Dissertations / Theses on the topic 'Temporary Network Structure'

Paper I: The power of star ideators: Does star ideator alone drive the success of ideas? The virtues and limits of star ideator presence in groups. Paper II: (with Magnusson, M.) The effect of diversity and group familiarity on performance in ideation groups. Paper III: (with Björk, J.) Temporary Network Structure and Group ideation performance - the effect of centrality and structural holes.

Contemporary urban life and functioning have become increasingly dependent on mobility. Having become an inherent constituent of urban dynamics, the role of urban moblity in influencing urban processes and morphology has increased dramat- ically. However, the relationship between urban mobility and spatial socio-economic structure has still not been thoroughly understood. This work will attempt to take a complex network theoretical approach to studying this intricate relationship through • the spatio-temporal evolution of ad-hoc developed network centralities based on the Google PageRank, • multilayer network regression with statistical random graphs respecting net- work structures for explaining urban mobility flows from urban socio-economic attributes, • and Graph Neural Networks for predicting mobility flows to or from a specific location in the city. Making both practical and theoretical contributions to urban science by offering methods for describing, monitoring, explaining, and predicting urban dynamics, this work will thus be aimed at providing a network theoretical framework for developing tools to facilitate better decision-making in urban planning and policy making.

Statistical models for data collected over space are widely available and commonly used. These models however usually assume relationships between observations depend on Euclidean distance between monitoring sites whose location is determined using two dimensional coordinates, and that relationships are not direction dependent. One example where these assumptions fail is when data are collected on river networks. In this situation, the location of monitoring sites along a river network relative to other sites is as important as the location in two dimensional space since it can be expected that spatial patterns will depend on the direction of water flow and distance between monitoring sites measured along the river network. Euclidean distance therefore might no longer be the most appropriate distance metric to consider. This is further complicated where it might be necessary to consider both Euclidean distance and distance along the river network if the observed variable is influenced by the land in which the river network is embedded. The Environment Agency (EA), established in 1996, is the government agency responsible for monitoring and improving the water quality in rivers situated in England (and Wales until 2013). A key responsibility of the EA is to ensure that efforts are made to improve and maintain water quality standards in compliance with EU regulations such as the Water Framework Directive (WFD, European Parliament (2000)) and Nitrates Directive (European Parliament, 1991). Environmental monitoring is costly and in many regions of the world funding for environmental monitoring is decreasing (Ferreyra et al., 2002). It is therefore important to develop statistical methods that can extract as much information as possible from existing or reduced monitoring networks. One way to do this is to identify common temporal patterns shared by many monitoring sites so that redundancy in the monitoring network could be reduced by removing non-informative sites exhibiting the same temporal patterns. In the case of river water quality, information about the shape of the river network, such as flow direction and connectivity of monitoring sites, could be incorporated into statistical techniques to improve statistical power and provide efficient inference without the increased cost of collecting more data. Reducing the volume of data required to estimate temporal trends would improve efficiency and provide cost savings to regulatory agencies. The overall aim of this thesis is to investigate how information about the spatial structure of river networks can be used to augment and improve the specfic trends obtained when using a variety of statistical techniques to estimate temporal trends in water quality data. Novel studies are designed to investigate the effect of accounting for river network structure within existing statistical techniques and, where necessary, statistical methodology is developed to show how this might be achieved. Chapter 1 provides an introduction to water quality monitoring and a description of several statistical methods that might be used for this. A discussion of statistical problems commonly encountered when modelling spatiotemporal data is also included. Following this, Chapter 2 applies a dimension reduction technique to investigate temporal trends and seasonal patterns shared among catchment areas in England and Wales. A novel comparison method is also developed to identify differences in the shape of temporal trends and seasonal patterns estimated using several different statistical methods, each of which incorporate spatial information in different ways. None of the statistical methods compared in Chapter 2 specifically account for features of spatial structure found in river networks: direction of water flow, relative influence of upstream monitoring sites on downstream sites, and stream distance. Chapter 3 therefore provides a detailed investigation and comparison of spatial covariance models that can be used to model spatial relationships found in river networks to standard spatial covariance models. Further investigation of the spatial covariance function is presented in Chapter 4 where a simulation study is used to assess how predictions from statistical models based on river network spatial covariance functions are affected by reducing the size of the monitoring network. A study is also developed to compare the predictive performance of statistical models based on a river network spatial covariance function to models based on spatial covariate information, but assuming spatial independence of monitoring sites. Chapters 3 and 4 therefore address the aim of assessing the improvement in information extracted from statistical models after the inclusion of information about river network structure. Following this, Chapter 5 combines the ideas of Chapters 2, 3 and 4 and proposes a novel statistical method where estimated common temporal patterns are adjusted for known spatial structure, identified in Chapters 3 and 4. Adjusting for known structure in the data means that spatial and temporal patterns independent of the river network structure can be more clearly identified since they are no longer confounded with known structure. The final chapter of this thesis provides a summary of the statistical methods investigated and developed within this thesis, identifies some limitations of the work carried out and suggests opportunities for future research. An Appendix provides details of many of the data processing steps required to obtain information about the river network structure in an appropriate form.

Many real-world applications in the social, biological, and physical sciences involve large systems of entities that interact together in some way. The number of components in these systems can be extremely large, so some simplification is typically needed for tractable analysis. A common representation of interacting entities is a network. In its simplest form, a network consists of a set of nodes that represent entities and a set of edges between pairs of nodes that represent interactions between those entities. In this thesis, we investigate clustering techniques for time-dependent networks. An important mesoscale feature in networks is communities. Most community-detection methods are designed for time-independent networks. A recent framework for representing temporal networks is multilayer networks. In this thesis, we focus primarily on community detection in temporal networks represented as multilayer networks. We investigate three main topics: a community-detection method known as multilayer modularity maximization, the development of a benchmark for community detection in temporal networks, and the application of multilayer modularity maximization to temporal financial asset-correlation networks. We first investigate theoretical and computational issues in multilayer modularity maximization. We introduce a diagnostic to measure persistence of community structure in a multilayer network partition and we show how communities one obtains with multilayer modularity maximization reflect a trade-off between time-independent community structure within layers and temporal persistence between layers. We discuss computational issues that can arise when solving this method in practice and we suggest ways to mitigate them. We then propose a benchmark for community detection in temporal networks and carry out various numerical experiments to compare the performance of different methods and computational heuristics on our benchmark. We end with an application of multilayer modularity maximization to temporal financial correlation networks.

Language acquisition is one of the core problems in artificial intelligence (AI) and it is generally accepted that any successful AI account of the mind will stand or fall depending on its ability to model human language. Simple Recurrent Networks (SRNs) are a class of so-called artificial neural networks that have a long history in language modelling via learning to predict the next word in a sentence. However, SRNs have also been shown to suffer from catastrophic forgetting, lack of syntactic systematicity and an inability to represent more than three levels of centre-embedding, due to the so-called 'vanishing gradients' problem. This problem is caused by the decay of past input information encoded within the error-gradients which vanish exponentially as additional input information is encountered and passed through the recurrent connections. That said, a number of architectural variations have been applied which may compensate for this issue, such as the Nonlinear Autoregressive Network with exogenous inputs (NARX) network and the multi-recurrent network (MRN). In addition to this, Echo State Networks (ESNs) are a relatively new class of recurrent neural network that do not suffer from the vanishing gradients problem and have been shown to exhibit state-of-the-art performance in tasks such as motor control, dynamic time series prediction, and more recently language processing. This research re-explores the class of SRNs and evaluates them against the state-of-the-art ESN to identify which model class is best able to induce the underlying finite-state automaton of the target grammar implicitly through the next word prediction task. In order to meet its aim, the research analyses the internal representations formed by each of the different models and explores the conditions under which they are able to carry information about long term sequential dependencies beyond what is found in the training data. The findings of the research are significant. It reveals that the traditional class of SRNs, trained with backpropagation through time, are superior to ESNs for the grammar prediction task. More specifically, the MRN, with its state-based memory of varying rigidity, is more able to learn the underlying grammar than any other model. An analysis of the MRN’s internal state reveals that this is due to its ability to maintain a constant variance within its state-based representation of the embedded aspects (or finite state machines) of the target grammar. The investigations show that in order to successfully induce complex context free grammars directly from sentence examples, then not only are a hidden layer and output layer recurrency required, but so is self-recurrency on the context layer to enable varying degrees of current and past state information, that are integrated over time.

Ecological network theory has developed from studies of static, binary trophic relationships to the analysis of quantitative, dynamic communities consisting of multiple link-types. Particularly, work has focused on the dynamic nature of ecological networks, which maintains stability in complex communities. However, there are few in situ network-level studies of the determinants of temporal and spatial variation in community structure. This thesis utilises data from a 10-year study of a host-parasitoid network and a collaborative study in an applied ecological setting to identify individual level factors important to network structure. The work aims towards an empirical, predictive framework linking adaptive foraging behaviour to ecological network structure. The results show that condition-dependent foraging behaviours structure host-parasitoid networks. The realised niches of the studied parasitoid species were generally biased towards larger host species and condition-dependent sex ratio allocation increased the likelihood that females would eclose from relatively larger hosts and males from relatively smaller hosts, which resulted in sex ratios deviating from Hamiltonian (50:50) predictions. Further, both of these aspects of behaviour are plastic, where parasitoid behaviour responded to environmental heterogeneity. Particularly, host preference behaviour conformed to an egg-/time-limitation framework, where the size dependency of the behaviour is greater when individuals have a greater likelihood of being egg-limited. Both the size-dependency and the plasticity of these behaviours differed significantly between secondary parasitoid species. This species identity effect interacted with landscape heterogeneity, which may explain some inter- and intra-specific variation in network structure. With respect to applied ecology, the results show that the benefits of natural vegetation for pest control are dependent upon the dispersal capabilities and the diet breadth of the pest and its natural enemies. The findings are evaluated towards a predictive framework for understanding the effects of future climate change on community structure and stability. We consider this framework in terms of applied ecology, particularly pest control ecosystem services provided by natural vegetation in an agricultural environment. The synergistic nature of the multiple determinants of network structure found in this thesis suggest that future studies should focus on the whole network, which is not necessarily the sum of its parts.

There has been significant interest in studying social interactions in online social networks, such as how people exchange opinions, disseminate information, and adopt certain behaviours. One phenomenon addressed is information diffusion: the way information is spread in social networks. Since their emergence, online social networks have been used by people to create and share content. They provide a set of functionalities that facilitate these and other tasks, allowing users to interact with each other. For researchers, these platforms became the basis for understanding complex human behaviours, one of which is the ‘urge’ to share content with others. Online social networks allow users to create and share various types of content daily. In fact, the bulk of the content displayed on these platforms is not original but shared. Thus, the ability to decipher the phenomenon of information diffusion became essential in diverse fields, such as marketeers who wish to create content that spreads, sociologists who wish to understand the underlying phenomenon, and web scientists who wish to understand the web as a sociotechnical entity. In its simplest form, the information diffusion process in online social networks consists of the content that spreads, the context that facilitates the spread, and the outcome of the process. The underlying structure on which the content spreads is the network of connections between users (the social network). Therefore, the structure of the diffusion is also a network that links users, and is based on information about who influences whom to spread the content. This network is known as the cascade. In the literature, diffusion and cascades are intersecting concepts, and they are often used interchangeably. However, this work differentiates the two. Diffusion is used to ii refer to the phenomenon while cascade is used to refer to the result of the diffusion, i.e. the structural representation of the diffusion process. This work investigates information diffusion on Tumblr, an online social network platform that provides reblogging functionality. Reblogging allows users to reblog posts, which creates a cascading behaviour that can be observed. The reblogging history is provided as a list of notes attached to each post and all of its reblogged copies. In practice, these notes have two parts: structural (who reblogged from whom) and temporal (when did the reblogging occur). These two aspects complement each other in providing an understanding of the diffusion process as it manifests in the form of a cascade. Studying such explicit cascades is important as it allows understanding the information diffusion, a phenomenon that occurs in many implicit forms on the Web. This work’s contributions include proposing an information diffusion framework that conceptualises the elements of the diffusion (namely, the content, context and cascade) and how they relate to each other. It also proposes construction models that create cascade networks minimal contextual information and missing/degraded data. In addition, this work provides a survey of the structural and temporal features of cascades, including their definitions and implications. It also investigates Tumblr as a platform for information diffusion, analyses the structural and temporal aspects of Tumblr’s cascades and compares its features with cascades obtained from other platforms. The main findings show that Tumblr’s most popular content create ‘large’ cascades that are deep, branching into a large number of separate and long paths, having a consistent number of reblogs at each depth and at each given time. These cascades gain their popularity throughout time in various ways; some of them feature high reblogging activities followed by idleness phases, others fluctuate more slowly accumulating rebloggings. Few cascades regain their popularity after long periods of idleness, while the majority have one outstanding popularity phase that is never repeated.

Increasing evidence indicates that biological neurons process information conveyed by the precise timings of individual spikes. Such observations have prompted studies on artificial networks of spiking neurons, or Spiking Neural Networks (SNNs), that use temporal encodings to represent input features. Potentially, SNNs used in this way are capable of increased computational power in comparison with rate-based networks. This thesis investigates general learning methods for SNNs which utilise the timings of single and multiple output spikes to encode information. To this end, three distinct contributions to SNN learning are made as follows. The first contribution is a proposed reward-modulated synaptic plasticity method for training SNNs to learn sequences of precisely-timed output spikes in response to spatio-temporal input patterns. Results demonstrate the high temporal accuracy of this method, even when synaptic weights in the network are modified by a delayed feedback signal. This method is potentially of biological significance, since synaptic strength modifications have been observed to be modulated by a reward signal, such as dopamine, in the nervous system. The second contribution proposes two new supervised learning rules for SNNs that perform input-output transformations of spatio-temporal spike patterns. Simulations demonstrate the rules are capable of encoding large numbers of input patterns as precisely timed output spikes, comparing favourably with existing work. The final contribution is a new supervised learning rule, termed MultilayerSpiker, for training SNNs containing hidden layers of spiking neurons to temporally encode spatio-temporal spike patterns using single or multiple output spikes. Simulations show MultilayerSpiker supports a very large number of encodings, that is a substantial improvement over existing spike-based multilayer rules, and provides increased classification accuracy when using the timings of multiple rather than single output spikes to identify input patterns.

La recherche et l'indexation de données en fonction d'une date ou d'une zone géographique permettent le partage et la découverte d'informations géolocalisées telles que l'on en trouve sur les réseaux sociaux comme Facebook, Flickr, ou Twitter. Cette réseau social connue sous le nom de Location Based Social Network (LBSN) s'applique à des millions d'utilisateurs qui partagent et envoient des requêtes ciblant des zones spatio-temporelles, permettant d'accéder à des données géolocalisées générées dans une zone géographique et dans un intervalle de temps donné. Un des principaux défis pour de telles applications est de fournir une architecture capable de traiter la multitude d'insertions et de requêtes spatio-temporelles générées par une grande quantité d'utilisateurs. A ces fins, les Tables de Hachage Distribué (DHT) et le paradigme Pair-à-Pair (P2P) sont autant de primitives qui forment la base pour les applications de grande envergure. Cependant, les DHTs sont mal adaptées aux requêtes ciblant des intervalles donnés; en effet, l'utilisation de fonctions de hachage sacrifie la localité des données au profit d'un meilleur équilibrage de la charge. Plusieurs solutions ajoutent le support de requêtes ciblant des ensembles aux DHTs. En revanche ces solutions ont tendance à générer un nombre de messages et une latence élevée pour des requêtes qui ciblent des intervalles. Cette thèse propose deux solutions à large échelle pour l'indexation des données géolocalisées
Indexing and retrieving data by location and time allows people to share and explore massive geotagged datasets observed on social networks such as Facebook, Flickr, and Twitter. This scenario known as a Location Based Social Network (LBSN) is composed of millions of users, sharing and performing location-temporal range queries in order to retrieve geotagged data generated inside a given geographic area and time interval. A key challenge is to provide a scalable architecture that allow to perform insertions and location-temporal range queries from a high number of users. In order to achieve this, Distributed Hash Tables (DHTs) and the Peer-to-Peer (P2P) computing paradigms provide a powerful building block for implementing large scale applications. However, DHTs are ill-suited for supporting range queries because the use of hash functions destroy data locality for the sake of load balance. Existing solutions that use a DHT as a building block allow to perform range queries. Nonetheless, they do not target location-temporal range queries and they exhibit poor performance in terms of query response time and message traffic. This thesis proposes two scalable solutions for indexing and retrieving geotagged data based on location and time

The approach developed by Eynon-Switzer (1983) to analyze the spatial-temporal structure of a data set obtained from the EPRI monitoring network is applied to a data set obtained from the MAP3S/PCN monitoring network. In this approach, a spatio-temporal stochastic model, including deterministic components for seasonal variation and rainfall washout, is fitted to the data. The results indicate that the model fails to capture some of the features of the underlying structure. In an effort to identify an appropriate model for the data, we examine the raw data in detail. An ANOVA model is fitted to the data. Different criteria such as Akaike, Schwarz, Mallows, etc, are used to identify the 'best' submodel (i.e. eliminate some terms in the full ANOVA model). The results indicate that it is possible to capture the deterministic component of the model with a much smaller model (i.e. fewer parameters). The normality of the residuals is also examined. The results indicate that the data from all stations except one can reasonably be approximated as coming from normal distributions.
Science, Faculty of
Statistics, Department of
Graduate

Nous avons étudié l’impact de la structure temporelle de l'activité neuronale sur la dynamique de réseaux recevant cette activité et impliqués dans la cognition. Nous avons caractérisé le code qui permet de lire l'information dans des signaux du cortex cingulaire antérieur dorsal (CCAd) simien, qui intervient dans les processus d'adaptation comportementale. Nos analyses suggèrent que la variabilité importante du nombre de potentiels d'action émis par les neurones, ainsi que la fiabilité temporelle conséquente de ces signaux, favorisent un décodage par des réseaux sensibles à la structure temporelle. De plus, quand nous avons séparé les données entre un groupe avec un grand nombre de potentiels d'action, et un groupe avec un faible nombre de potentiels d'action, nous n'avons pas trouvé pas de différence robuste du comportement du singe entre ces deux groupes. Par contre, lorsque l'activité d'un neurone devenait moins semblable à la réponse typique de ce neurone, le singe semblent répondait plus lentement pendant la tâche comportementale. L'activité d'un neurone semblait pouvoir se différencier de sa réponse typique par une augmentation ou une diminution du nombre de potentiels d'actions, ou par des imprécisions sur le temps d'émission des potentiels d'action. Nos résultats suggèrent que les réseaux neuronaux qui décodent les signaux du CCAd détectent des motifs spatiotemporels. Enfin, nous avons ensuite analysé mathématiquement des modèles de réseaux de neurones récurrents dans le but mieux comprendre l’impact des signaux du CCAd sur le décodeur. Le modèle de neurone utilisé peut reproduire la réponse dynamique de neurones biologique par l’inclusion d’une adaptation neurale
We investigated the putative function of the fine temporal dynamics of neuronal networks for implementing cognitive processes. First, we characterized the coding properties of spike trains recorded from the dorsal Anterior Cingulate Cortex (dACC) of monkeys. dACC is thought to trigger behavioral adaptation. We found evidence for (i) high spike count variability and (ii) temporal reliability (favored by temporal correlations) which respectively hindered and favored information transmission when monkeys were cued to switch the behavioral strategy. Also, we investigated the nature of the neuronal variability that was predictive of behavioral variability. High vs. low firing rates were not robustly associated with different behavioral responses, while deviations from a neuron-specific prototypical spike train predicted slower responses of the monkeys. These deviations could be due to increased or decreased spike count, as well as to jitters in spike times. Our results support the hypothesis of a complex spatiotemporal coding of behavioral adaptation by dACC, and suggest that dACC signals are unlikely to be decoded by a neural integrator. Second, we further investigated the impact of dACC temporal signals on the downstream decoder by developing mean-field equations to analyze network dynamics. We used an adapting single neuron model that mimics the response of cortical neurons to realistic dynamic synaptic-like currents. We approximated the time-dependent population rate for recurrent networks in an asynchronous irregular state. This constitutes an important step towards a theoretical study of the effect of temporal drives on networks which could mediate cognitive functions

La caractérisation de la structure du cerveau humain et les motifs fonctionnelles qu’il fait apparaitre est un défi central pour une meilleure compréhension des pathologies du réseau cérébral telle que l’épilepsie du lobe temporal. Cette caractérisation pourrait aider à améliorer la prédictibilité clinique des résultats d’une chirurgie visant à traiter l’épilepsie.Le fonctionnement du cerveau peut être étudié par l’électroencéphalographie (EEG) ou par l’imagerie de résonance magnétique fonctionnelle (IRMf), alors que la structure peut être caractérisé par l’IRM de diffusion (IRMd). Nous avons utilisé ces modalités pour mesurer le fonctionnement du cerveau pendant une tache de reconnaissance de visages et pendant le repos dans le but de faire le lien entre les modalités d’une façon optimale en termes de résolution temporale et spatiale. Avec cette approche on a mis en évidence une perturbation des relations structure-fonction chez les patients épileptiques.Ce travail a contribué à améliorer la compréhension de l’épilepsie comme une maladie de réseau qui affecte le cerveau à large échelle et non pas au niveau d’un foyer épileptique local. Dans le futur, ces résultats pourraient être utilisés pour guider des interventions chirurgicales mais ils fournissent également des approches nouvelles pour évaluer des traitements pharmacologiques selon ses implications fonctionnelles à l’échelle du cerveau entier
The understanding human brain structure and the function patterns arising from it is a central challenge to better characterize brain network pathologies such as temporal lobe epilepsies, which could help to improve the clinical predictability of epileptic surgery outcome.Brain functioning can be accessed by both electroencephalography (EEG) or functional magnetic resonance imaging (fMRI), while brain structure can be measured with diffusion MRI (dMRI). We use these modalities to measure brain functioning during a face recognition task and in rest in order to link the different modalities in an optimal temporal and spatial manner. We discovered disruption of the network processing famous faces as well a disruption of the structure-function relation during rest in epileptic patients.This work broadened the understanding of epilepsy as a network disease that changes the brain on a large scale not limited to a local epileptic focus. In the future these results could be used to guide clinical intervention during epilepsy surgery but also they provide new approaches to evaluate pharmacological treatment on its functional implications on a whole brain scale

Die flexible Anwendung von Wissen in neuen Alltagssituationen ist eine notwendige kognitive Fähigkeit. Bisherige Studien betonen die zentrale Rolle des Hippocampus beim Lernen und Verknüpfen neuer Informationen mit bereits vorhandenem Wissen. Die funktionelle Integrität des Hippocampus ändert sich jedoch im Laufe des Lebens bzw. wird durch neuropsychiatrische Erkrankungen häufig beeinflusst. Die betroffenen Personen müssen deswegen adaptive Strategien entwickeln, um behaviorale Ziele weiter zu erreichen. Daher befasst sich meine Doktorarbeit mit Adaptationsprozessen im sich entwickelnden Gehirn und im vollständig entwickelten Gehirn mit einer hippocampalen Dysfunktion. Diese Synopsis umfasst dazu drei Studien: (1) zu behavioralen Strategien im sich entwickelnden Gehirn, (2) zu behavioralen Strategien im vollständig entwickelten Gehirn nach einer Läsion und (3) zu strukturellen Veränderungen im vollständig entwickelten Gehirn nach einer Läsion. Studie 1 zeigt einen altersgebundenen Wechsel beim assoziativen Gedächtnis: Kinder, Jugendliche und junge Erwachsene benutzen verschiedene Gedächtnisstrategien beim Integrieren von Gedächtnisinhalten. Studie 2 zeigt, dass die beobachteten Gedächtnisbeeinträchtigungen bei Patienten mit rechtsseitigen hippocampalen Läsionen sich nicht alleine durch ein Defizit des assoziativen Gedächtnisses erklären lassen, sondern auf einen zusätzlichen hippocampalen Beitrag zur Gedächtnisintegration zurückzuführen sind. Studie 3 zeigt, dass sich postoperative Adaptationsprozesse auf struktureller Ebene in überraschend kurzer Zeit ereignen und dass die strukturelle Reorganisation nicht nur im Hippocampus, sondern auch in entfernteren Hirnregionen, die mit dem Hippocampus verbunden sind, stattfindet. Zusammenfassend zeigen die Ergebnisse der drei Studien, dass Adaptationsprozesse im sich entwickelnden Gehirn sowie bei Erwachsenen mit einer hippocampalen Dysfunktion sowohl auf der behavioralen als auch auf der strukturellen Ebene auftreten.
Applying knowledge flexibly to new situations is a cognitive faculty that is necessary in every-day life. Previous findings emphasise the crucial role the hippocampus plays in learning and linking new information with pre-existing knowledge. However, the functional integrity of the hippocampus changes over the lifespan and is frequently affected by neuropsychiatric disorders. The affected subjects must, therefore, develop adaptive strategies to achieve behavioural goals. Thus, my doctoral thesis deals with adaptation processes in the developing brain and in adult brains with a hippocampal dysfunction. This synopsis encompasses three studies on: (1) behavioural strategies in the developing brain, (2) behavioural strategies in the lesioned fully developed brain, and (3) structural changes in the lesioned fully developed brain. Study 1 suggests an age-related shift in the associative memory: Children, adolescents, and young adults use different memory strategies when integrating information. Study 2 suggests that the memory deficits observed in patients with right-sided hippocampal lesions are not merely a consequence of an impaired associative memory but rather result from an additional hippocampal contribution to the memory integration. Study 3 suggests that postoperative structural adaptation processes occur on a surprisingly short time-scale, and this structural reorganisation happens not only in the hippocampus but also in distant brain areas connected to the hippocampus. In conclusion, findings from these three studies show that adaptation processes in the developing brain and in adult brains with hippocampal dysfunction occur on both the behavioural and the structural level.

Cette thèse introduit différentes méthodes de vérification (ou model-checking) sur des modèles de systèmes à pile. En effet, les systèmes à pile (pushdown systems) modélisent naturellement les programmes séquentiels grâce à une pile infinie qui peut simuler la pile d'appel du logiciel. La première partie de cette thèse se concentre sur la vérification sur des systèmes à pile de la logique HyperLTL, qui enrichit la logique temporelle LTL de quantificateurs universels et existentiels sur des variables de chemin. Il a été prouvé que le problème de la vérification de la logique HyperLTL sur des systèmes d'états finis est décidable ; nous montrons que ce problème est en revanche indécidable pour les systèmes à pile ainsi que pour la sous-classe des systèmes à pile visibles (visibly pushdown systems). Nous introduisons donc des algorithmes d'approximation de ce problème, que nous appliquons ensuite à la vérification de politiques de sécurité. Dans la seconde partie de cette thèse, dans la mesure où la représentation de la pile d'appel par les systèmes à pile est approximative, nous introduisons les systèmes à surpile (pushdown systems with an upper stack) ; dans ce modèle, les symboles retirés de la pile d'appel persistent dans la zone mémoire au dessus du pointeur de pile, et peuvent être plus tard écrasés par des appels sur la pile. Nous montrons que les ensembles de successeurs post* et de prédécesseurs pre* d'un ensemble régulier de configurations ne sont pas réguliers pour ce modèle, mais que post* est toutefois contextuel (context-sensitive), et que l'on peut ainsi décider de l'accessibilité d'une configuration. Nous introduisons donc des algorithmes de sur-approximation de post* et de sous-approximation de pre*, que nous appliquons à la détection de débordements de pile et de manipulations nuisibles du pointeur de pile. Enfin, dans le but d'analyser des programmes avec plusieurs fils d'exécution, nous introduisons le modèle des réseaux à piles dynamiques synchronisés (synchronized dynamic pushdown networks), que l'on peut voir comme un réseau de systèmes à pile capables d'effectuer des changements d'états synchronisés, de créer de nouveaux systèmes à piles, et d'effectuer des actions internes sur leur pile. Le problème de l'accessibilité étant naturellement indécidable pour un tel modèle, nous calculons une abstraction des chemins d'exécutions entre deux ensembles réguliers de configurations. Nous appliquons ensuite cette méthode à un processus itératif de raffinement des abstractions
In this thesis, we propose different model-checking techniques for pushdown system models. Pushdown systems (PDSs) are indeed known to be a natural model for sequential programs, as they feature an unbounded stack that can simulate the assembly stack of an actual program. Our first contribution consists in model-checking the logic HyperLTL that adds existential and universal quantifiers on path variables to LTL against pushdown systems (PDSs). The model-checking problem of HyperLTL has been shown to be decidable for finite state systems. We prove that this result does not hold for pushdown systems nor for the subclass of visibly pushdown systems. Therefore, we introduce approximation algorithms for the model-checking problem, and show how these can be used to check security policies. In the second part of this thesis, as pushdown systems can fail to accurately represent the way an assembly stack actually operates, we introduce pushdown systems with an upper stack (UPDSs), a model where symbols popped from the stack are not destroyed but instead remain just above its top, and may be overwritten by later push rules. We prove that the sets of successors post* and predecessors pre* of a regular set of configurations of such a system are not always regular, but that post* is context-sensitive, hence, we can decide whether a single configuration is forward reachable or not. We then present methods to overapproximate post* and under-approximate pre*. Finally, we show how these approximations can be used to detect stack overflows and stack pointer manipulations with malicious intent. Finally, in order to analyse multi-threaded programs, we introduce in this thesis a model called synchronized dynamic pushdown networks (SDPNs) that can be seen as a network of pushdown processes executing synchronized transitions, spawning new pushdown processes, and performing internal pushdown actions. The reachability problem for this model is obviously undecidable. Therefore, we compute an abstraction of the execution paths between two regular sets of configurations. We then apply this abstraction framework to a iterative abstraction refinement scheme

This thesis draws on David Harvey’s concept of “accumulation by dispossession” and an international political economy (IPE) approach centred on the institutional arrangements and power structures that privilege certain actors and values, in order to critique current capitalist practices of primitive accumulation by the global corporate extractive industry. The thesis examines how accumulation by dispossession by the global extractive industry is facilitated by the “free entry” or “free mining” principle. It does so by focusing on Canada as a leader in the global extractive industry and the spread of this country’s mining laws to other countries – in other words, the transnationalisation of norms in the global extractive industry – so as to maintain a consistent and familiar operating environment for Canadian extractive companies. The transnationalisation of norms is further promoted by key international institutions such as the World Bank, which is also the world’s largest development lender and also plays a key role in shaping the regulations that govern natural resource extraction. The thesis briefly investigates some Canadian examples of resource extraction projects, in order to demonstrate the weaknesses of Canadian mining laws, particularly the lack of protection of landowners’ rights under the free entry system and the subsequent need for “free, prior and informed consent” (FPIC). The thesis also considers some of the challenges to the adoption and implementation of the right to FPIC. These challenges include embedded institutional structures like the free entry mining system, international political economy (IPE) as shaped by international institutions and powerful corporations, as well as concerns regarding ‘local’ power structures or the legitimacy of representatives of communities affected by extractive projects. The thesis concludes that in order for Canada to be truly recognized as a leader in the global extractive industry, it must establish legal norms domestically to ensure that Canadian mining companies and residents can be held accountable when there is evidence of environmental and/or human rights violations associated with the activities of Canadian mining companies abroad. The thesis also concludes that Canada needs to address underlying structural issues such as the free entry mining system and implement FPIC, in order to curb “accumulation by dispossession” by the extractive industry, both domestically and abroad.

博士
國立臺灣大學
電機工程學研究所
106
Structural network analysis for temporal social networks is an essential discipline for comprehending human behaviors and interactions on social networks. For systematically interpreting the temporal networks, we raise two fundamental questions and propose a general framework to track, model, and predict the structures of time-varying networks. In this dissertation, both temporal user-item (bipartite) networks and temporal social (unipartite) networks are scrutinized respectively. We introduce temporal bipartite projection (TBP) to socially aggregate the temporal information among users and represent the item transition tendencies within an item transition graph (ITG). Based on the ITG, we propose a scoring function called STEP (Score for TEmporal Prediction) for each user-item pair which is for performing the new link prediction task. Furthermore, we introduce temporal Laplacian eigenmaps (TLE) for determining the sequence of latent feature vectors for each node from temporal networks. A general prediction framework is proposed based on the results of TLE, which use the Finite Impulse Response (FIR) filter to learn the dynamics of evolving latent feature vectors of users. Then, the predicted feature vectors are used for various network analysis applications, including community detection, link prediction, and node ranking. Besides, we also use the recurrent neural networks (RNNs) to model the temporal latent feature vectors for better accuracy. To validate the effectiveness of our frameworks, we conduct various experiments based on our synthetic datasets and real-world datasets such as DBLP, Flickr, Delicious for temporal user-item networks and Infectious, Haggle, Reality Mining for temporal social networks. Our experimental results show that our framework is very effective in tracking latent feature vectors and predicting future network structures.

Localization is a fundamental operation for many wireless networks. While GPS is widely used for location determination, it is unavailable in many environments either due to its high cost or the lack of line of sight to the satellites (e.g., indoors, under the ground, or in a downtown canyon). The limitations of GPS have motivated researchers to develop many localization schemes to infer locations based on measured wireless signals. However, most of these existing schemes focus on localization in static wireless networks. As many wireless networks are mobile (e.g., mobile sensor networks, disaster recovery networks, and vehicular networks), we focus on localization in mobile networks in this thesis. We analyze real mobility traces and find that they exhibit temporal stability and low-rank structure. Motivated by this observation, we develop three novel localization schemes to accurately determine locations in mobile networks: 1. Low Rank based Localization (LRL), which exploits the low-rank structure in mobility. 2. Temporal Stability based Localization (TSL), which leverages the temporal stability. 3. Temporal Stability and Low Rank based Localization (TSLRL), which incorporates both the temporal stability and the low-rank structure. These localization schemes are general and can leverage either mere connectivity (i.e., range-free localization) or distance estimation between neighbors (i.e., range-based localization). Using extensive simulations and testbed experiments, we show that our new schemes significantly outperform state-of-the-art localization schemes under a wide range of scenarios and are robust to measurement errors.
text

博士
國立中央大學
地球物理研究所
93
The seismicity is low in northwestern Taiwan for a long time. However, recent studies indicated the geological characteristics in this area are special. Besides that, many important facilities, industrial and economic cities are in this area. Significant intensity was recorded at seismic stations in Hsinchu area during the 1999 Chi-Chi earthquake. And the disastrous Hsinchu-Taichung earthquake in 1935 occurred in the southern part of this area. Thus, potential recurrence of large earthquakes in this area becomes important topics. In response, the seismicity and subsurface structures are analyzed in this study. A temporary seismic network including ten seismic stations was deployed in Taoyuan, Hsinchu and Miaoli counties in northwestern Taiwan since January 2001. Each station has one triaxial accelerograph, three external one-component velocity sensors, global positioning system, and data storage device. Seismic records with absolute timing are critical to obtain accurate earthquake locations. Dense station distribution is necessary to get earthquakes with lower magnitude from clear seismic signals recorded in local area. Based on considerations of available instruments and recording sites, the stations were deployed uniformly in the study area. In this study, the arrival time data of the temporary seismic network and CWBSN are combined to locate earthquake. Two dense earthquake clusters were relocated to compare several geological cross sections. In order to study the stress patterns, numerous focal mechanisms were determined by waveform inversion. Finally, the thicknesses of alluvium were estimated by using the P-wave travel-time residuals from earthquake location and the dominant frequencies identified from spectra of ground acceleration. By combining the temporary seismic network and CWBSN data, the results of earthquake location show significant convergent in focal depths due to adding of the near source arrival time data. Most of the hypocenters are located shallower than 15 km at depth. Relocation of two dense clusters using the JHD and DD methods removed systematic bias due to one-dimensional velocity model. The events were shifted toward northwest in the horizontal direction and became clustered at depth from 5 to 10 km. The station corrections of JHD reflected the difference in geology of the northwest and southeast parts in the study area. It is also consistent with distinct topographic features. By comparing the relocated events with several geological cross sections, we found that the seismicity and subsurface structures are related. To determine the focal mechanisms using waveform inversion, the acceleration records were used. The acceleration records of a single station with three-component sensors are doubly integrated to get displacement waveforms. Then the focal mechanisms are determined by waveform inversion. In total, 88 focal mechanisms were determined with local magnitudes from 1.35 to 3.33. The widespread presences of complex fault types of focal mechanisms imply that the microearthquakes might be triggered by subfaults. Finally, the thicknesses of alluvium were estimated by correlating the velocity and acceleration data. Results from the averaged P-wave travel-time residuals from earthquake location and from the dominant frequency from Fourier spectra are not consistent to each other. The thicknesses of alluvium obtained by these two methods are different. However, the averaged P-wave travel-time residuals can be correlated with geology and topography. There is good relation between travel-time residuals and geological cross-sections. The results imply that the averaged P-wave travel-time residuals can be used to prospect the subsurface structures.

Over the last decade advances in multineuron simultaneous recording techniques have produced huge amounts of data. This has led to the investigation of probable temporal relationships between spike times of neurons as manifestations of the underlying network structure. But the huge dimensionality of data makes the search for patterns difficult. Although this difficulty may be surpassed by employing massive computing resources, understanding the significance and relation of these temporal patterns to the underlying network structure and the causative activity is still difficult. To find such relationships in networks of excitatory neurons, a simplified network structure of feedforward chains called "Synfire chains" has been frequently employed. But in a recurrently connected network where activity from feedback connections is comparable to the feedforward chain, the basic assumptions underlying synfire chains are violated. In the first part of this thesis we propose the first-spike-latency based analysis as a low complexity method of studying the temporal relationships between neurons. Firstly, spike latencies being temporal delays measured at a particular epoch of time (onset of activity after a quiescent period) are a small subset of all the temporal information available in spike trains, thereby hugely reducing the amount of data that needs to be analyzed. We also define for the first time, "Synconset waves and chains" as a sequence of first-spike-times and the causative neuron chain. Using simulations, we show the efficacy of the synconset paradigm in unraveling feedforward chains of excitatory neurons even in a recurrent network. We further create a framework for going back and forth between network structure and the observed first-spike-latency patterns. To quantify these associations between network structure and dynamics we propose a likelihood measure based on Bayesian reasoning. This quantification is agnostic to the methods of association used and as such can be used with any of the existing approaches. We also show the benefits of such an analysis when the recorded data is subsampled, as is the case with most physiological recordings. In the subsequent part of our thesis we show two sample applications of first-spike-latency analysis on data acquired from multielectrode arrays. Our first application dwells on the intricacies of extracting first-spike-latency patterns from multineuron recordings using recordings of glutamate injured cultures. We study the significance of these patterns extracted vis-a-vis patterns that may be obtained from exponential spike latency distributions and show the differences between patterns obtained in injured and control cultures. In a subsequent application, we study the evolution of latency patterns over several days during the lifetime of a dissociated hippocampal culture.

In order to understand how the neural system encodes and processes information, research has focused on the study of neural representations of simple stimuli, paying no particular attention to it's temporal structure, with the assumption that a deeper understanding of how the neural system processes simpli fied stimuli will lead to an understanding of how the brain functions as a whole [1]. However, time is intrinsically bound to neural processing as all sensory, motor, and cognitive processes are inherently dynamic. Despite the importance of neural and stimulus dynamics, little is known of how the neural system represents rich spatio-temporal stimulus, which ultimately link the neural system to a continuously changing environment. The purpose of this thesis is to understand whether and how temporally-structured neural activity modulates the processing of information within the brain, proposing in turn that, the precise interaction between the spatio-temporal structure of the stimulus and the neural system is particularly relevant, particularly when considering the ongoing plasticity mechanisms which allow the neural system to learn from experience. In order to answer these questions, three studies were conducted. First, we studied the impact of spiking temporal structure on a single neuron spiking response, and explored in which way the functional connections to pre-synaptic neurons are modulated through adaptation. Our results suggest that, in a generic spiking neuron, the temporal structure of pre-synaptic excitatory and inhibitory neurons modulate both the spiking response of that same neuron and, most importantly, the speed and strength of learning. In the second, we present a generic model of a spiking neural network that processes rich spatio-temporal stimuli, and explored whether the processing of stimulus within the network is modulated due to the interaction with an external dynamical system (i.e. extracellular media), as well as several plasticity mechanisms. Our results indicate that the memory capacity, that re ects a dynamic short-term memory of incoming stimuli, can be extended on the presence of plasticity and through the interaction with an external dynamical system, while maintaining the network dynamics in a regime suitable for information processing. Finally, we characterized cortical signals of human subjects (electroencephalography, EEG) associated to a visual categorization task. Among other aspects, we studied whether changes in the dynamics of the stimulus leads to a changes in the neural processing at the cortical level, and introduced the relevance of large-scale integration for cognitive processing. Our results suggest that the dynamic synchronization across distributed cortical areas is stimulus specific and specifically linked to perceptual grouping. Taken together, the results presented here suggest that the temporal structure of the stimulus modulates how the neural system encodes and processes information within single neurons, network of neurons and cortical areas. In particular, the results indicate that timing modulates single neuron connectivity structures, the memory capability of networks of neurons, and the cortical representation of a visual stimuli. While the learning of invariant representations remains as the best framework to account for a number of neural processes (e.g. long-term memory [2]), the reported studies seem to provide support the idea that, at least to some extent, the neural system functions in a non-stationary fashion, where the processing of information is modulated by the stimulus dynamics itself. Altogether, this thesis highlights the relevance of understanding adaptive processes and their interaction with the temporal structure of the stimulus, arguing that a further understanding how the neural system processes dynamic stimuli is crucial for the further understanding of neural processing itself, and any theory that aims to understand neural processing should consider the processing of dynamic signals. 1. Frankish, K., and Ramsey, W. The Cambridge Handbook of Cognitive Science. Cambridge University Press, 2012. // 2. McGaugh, J. L. Memory{a Century of Consolidation. Science 287, 5451 (Jan. 2000), 248{251.

Indiana University-Purdue University Indianapolis (IUPUI)
With the increasing availability of Electronic Health Records (EHRs) and advances in deep learning techniques, developing deep predictive models that use EHR data to solve healthcare problems has gained momentum in recent years. The majority of clinical predictive models benefit from structured data in EHR (e.g., lab measurements and medications). Still, learning clinical outcomes from all possible information sources is one of the main challenges when building predictive models. This work focuses mainly on two sources of information that have been underused by researchers; unstructured data (e.g., clinical notes) and a patient network. We propose a novel hybrid deep learning model, DeepNote-GNN, that integrates clinical notes information and patient network topological structure to improve 30-day hospital readmission prediction. DeepNote-GNN is a robust deep learning framework consisting of two modules: DeepNote and patient network. DeepNote extracts deep representations of clinical notes using a feature aggregation unit on top of a state-of-the-art Natural Language Processing (NLP) technique - BERT. By exploiting these deep representations, a patient network is built, and Graph Neural Network (GNN) is used to train the network for hospital readmission predictions. Performance evaluation on the MIMIC-III dataset demonstrates that DeepNote-GNN achieves superior results compared to the state-of-the-art baselines on the 30-day hospital readmission task. We extensively analyze the DeepNote-GNN model to illustrate the effectiveness and contribution of each component of it. The model analysis shows that patient network has a significant contribution to the overall performance, and DeepNote-GNN is robust and can consistently perform well on the 30-day readmission prediction task. To evaluate the generalization of DeepNote and patient network modules on new prediction tasks, we create a multimodal model and train it on structured and unstructured data of MIMIC-III dataset to predict patient mortality and Length of Stay (LOS). Our proposed multimodal model consists of four components: DeepNote, patient network, DeepTemporal, and score aggregation. While DeepNote keeps its functionality and extracts representations of clinical notes, we build a DeepTemporal module using a fully connected layer stacked on top of a one-layer Gated Recurrent Unit (GRU) to extract the deep representations of temporal signals. Independent to DeepTemporal, we extract feature vectors of temporal signals and use them to build a patient network. Finally, the DeepNote, DeepTemporal, and patient network scores are linearly aggregated to fit the multimodal model on downstream prediction tasks. Our results are very competitive to the baseline model. The multimodal model analysis reveals that unstructured text data better help to estimate predictions than temporal signals. Moreover, there is no limitation in applying a patient network on structured data. In comparison to other modules, the patient network makes a more significant contribution to prediction tasks. We believe that our efforts in this work have opened up a new study area that can be used to enhance the performance of clinical predictive models.

Adaptability and flexibility are some of the most important human characteristics. Learning based on new experiences enables adaptation by changing the structural connectivity of the brain through plasticity mechanisms. But the human brain can also adapt to new tasks and situations in a matter of milliseconds by dynamic coordination of functional activation. To understand how this flexibility can be achieved in the computations performed by neural networks, we have to understand how the relatively fixed structural backbone interacts with the functional dynamics. In this thesis, I will analyze these interactions between the structural network connectivity and functional activations and their dynamic interactions on different levels of abstraction and spatial and temporal scales. One of the big questions in neuroscience is how functional interactions in the brain can adapt instantly to different tasks while the brain structure remains almost static. To improve our knowledge of the neural mechanisms involved, I will first analyze how dynamics in functional brain activations can be simulated based on the structural brain connectivity obtained with diffusion tensor imaging. In particular, I will show that a dynamic model of functional connectivity in the human cortex is more predictive of empirically measured functional connectivity than a stationary model of functional dynamics. More specifically, the simulations of a coupled oscillator model predict 54\% of the variance in the empirically measured EEG functional connectivity. Hypotheses of temporal coding have been proposed for the computational role of these dynamic oscillatory interactions on fast timescales. These oscillatory interactions play a role in the dynamic coordination between brain areas as well as between cortical columns or individual cells. Here I will extend neural network models, which learn unsupervised from statistics of natural stimuli, with phase variables that allow temporal coding in distributed representations. The analysis shows that synchronization of these phase variables provides a useful mechanism for binding of activated neurons, contextual coding, and figure ground segregation. Importantly, these results could also provide new insights for improvements of deep learning methods for machine learning tasks. The dynamic coordination in neural networks has also large influences on behavior and cognition. In a behavioral experiment, we analyzed multisensory integration between a native and an augmented sense. The participants were blindfolded and had to estimate their rotation angle based on their native vestibular input and the augmented information. Our results show that subjects alternate in the use between these modalities, indicating that subjects dynamically coordinate the information transfer of the involved brain regions. Dynamic coordination is also highly relevant for the consolidation and retrieval of associative memories. In this regard, I investigated the beneficial effects of sleep for memory consolidation in an electroencephalography (EEG) study. Importantly, the results demonstrate that sleep leads to reduced event-related theta and gamma power in the cortical EEG during the retrieval of associative memories, which could indicate the consolidation of information from hippocampal to neocortical networks. This highlights that cognitive flexibility comprises both dynamic organization on fast timescales and structural changes on slow timescales. Overall, the computational and empirical experiments demonstrate how the brain evolved to a system that can flexibly adapt to any situation in a matter of milliseconds. This flexibility in information processing is enabled by an effective interplay between the structure of the neural network, the functional activations, and the dynamic interactions on fast time scales.

Difficulty understanding speech in background noise is the most common hearing complaint. Elucidating the neurophysiological mechanisms underlying speech intelligibility in everyday environments with multiple sound sources and distortions is hence important for any technology that aims to improve real-world listening. Using a combination of behavioral, electroencephalography (EEG), and computational modeling experiments, this dissertation provides insight into how the brain analyzes such complex scenes, and what roles different acoustic cues play in facilitating this process and in conveying phonetic content. Experiment #1 showed that brain oscillations selectively track the temporal envelopes (i.e., modulations) of attended speech in a mixture of competing talkers, and that the strength and pattern of this attention effect differs between individuals. Experiment #2 showed that the fidelity of neural tracking of attended-speech envelopes is strongly shaped by the modulations in interfering sounds as well as the temporal fine structure (TFS) conveyed by the cochlea, and predicts speech intelligibility in diverse listening environments. Results from Experiments #1 and #2 support the theory that temporal coherence of sound elements across envelopes and/or TFS shapes scene analysis and speech intelligibility. Experiment #3 tested this theory further by measuring and computationally modeling consonant categorization behavior in a range of background noises and distortions. We found that a physiologically plausible model that incorporated temporal-coherence effects predicted consonant confusions better than conventional speech-intelligibility models, providing independent evidence that temporal coherence influences scene analysis. Finally, results from Experiment #3 also showed that TFS is used to extract speech content (voicing) for consonant categorization even when intact envelope cues are available. Together, the novel insights provided by our results can guide future models of speech intelligibility and scene analysis, clinical diagnostics, improved assistive listening devices, and other audio technologies.