La présente étude avait pour but d’explorer les modulations fonctionnelles putaminales du signal de spectroscopie par résonance magnétique (SRM) combiné du glutamate et de la glutamine (Glx), ainsi que de l’acide γ-aminobutyrique (GABA) en lien avec l’apprentissage d’une séquence motrice. Nous avons émis l’hypothèse que les concentrations de Glx seraient spécifiquement augmentées pendant et après la pratique d’une telle tâche, et ce comparativement à une condition d’exécution motrice simple conçue pour minimiser l’apprentissage. La tâche d’appuis séquentiels des doigts (« finger taping task ») utilisée est connue pour induire un apprentissage moteur évoluant en phases, avec une progression initialement rapide lors de la première session d’entraînement (phase rapide), puis lente lors de sessions subséquentes (phase lente). Cet apprentissage est également conçu comme dépendant de processus « on-line » (pendant la pratique) d’acquisition et « off-line » (entre les périodes de pratique) de consolidation de la trace mnésique de l’habilité motrice. Une grande quantité de données impliquent le système de neurotransmission glutamatergique, principalement par l’action de ses récepteurs N-Méthyl-D-aspartate (NMDAR) et métabotropiques (mGluR), dans une multitude de domaine de la mémoire. Quelques-unes de ces études suggèrent que cette relation s’applique aussi à des mémoires de type motrice ou dépendante du striatum. De plus, certains travaux chez l’animal montrent qu’une hausse des concentrations de glutamate et de glutamine peut être associée à l’acquisition et/ou consolidation d’une trace mnésique. Nos mesures de SRM à 3.0 Tesla, dont la qualité ne s’est avérée satisfaisante que pour le Glx, démontrent qu’une telle modulation des concentrations de Glx est effectivement détectable dans le putamen après la performance d’une tâche motrice. Elles ne nous permettent toutefois pas de dissocier cet effet putativement attribuable à la plasticité du putamen associée à l’apprentissage moteur de séquence, de celui de la simple activation neuronale causée par l’exécution motrice. L’interprétation de l’interaction non significative, montrant une plus grande modulation par la tâche motrice simple, mène cependant à l’hypothèse alternative que la plasticité glutamatergique détectée est potentiellement plus spécifique à la phase lente de l’apprentissage, suggérant qu’une seconde expérience ainsi orientée et utilisant une méthode de SRM plus sensible au Glx aurait donc de meilleures chances d’offrir des résultats concluants. / The present study explored motor learning-related functional changes in putaminal combined glutamate and glutamine (Glx) and γ-Aminobutyric acid (GABA) magnetic resonance spectroscopy (MRS) signal. It was hypothesized that Glx concentrations would specifically increase during and after learning of a sequential finger tapping task (sFTT), as compared to execution of a simple motor task designed to elicit minimal learning. Learning of sFTT is known to evolve in an initial fast progressing stage during the first practice session (fast learning stage), followed by a slower progression during later sessions (slow learning stage). It is also thought to depend on both on-line (during practice sessions) acquisition and off-line (between practice sessions) consolidation processes to create, transform and assure retention of a motor skill memory trace. A body of data implicates glutamatergic neurotransmission, especially through its N-Methyl-D-aspartate (NMDAR) and metabotropic (mGluR) receptors, in many memory systems, some of which apply to motor learning and striatal-dependant learning. Moreover, some animal studies suggest that Glx concentrations can be upregulated in relation to memory acquisition and/or consolidation. Our MRS acquisitions, of which the quality happened to be sufficient only for Glx quantification, allowed the detection of an augmentation in putaminal Glx occurring after motor task execution. However, our data could not ascribe this modulation specifically to motor learning related plastic changes, at the exclusion of simple neural activation related to motor execution. Nevertheless, the interpretation of the non-significant interaction, showing a larger Glx change in response to the simple motor task compared to sFTT, leads to the possibility that the detected glutamatergic plasticity may be specifically associated to the slow learning phase. We therefore suggest that testing this alternate hypothesis in a second experiment, using an MRS technique with more sensibility to Glx could yield more convincing results.

Cortical and subcortical microcircuits are continuously modified throughout life. Despite ongoing changes these networks stubbornly maintain their functions, which persist although destabilizing synaptic and nonsynaptic mechanisms should ostensibly propel them towards runaway excitation or quiescence. What dynamical phenomena exist to act together to balance such learning with information processing? What types of activity patterns do they underpin, and how do these patterns relate to our perceptual experiences? What enables learning and memory operations to occur despite such massive and constant neural reorganization? Progress towards answering many of these questions can be pursued through large-scale neuronal simulations.    In this thesis, a Hebbian learning rule for spiking neurons inspired by statistical inference is introduced. The spike-based version of the Bayesian Confidence Propagation Neural Network (BCPNN) learning rule involves changes in both synaptic strengths and intrinsic neuronal currents. The model is motivated by molecular cascades whose functional outcomes are mapped onto biological mechanisms such as Hebbian and homeostatic plasticity, neuromodulation, and intrinsic excitability. Temporally interacting memory traces enable spike-timing dependence, a stable learning regime that remains competitive, postsynaptic activity regulation, spike-based reinforcement learning and intrinsic graded persistent firing levels.    The thesis seeks to demonstrate how multiple interacting plasticity mechanisms can coordinate reinforcement, auto- and hetero-associative learning within large-scale, spiking, plastic neuronal networks. Spiking neural networks can represent information in the form of probability distributions, and a biophysical realization of Bayesian computation can help reconcile disparate experimental observations. / <p>QC 20170421</p>

Supplying the right amount of taxis in the right place at the right time is very important for taxi companies. In this paper, the machine learning model Taxi Demand Net (TDNet) is presented which predicts short-term taxi demand in different zones of a city. It is based on WaveNet which is a causal dilated convolutional neural net for time-series generation. TDNet uses historical demand from the last years and transforms features such as time of day, day of week and day of month into 26-hour taxi demand forecasts for all zones in a city. It has been applied to one city in northern Europe and one in South America. In northern europe, an error of one taxi or less per hour per zone was achieved in 64% of the cases, in South America the number was 40%. In both cities, it beat the SARIMA and stacked ensemble benchmarks. This performance has been achieved by tuning the hyperparameters with a Bayesian optimization algorithm. Additionally, weather and holiday features were added as input features in the northern European city and they did not improve the accuracy of TDNet.

Korrelationskodierung im Hippokampus bildet möglicherweise die neuronale Basis für episodisches Gedächtnis. In dieser Arbeit untersuchen wir zwei Phänomene der Korrelationskodierung: Phasenpräzession und Sequenzwiederholungen. Phasenpräzession bezeichnet die Abnahme der Phase des Aktionspotentials einer Ortszelle relativ zur Theta-Oszillation. Sequenzwiederholung beschreibt die Aktivität von Ortszellen in Ruhephasen; dabei werden vorangegangene Orts- Sequenzen in umgekehrter Reihenfolge wiederholt. Wir untersuchen Phasenpräzession in einzelnen Versuchsdurchläufen. In bisherigen Studien wurden Daten zur Phasenpräzession in vielen Versuchsdurchläufen zusammengelegt. Wir zeigen, dass dies zu einer verzerrten Schätzung von grundlegenden Eigenschaften der Phasenpräzession führen kann. Weiterhin demonstrieren wir eine starke Variabilität der Phasenpräzession zwischen verschiedenen Versuchsdurchläufen. Daher ist Phasenpräzession besser geeignet zeitlich strukturierte Sequenzen zu lernen, als man aufgrund der zusammengelegten Daten vermutet hatte. Desweiteren untersuchen wir die Beziehung von Phasenpräzession in unterschiedlichen Teilen des Hippokampus. Wir zeigen, dass die extrazellulären Theta- Oszillationen in CA3 und CA1 außer Phase sind. Dennoch geschieht Phasenpräzession in beiden Regionen fast gleichzeitig, und CA3 Zellen feuern oft kurz vor CA1 Zellen. Diese zeitliche Beziehung ist im Einklang mit einer Vererbung von Phasenpräzession von CA3 nach CA1. Wir entwickeln ein mechanistisches Modell für Sequenzwiederholungen in umgekehrter Reihenfolge basierend auf Kurzzeitfazilitierung. Mit Hilfe des Tempotrons beweisen wir, dass die entstehenden zeitlichen Muster geeignet sind, um von nachgeschalteten Strukturen ausgelesen zu werden. Das Modell sagt voraus, dass im Gyrus Dentatus synchrone Zellaktivität kurz vor einer Sequenzwiederholung in CA3 zu sehen ist, und es zeigt, dass Sequenzwiederholungen zum Lernen von zeitlichen Mustern genutzt werden können. / Hippocampal correlation coding is a putative neural mechanism underlying episodic memory. Here, we look at two related phenomena: phase precession and reverse replay of sequences. Phase precession refers to the decrease of the firing phase of a place cell with respect to the local theta rhythm during the crossing of the place field. Reverse replay refers to reactivation of previously experienced place field sequences in reverse order during awake resting periods. First, we study properties of phase precession in single trials. Usually, phase precession is studied on the basis of data in which many place field traversals are pooled together. We find that single-trial and pooled-trial phase precession are different with respect to phase-position correlation, phase-time correlation, and phase range. We demonstrate that phase precession exhibits a large trial-to-trial variability and that pooling trials changes basic measures of phase precession. These findings indicate that single trials may be better suited for encoding temporally structured events than is suggested by the pooled data. Second, we examine the coordination of phase precession among subregions of the hippocampus. We find that the local theta rhythms in CA3 and CA1 are almost antiphasic. Still, phase precession in the two regions occurs with only a small phase shift, and CA3 cells tend to fire a few milliseconds before CA1 cells. These results suggest that phase precession in CA1 might be inherited from CA3. Finally, we present a model of reverse replay based on short-term facilitation. The model compresses temporal patterns from a behavioral time scale of seconds to shorter time scales relevant for synaptic plasticity. We demonstrate that the compressed patterns can be learned by the tempotron learning rule. The model provides testable predictions (synchronous activation of dentate gyrus during sharp wave-ripples) and functional interpretations of hippocampal activity (temporal pattern learning).

Prever eventos ambientais, com base em memórias sobre regularidades passadas, é uma das funções fundamentais de sistemas nervosos complexos. Eventos ordenados serialmente ou sequências estruturadas de estímulos permitem extrair informação passível de descrição formal que define seu padrão serial. Esse padrão inclui informações temporais e espaciais que facultam prever os próximos eventos da sequência, possibilitando a preparação prévia do organismo para lidar com sua ocorrência. Não surpreende que animais, incluindo o ser humano, aprendam, de maneira relativamente rápida, sobre regras e estruturas de padrões sequenciais de estímulos. O uso de tarefas de tempo de reação serial (TRS) é recorrente em estudos envolvendo a formação de associações, antecipação, atenção, as bases da memória e aprendizagem de relações complexas. Resumidamente, voluntários devem responder a estímulos apresentados em sequências repetitivas ou aleatórias. Com o treino, há redução no tempo de reação a cada estímulo, refletindo a aprendizagem de relações percepto-motoras. Essa redução, porém, é maior na sequência repetitiva em relação à sequência aleatória, indicando um aprendizado também sobre a sequência repetitiva, mesmo quando o voluntário não a percebe (conscientemente) e seja incapaz de relatar sua existência. Trata-se, portanto, de uma aquisição (inicialmente) implícita. A complexidade de uma sequência de estímulos pode ser expressa quantitativamente por meio de uma ferramenta matemática proposta por Shannon (1948), a entropia informacional (EI), que considera, entre outras coisas, a probabilidade de ocorrência dos estímulos em diferentes níveis. No presente trabalho, avaliamos em que extensão o conceito de EI permite prever o desempenho de ratos na tarefa de TRS envolvendo sequências com diferentes níveis de complexidade. Ratos foram treinados a reagir (1) a uma sequência repetitiva de estímulos, cuja quantidade de EI no nível 1 (que relaciona os estímulos da sequência 2 a 2) era 2,75. Após atingirem um nível assintótico de desempenho, os animais foram expostos (2) a sequências variáveis de estímulos com a mesma quantidade de EI no nível 1, porém, com maior quantidade de EI no nível 2 (que relaciona os estímulos da sequência 3 a 3). Numa etapa posterior os animais foram expostos (3) a uma nova sequência repetitiva de estímulos, cuja quantidade de EI no nível 1 era 3,00; por fim, os animais foram submetidos (4) a sequências variáveis com a mesma quantidade de EI no nível 1em relação à sequência anterior, porém, com maior quantidade de EI no nível 2 . Os resultados mostraram que os ratos aprenderam sobre os padrões seriais e, mais interessante, que seu desempenho esteve fortemente correlacionado à quantidade de EI no nível 2. Em outras palavras, quanto maior a EI, pior o desempenho dos animais tanto em termos do tempo de reação como em termos da percentagem de respostas corretas. Portanto, o conceito de EI permite não apenas quantificar a complexidade de sequências empregadas em estudos envolvendo aprendizagem serial, mas também prever o desempenho dos animais. / Prediction of environmental events, relying on memories of past regularities, is one of the fundamental functions of complex nervous systems. Sequences of serially ordered stimuli allow extracting information that defines its serial pattern. These patterns allow prediction of the next item in a sequence of events, facultating previous preparation to deal with its occurrence. Not surprisingly, animals, including humans, can identify rules present in serial structures of stimuli. Serial reaction time tasks (SRTT) have been extensively used in studies involving association, anticipation, attention, and learning and memory. Typically, subjects have to react to stimuli presented either in random or in repetitive sequences. As training proceeds, reaction time to each stimulus decreases, reflecting acquisition of this perceptual-motor skill. However, reaction time reduction is greater for repetitive sequences relative to the random sequences, indicating acquisition about the repetitive structure of the sequence. In human beings, this may occur even when the subject in uncapable of reporting the existence of a sequence, indicating that the acquisition was (at least initially) implicit rather than explicit. The complexity of a sequence of stimuli, at different levels, may be quantifyed by means of a mathematical tool proposed by Shannon (1948), the information entropy (IE). In this study we evaluated to which extent IE can predict performance of rats in SRTT involving sequences of stimuli organized at different levels of complexity. Rats were trained to react (1) a repeated sequence of stimuli which IE at the level \"1\" (i.e., expressing to which extent a given item allow prediction of the next) was 2.75. After reaching an asymptotic level of performance, the animals were exposed (2) a variable sequence of stimuli with the same amount of IE in the level \"1\", but with more IE in the level \"2\" (i.e., expressing to which extent two given items allow prediction of the next). Later the animals were exposed to (3) a new repeated sequence of stimuli, which IE at the level \"1\" was 3.00. Finally, the animals were submitted to (4) a random sequence of stimuli with the same amount of IE at the level \"1\", i.e., 3.00, but with greater IE in level 2. Results showed that rats learned about the serial patterns and, more interestingly, their performance strongly correlated to the amount of IE at the level \"2 \", both in terms of reaction times and in terms of percentage of correct responses. Therefore, IE allows not only to quantify complexity of sequences in studies involving serial learning, but also to predict performance of the subjects.

La thèse propose une approche de l'apprentissage temporel par des mécanismes d'auto-organisation à grain fin. Le manuscrit situe dans un premier temps le travail dans la perspective de contribuer à promouvoir une informatique cellulaire. Il s'agit d'une informatique où les calculs se répartissent en un grand nombre de calculs élémentaires, exécutés en parallèle, échangeant de l'information entre eux. Le caractère cellulaire tient à ce qu'en plus d’être à grain fin, une telle architecture assure que les connexions entre calculateurs respectent une topologie spatiale, en accord avec les contraintes des évolutions technologiques futures des matériels. Dans le manuscrit, la plupart des architectures informatiques distribuées sont examinées suivant cette perspective, pour conclure que peu d'entre elles relèvent strictement du paradigme cellulaire.Nous nous sommes intéressé à la capacité d'apprentissage de ces architectures, du fait de l'importance de cette notion dans le domaine connexe des réseaux de neurones par exemple, sans oublier toutefois que les systèmes cellulaires sont par construction des systèmes complexes dynamiques. Cette composante dynamique incontournable a motivé notre focalisation sur l'apprentissage temporel, dont nous avons passé en revue les déclinaisons dans les domaines des réseaux de neurones supervisés et des cartes auto-organisatrices.Nous avons finalement proposé une architecture qui contribue à la promotion du calcul cellulaire en ce sens qu'elle exhibe des propriétés d'auto-organisation pour l'extraction de la représentation des états du système dynamique qui lui fournit ses entrées, même si ces dernières sont ambiguës et ne reflètent que partiellement cet état. Du fait de la présence d'un cluster pour nos simulations, nous avons pu mettre en œuvre une architecture complexe, et voir émerger des phénomènes nouveaux. Sur la base de ces résultats, nous développons une critique qui ouvre des perspectives sur la suite à donner à nos travaux. / The thesis proposes a sequence learning approach that uses the mechanism of fine grain self-organization. The manuscript initially starts by situating this effort in the perspective of contributing to the promotion of cellular computing paradigm in computer science. Computation within this paradigm is divided into a large number of elementary calculations carried out in parallel by computing cells, with information exchange between them.In addition to their fine grain nature, the cellular nature of such architectures lies in the spatial topology of the connections between cells that complies with to the constraints of the technological evolution of hardware in the future. In the manuscript, most of the distributed architecture known in computer science are examined following this perspective, to find that very few of them fall within the cellular paradigm.We are interested in the learning capacity of these architectures, because of the importance of this notion in the related domain of neural networks for example, without forgetting, however, that cellular systems are complex dynamical systems by construction.This inevitable dynamical component has motivated our focus on the learning of temporal sequences, for which we reviewed the different models in the domains of neural networks and self-organization maps.At the end, we proposed an architecture that contributes to the promotion of cellular computing in the sense that it exhibits self-organization properties employed in the extraction of a representation of a dynamical system states that provides the architecture with its entries, even if the latter are ambiguous such that they partially reflect the system state. We profited from an existing supercomputer to simulate complex architecture, that indeed exhibited a new emergent behavior. Based on these results we pursued a critical study that sets the perspective for future work.

Denna avhandling i datalogi föreslår modeller för hur vissa beräkningsmässiga uppgifter kan utföras av hjärnbarken. Utgångspunkten är dels kända fakta om hur en area i hjärnbarken är uppbyggd och fungerar, dels etablerade modellklasser inom beräkningsneurobiologi, såsom attraktorminnen och system för gles kodning. Ett neuralt nätverk som producerar en effektiv gles kod i binär mening för sensoriska, särskilt visuella, intryck presenteras. Jag visar att detta nätverk, när det har tränats med naturliga bilder, reproducerar vissa egenskaper (receptiva fält) hos nervceller i lager IV i den primära synbarken och att de koder som det producerar är lämpliga för lagring i associativa minnesmodeller. Vidare visar jag hur ett enkelt autoassociativt minne kan modifieras till att fungera som ett generellt sekvenslärande system genom att utrustas med synapsdynamik. Jag undersöker hur ett abstrakt attraktorminnessystem kan implementeras i en detaljerad modell baserad på data om hjärnbarken. Denna modell kan sedan analyseras med verktyg som simulerar experiment som kan utföras på en riktig hjärnbark. Hypotesen att hjärnbarken till avsevärd del fungerar som ett attraktorminne undersöks och visar sig leda till prediktioner för dess kopplingsstruktur. Jag diskuterar också metodologiska aspekter på beräkningsneurobiologin idag. / In this thesis I take a modular approach to cortical function. I investigate how the cerebral cortex may realise a number of basic computational tasks, within the framework of its generic architecture. I present novel mechanisms for certain assumed computational capabilities of the cerebral cortex, building on the established notions of attractor memory and sparse coding. A sparse binary coding network for generating efficient representations of sensory input is presented. It is demonstrated that this network model well reproduces the simple cell receptive field shapes seen in the primary visual cortex and that its representations are efficient with respect to storage in associative memory. I show how an autoassociative memory, augmented with dynamical synapses, can function as a general sequence learning network. I demonstrate how an abstract attractor memory system may be realised on the microcircuit level -- and how it may be analysed using tools similar to those used experimentally. I outline some predictions from the hypothesis that the macroscopic connectivity of the cortex is optimised for attractor memory function. I also discuss methodological aspects of modelling in computational neuroscience. / QC 20100916

Using Hebb’s (1961) paradigm, it has been shown that older adults (OAs) fail to learn recurrent visuospatial supra-span sequence information (Turcotte, Gagnon, & Poirier, 2005); a deficit which has not been demonstrated on verbal versions of the same task or in younger adults (YAs). Since the Hebb paradigm is thought to rely on working memory and thus attention (Conway & Engle, 1996), one interpretation concerns an OA’s capacity to allocate the necessary attentional resources to carry out the various components of the task. Five studies investigated this proposal. The first three (Article 1) examined attention in a general manner by reducing the amount of attentional resources that a YA could devote to carrying out the visuospatial Hebb supra-span sequence learning task through the implementation of a verbal dual task (DT) procedure. The fourth (Article 2) further investigated the role of attention by using a DT induced at retrieval that overlapped extensively with the requirements (spatial and response features) of the visuospatial Hebb task. The final study (Article 3) aimed to use our previous findings to demonstrate learning among OAs in a visuospatial Hebb learning paradigm in which the motor response was replaced by a verbal response. Our findings confirm that attentional resources employed at the retrieval phase of the task appear to be particularly important for the demonstration of visuospatial sequence learning. The inclusion of a spatial and motor based DT at recall eliminated learning of the repeated sequence in YAs. Interestingly, the learning deficit of OAs was partially eliminated when the motor and spatial requirements at retrieval were reduced. Our findings offer strong support to the contention that supra-span learning of the Hebb type is not altered by the effect of age. However, learning deficits can be observed among OAs when the retrieval component of the task overly taxes attention-related processes. In the case of the visuospatial sequences, the basis of the deficit likely concerns an individual’s capacity to discriminate between responses made to previously presented sequences versus those that need to be made in reaction to the just seen sequence.

Using Hebb’s (1961) paradigm, it has been shown that older adults (OAs) fail to learn recurrent visuospatial supra-span sequence information (Turcotte, Gagnon, & Poirier, 2005); a deficit which has not been demonstrated on verbal versions of the same task or in younger adults (YAs). Since the Hebb paradigm is thought to rely on working memory and thus attention (Conway & Engle, 1996), one interpretation concerns an OA’s capacity to allocate the necessary attentional resources to carry out the various components of the task. Five studies investigated this proposal. The first three (Article 1) examined attention in a general manner by reducing the amount of attentional resources that a YA could devote to carrying out the visuospatial Hebb supra-span sequence learning task through the implementation of a verbal dual task (DT) procedure. The fourth (Article 2) further investigated the role of attention by using a DT induced at retrieval that overlapped extensively with the requirements (spatial and response features) of the visuospatial Hebb task. The final study (Article 3) aimed to use our previous findings to demonstrate learning among OAs in a visuospatial Hebb learning paradigm in which the motor response was replaced by a verbal response. Our findings confirm that attentional resources employed at the retrieval phase of the task appear to be particularly important for the demonstration of visuospatial sequence learning. The inclusion of a spatial and motor based DT at recall eliminated learning of the repeated sequence in YAs. Interestingly, the learning deficit of OAs was partially eliminated when the motor and spatial requirements at retrieval were reduced. Our findings offer strong support to the contention that supra-span learning of the Hebb type is not altered by the effect of age. However, learning deficits can be observed among OAs when the retrieval component of the task overly taxes attention-related processes. In the case of the visuospatial sequences, the basis of the deficit likely concerns an individual’s capacity to discriminate between responses made to previously presented sequences versus those that need to be made in reaction to the just seen sequence.

The anomaly detection task is to recognize the presence of an unusual (and potentially hazardous) state within the behaviors or activities of a computer user, system, or network with respect to some model of normal behavior which may be either hard-coded or learned from observation. An anomaly detection agent faces many learning problems including learning from streams of temporal data, learning from instances of a single class, and adaptation to a dynamically changing concept. The domain is complicated by considerations of the trusted insider problem (recognizing the difference between innocuous and malicious behavior changes on the part of a trusted user). This study introduces the anomaly detection in web applications and formulates it as a machine learning task on temporal sequence data. In this study the goal is to develop a model or profile of normal working state of web application user and to detect anomalous conditions as deviations from the expected behavior patterns. We focus, here, on learning models of normality at the user behavioral level, as observed through a web application. In this study we introduce some sensors intended to function as a focus of attention unit at the lowest level of a classification hierarchy using Finite State Markov Chains and Hidden Markov Models and discuss the success of these sensors.