPCD and PFCD) with the marker flavour that was previously paired with the HPCD (Mixednp-f) when kept on the PFMD. In addition, it was hypothesized that intracranial injection of GRF into the SCN/MPOA would result in a selective increase in HPCD as well as Mixednp-t consumption. Results demonstrated that rats did in fact selectively increase their consumption of the flavoured HPCD and Mixednp-f when kept on the NCMD. These findings indicate that the rats successfully learned about the nutrient content of the conditioning diets and were able to associate a distinct marker flavour with the nutrient content of the diets. However, the results failed to support previous findings that GRF increases protein intake. In contrast, the administration of GRF significantly reduced consumption of HPCD during the first hour of testing as compared to the no injection condition. In addition, no differences in the intake of the HPCD were found between the GRF and vehicle condition. Because GRF did not selectively increase HPCD consumption, it was not surprising that GRF also did not increase MixedHP-rintake. What was interesting was that administration of GRF and vehicle did not reduc^Mixednp-f consumption as it had decreased HPCD consumption.

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

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

Evolution of Various Library Instruction Strategies: Using Student Feedback to Create and Enhance Online Active Learning Assignments

This case study traces the evolution of library assignments for biological science students from paper-based workbooks in a blended (hands-on) workshop to blended learning workshops using online assignments to online active learning modules which are stand-alone without any face-to-face instruction. As the assignments evolved to adapt to online learning supporting materials in the form of PDFs (portable document format), screen captures and screencasting were embedded into the questions as teaching moments to replace face-to-face instruction. Many aspects of the evolution of the assignment were based on student feedback from evaluations, input from senior lab demonstrators and teaching assistants, and statistical analysis of the students’ performance on the assignment. Advantages and disadvantages of paper-based and online assignments are discussed. An important factor for successful online learning may be the ability to get assistance.

Cholinergic enhancement of perceptual learning : behavioral, physiological, and neuro-pharmacological study in the rat primary visual cortex

Les cortices sensoriels sont des régions cérébrales essentielles pour la perception. En particulier, le cortex visuel traite l’information visuelle en provenance de la rétine qui transite par le thalamus. Les neurones sont les unités fonctionnelles qui transforment l'information sensorielle en signaux électriques, la transfèrent vers le cortex et l'intègrent. Les neurones du cortex visuel sont spécialisés et analysent différents aspects des stimuli visuels. La force des connections entre les neurones peut être modulée par la persistance de l'activité pré-synaptique et induit une augmentation ou une diminution du signal post-synaptique à long terme. Ces modifications de la connectivité synaptique peuvent induire la réorganisation de la carte corticale, c’est à dire la représentation de ce stimulus et la puissance de son traitement cortical. Cette réorganisation est connue sous le nom de plasticité corticale. Elle est particulièrement active durant la période de développement, mais elle s’observe aussi chez l’adulte, par exemple durant l’apprentissage. Le neurotransmetteur acétylcholine (ACh) est impliqué dans de nombreuses fonctions cognitives telles que l’apprentissage ou l’attention et il est important pour la plasticité corticale. En particulier, les récepteurs nicotiniques et muscariniques du sous-type M1 et M2 sont les récepteurs cholinergiques impliqués dans l’induction de la plasticité corticale. L’objectif principal de la présente thèse est de déterminer les mécanismes de plasticité corticale induits par la stimulation du système cholinergique au niveau du télencéphale basal et de définir les effets sur l’amélioration de la perception sensorielle. Afin d’induire la plasticité corticale, j’ai jumelé des stimulations visuelles à des injections intracorticales d’agoniste cholinergique (carbachol) ou à une stimulation du télencéphale basal (neurones cholinergiques qui innervent le cortex visuel primaire). J'ai analysé les potentiels évoqués visuels (PEVs) dans le cortex visuel primaire des rats pendant 4 à 8 heures après le couplage. Afin de préciser l’action de l’ACh sur l’activité des PEVs dans V1, j’ai injecté individuellement l’antagoniste des récepteurs muscariniques, nicotiniques, α7 ou NMDA avant l’infusion de carbachol. La stimulation du système cholinergique jumelée avec une stimulation visuelle augmente l’amplitude des PEVs durant plus de 8h. Le blocage des récepteurs muscarinique, nicotinique et NMDA abolit complètement cette amélioration, tandis que l’inhibition des récepteurs α7 a induit une augmentation instantanée des PEVs. Ces résultats suggèrent que l'ACh facilite à long terme la réponse aux stimuli visuels et que cette facilitation implique les récepteurs nicotiniques, muscariniques et une interaction avec les récepteur NMDA dans le cortex visuel. Ces mécanismes sont semblables à la potentiation à long-terme, évènement physiologique lié à l’apprentissage. L’étape suivante était d’évaluer si l’effet de l’amplification cholinergique de l’entrée de l’information visuelle résultait non seulement en une modification de l’activité corticale mais aussi de la perception visuelle. J’ai donc mesuré l’amélioration de l’acuité visuelle de rats adultes éveillés exposés durant 10 minutes par jour pendant deux semaines à un stimulus visuel de type «réseau sinusoïdal» couplé à une stimulation électrique du télencéphale basal. L’acuité visuelle a été mesurée avant et après le couplage des stimulations visuelle et cholinergique à l’aide d’une tâche de discrimination visuelle. L’acuité visuelle du rat pour le stimulus d’entrainement a été augmentée après la période d’entrainement. L’augmentation de l’acuité visuelle n’a pas été observée lorsque la stimulation visuelle seule ou celle du télencéphale basal seul, ni lorsque les fibres cholinergiques ont été lésées avant la stimulation visuelle. Une augmentation à long terme de la réactivité corticale du cortex visuel primaire des neurones pyramidaux et des interneurones GABAergiques a été montrée par l’immunoréactivité au c-Fos. Ainsi, lorsque couplé à un entrainement visuel, le système cholinergique améliore les performances visuelles pour l’orientation et ce probablement par l’optimisation du processus d’attention et de plasticité corticale dans l’aire V1. Afin d’étudier les mécanismes pharmacologiques impliqués dans l’amélioration de la perception visuelle, j’ai comparé les PEVs avant et après le couplage de la stimulation visuelle/cholinergique en présence d’agonistes/antagonistes sélectifs. Les injections intracorticales des différents agents pharmacologiques pendant le couplage ont montré que les récepteurs nicotiniques et M1 muscariniques amplifient la réponse corticale tandis que les récepteurs M2 muscariniques inhibent les neurones GABAergiques induisant un effet excitateur. L’infusion d’antagoniste du GABA corrobore l’hypothèse que le système inhibiteur est essentiel pour induire la plasticité corticale. Ces résultats démontrent que l’entrainement visuel jumelé avec la stimulation cholinergique améliore la plasticité corticale et qu’elle est contrôlée par les récepteurs nicotinique et muscariniques M1 et M2. Mes résultats suggèrent que le système cholinergique est un système neuromodulateur qui peut améliorer la perception sensorielle lors d’un apprentissage perceptuel. Les mécanismes d’amélioration perceptuelle induits par l’acétylcholine sont liés aux processus d’attention, de potentialisation à long-terme et de modulation de la balance d’influx excitateur/inhibiteur. En particulier, le couplage de l’activité cholinergique avec une stimulation visuelle augmente le ratio de signal / bruit et ainsi la détection de cibles. L’augmentation de la concentration cholinergique corticale potentialise l’afférence thalamocorticale, ce qui facilite le traitement d’un nouveau stimulus et diminue la signalisation cortico-corticale minimisant ainsi la modulation latérale. Ceci est contrôlé par différents sous-types de récepteurs cholinergiques situés sur les neurones GABAergiques ou glutamatergiques des différentes couches corticales. La présente thèse montre qu’une stimulation électrique dans le télencéphale basal a un effet similaire à l’infusion d’agoniste cholinergique et qu’un couplage de stimulations visuelle et cholinergique induit la plasticité corticale. Ce jumelage répété de stimulations visuelle/cholinergique augmente la capacité de discrimination visuelle et améliore la perception. Cette amélioration est corrélée à une amplification de l’activité neuronale démontrée par immunocytochimie du c-Fos. L’immunocytochimie montre aussi une différence entre l’activité des neurones glutamatergiques et GABAergiques dans les différentes couches corticales. L’injection pharmacologique pendant la stimulation visuelle/cholinergique suggère que les récepteurs nicotiniques, muscariniques M1 peuvent amplifier la réponse excitatrice tandis que les récepteurs M2 contrôlent l’activation GABAergique. Ainsi, le système cholinergique activé au cours du processus visuel induit des mécanismes de plasticité corticale et peut ainsi améliorer la capacité perceptive. De meilleures connaissances sur ces actions ouvrent la possibilité d’accélérer la restauration des fonctions visuelles lors d’un déficit ou d’amplifier la fonction cognitive.

Molecular mechanisms underlying deficient neurogenesis in Alzheimer’s disease

La neurogenèse est présente, dans le cerveau adulte, dans la zone sous-ventriculaire (ZSV) encadrant les ventricules latéraux et dans le gyrus dentelé (GD) de l’hippocampe, permettant l’apprentissage, la mémoire et la fonction olfactive. Ces micro-environnements possèdent des signaux contrôlant l’auto-renouvellement des cellules souches neurales (CSN), leur prolifération, leur destin et leur différenciation. Or, lors du vieillissement, les capacités régénératives et homéostatiques et la neurogenèse déclinent. Les patients atteints de la maladie d’Alzheimer (MA), comme le modèle animal reproduisant cette maladie (3xTg-AD), montrent une accélération des phénotypes liés au vieillissement dont une diminution de la neurogenèse. Notre hypothèse est que la découverte des mécanismes affectant la neurogenèse, lors du vieillissement et de la MA, pourrait fournir de nouvelles cibles thérapeutiques pour prévenir le déclin cognitif. Les études sur l’âge d’apparition et les mécanismes altérant la neurogenèse dans la MA sont contrastées et nous ont guidé vers deux études. L’examen des changements dans les étapes de la neurogenèse lors du vieillissement et du développement de la neuropathologie. Nous avons étudié la ZSV, les bulbes olfactifs et le GD de souris femelles de 11 et 18 mois, et l’apparition des deux pathologies associées à la MA : les plaques amyloïdes et les enchevêtrements neurofibrillaires. Nous avons découvert que les souris 3xTg-AD possèdent moins de cellules en prolifération, de progéniteurs et de neuroblastes, induisant une diminution de l’intégration de nouvelles cellules dans le GD et les bulbes olfactifs. Notons que le taux de neurogenèse chez ces souris de 11 mois est similaire à celui des souris de phénotype sauvage de 18 mois, indiquant une accélération des changements liés au vieillissement dans la MA. Dans la ZSV, nous avons aussi démontré une accumulation de gouttelettes lipidiques, suggérant des changements dans l’organisation et le métabolisme de la niche. Enfin, nous avons démontré que le déficit de la neurogenèse apparait lors des premières étapes de la MA, avant l’apparition des plaques amyloïdes et des enchevêtrements neurofibrillaires. A l’examen des mécanismes inhibant la neurogenèse lors de la MA, nous voyons que chez des souris de 5 mois, le déficit de la neurogenèse dans la ZSV et le GD est corrélé avec l’accumulation de lipides, qui coïncide avec l’apparition du déclin cognitif. Nous avons aussi découvert que dans le cerveau humain de patients atteints de la MA et dans les 3xTg-AD, des gouttelettes lipidiques s’accumulaient dans les cellules épendymaires, représentant le principal soutien des CSN de la niche. Ces lipides sont des triglycérides enrichis en acide oléique qui proviennent de la niche et pas d’une défaillance du système périphérique. De plus, l’infusion locale d’acide oléique chez des souris de phénotype sauvage permet de reproduire l’accumulation de triglycérides dans les cellules épendymaires, comme dans la MA. Ces gouttelettes induisent un dérèglement de la voie de signalisation Akt-FoxO3 dans les CSN, menant à l’inhibition de leur activation in vitro et in vivo. Ces résultats permettent une meilleure compréhension de la régulation de la neurogenèse par le métabolisme lipidique. Nous avons démontré un nouveau mécanisme par lequel l’accumulation des lipides dans la ZSV induit une inhibition des capacités de prolifération et de régénération des CSN lors de la MA. Les travaux futurs permettront de comprendre comment et pourquoi le métabolisme lipidique du cerveau est altéré dans la MA, ce qui pourrait offrir de nouvelles voies thérapeutiques pour la prévention et la régénération.

Advances in scaling deep learning algorithms

Les algorithmes d'apprentissage profond forment un nouvel ensemble de méthodes puissantes pour l'apprentissage automatique. L'idée est de combiner des couches de facteurs latents en hierarchies. Cela requiert souvent un coût computationel plus elevé et augmente aussi le nombre de paramètres du modèle. Ainsi, l'utilisation de ces méthodes sur des problèmes à plus grande échelle demande de réduire leur coût et aussi d'améliorer leur régularisation et leur optimization. Cette thèse adresse cette question sur ces trois perspectives. Nous étudions tout d'abord le problème de réduire le coût de certains algorithmes profonds. Nous proposons deux méthodes pour entrainer des machines de Boltzmann restreintes et des auto-encodeurs débruitants sur des distributions sparses à haute dimension. Ceci est important pour l'application de ces algorithmes pour le traitement de langues naturelles. Ces deux méthodes (Dauphin et al., 2011; Dauphin and Bengio, 2013) utilisent l'échantillonage par importance pour échantilloner l'objectif de ces modèles. Nous observons que cela réduit significativement le temps d'entrainement. L'accéleration atteint 2 ordres de magnitude sur plusieurs bancs d'essai. Deuxièmement, nous introduisont un puissant régularisateur pour les méthodes profondes. Les résultats expérimentaux démontrent qu'un bon régularisateur est crucial pour obtenir de bonnes performances avec des gros réseaux (Hinton et al., 2012). Dans Rifai et al. (2011), nous proposons un nouveau régularisateur qui combine l'apprentissage non-supervisé et la propagation de tangente (Simard et al., 1992). Cette méthode exploite des principes géometriques et permit au moment de la publication d'atteindre des résultats à l'état de l'art. Finalement, nous considérons le problème d'optimiser des surfaces non-convexes à haute dimensionalité comme celle des réseaux de neurones. Tradionellement, l'abondance de minimum locaux était considéré comme la principale difficulté dans ces problèmes. Dans Dauphin et al. (2014a) nous argumentons à partir de résultats en statistique physique, de la théorie des matrices aléatoires, de la théorie des réseaux de neurones et à partir de résultats expérimentaux qu'une difficulté plus profonde provient de la prolifération de points-selle. Dans ce papier nous proposons aussi une nouvelle méthode pour l'optimisation non-convexe.

Learning Commonsense Categorical Knowledge in a Thread Memory System

If we are to understand how we can build machines capable of broad purpose learning and reasoning, we must first aim to build systems that can represent, acquire, and reason about the kinds of commonsense knowledge that we humans have about the world. This endeavor suggests steps such as identifying the kinds of knowledge people commonly have about the world, constructing suitable knowledge representations, and exploring the mechanisms that people use to make judgments about the everyday world. In this work, I contribute to these goals by proposing an architecture for a system that can learn commonsense knowledge about the properties and behavior of objects in the world. The architecture described here augments previous machine learning systems in four ways: (1) it relies on a seven dimensional notion of context, built from information recently given to the system, to learn and reason about objects' properties; (2) it has multiple methods that it can use to reason about objects, so that when one method fails, it can fall back on others; (3) it illustrates the usefulness of reasoning about objects by thinking about their similarity to other, better known objects, and by inferring properties of objects from the categories that they belong to; and (4) it represents an attempt to build an autonomous learner and reasoner, that sets its own goals for learning about the world and deduces new facts by reflecting on its acquired knowledge. This thesis describes this architecture, as well as a first implementation, that can learn from sentences such as ``A blue bird flew to the tree'' and ``The small bird flew to the cage'' that birds can fly. One of the main contributions of this work lies in suggesting a further set of salient ideas about how we can build broader purpose commonsense artificial learners and reasoners.

Learning Linear, Sparse, Factorial Codes

In previous work (Olshausen & Field 1996), an algorithm was described for learning linear sparse codes which, when trained on natural images, produces a set of basis functions that are spatially localized, oriented, and bandpass (i.e., wavelet-like). This note shows how the algorithm may be interpreted within a maximum-likelihood framework. Several useful insights emerge from this connection: it makes explicit the relation to statistical independence (i.e., factorial coding), it shows a formal relationship to the algorithm of Bell and Sejnowski (1995), and it suggests how to adapt parameters that were previously fixed.

Observations on Cortical Mechanisms for Object Recognition andsLearning

This paper sketches a hypothetical cortical architecture for visual 3D object recognition based on a recent computational model. The view-centered scheme relies on modules for learning from examples, such as Hyperbf-like networks. Such models capture a class of explanations we call Memory-Based Models (MBM) that contains sparse population coding, memory-based recognition, and codebooks of prototypes. Unlike the sigmoidal units of some artificial neural networks, the units of MBMs are consistent with the description of cortical neurons. We describe how an example of MBM may be realized in terms of cortical circuitry and biophysical mechanisms, consistent with psychophysical and physiological data.