Дисертації з теми "Dynamic Representation Learning"

Enterprises use live video streaming as a mean of communication. Streaming high-quality video to thousands of devices in a corporate network is not an easy task; the bandwidth requirements often exceed the network capacity. For that matter, Peer-To-Peer (P2P) networks have been proven beneficial, as peers can exchange content efficiently by utilizing the topology of the corporate network. However, such networks are dynamic and their topology might not always be known. In this project we propose ABD, a new dynamic graph representation learning approach, which aims to estimate the bandwidth capacity between peers in a corporate network. The architecture of ABDis adapted to the properties of corporate networks. The model is composed of an attention mechanism and a decoder. The attention mechanism produces node embeddings, while the decoder converts those embeddings into bandwidth predictions. The model aims to capture both the dynamicity and the structure of the dynamic network, using an advanced training process. The performance of ABD is tested with two dynamic graphs which were produced by real corporate networks. Our results show that ABD achieves better results when compared to existing state-of-the-art dynamic graph representation learning models.
Företag använder live video streaming för både intern och extern kommunikation. Strömmning av hög kvalitet video till tusentals tittare i ett företagsnätverk är inte enkelt eftersom bandbreddskraven ofta överstiger kapaciteten på nätverket. För att minska lasten på nätverket har Peer-to-Peer (P2P) nätverk visat sig vara en lösning. Här anpassar sig P2P nätverket efter företagsnätverkets struktur och kan därigenom utbyta video data på ett effektivt sätt. Anpassning till ett företagsnätverk är ett utmanande problem eftersom dom är dynamiska med förändring över tid och kännedom över topologin är inte alltid tillgänglig. I det här projektet föreslår vi en ny lösning, ABD, en dynamisk approach baserat på inlärning av grafrepresentationer. Vi försöker estimera den bandbreddskapacitet som finns mellan två peers eller tittare. Architekturen av ABD anpassar sig till egenskaperna av företagsnätverket. Själva modellen bakom ABD använder en koncentrationsmekanism och en avkodare. Attention mekanismen producerar node embeddings, medan avkodaren konverterar embeddings till estimeringar av bandbredden. Modellen fångar upp dynamiken och strukturen av nätverket med hjälp av en avancerad träningsprocess. Effektiviteten av ABD är testad på två dynamiska nätverksgrafer baserat på data från riktiga företagsnätverk. Enligt våra experiment har ABD bättre resultat när man jämför med andra state-of the-art modeller för inlärning av dynamisk grafrepresentation.

Forming an effective multi-robot team to perform a task is a key problem in many domains. The performance of a multi-robot team depends on the robots the team is composed of, where each robot has different capabilities. Team performance has previously been modeled as the sum of single-robot capabilities, and these capabilities are assumed to be known. Is team performance just the sum of single-robot capabilities? This thesis is motivated by instances where agents perform differently depending on their teammates, i.e., there is synergy in the team. For example, in human sports teams, a well-trained team performs better than an allstars team composed of top players from around the world. This thesis introduces a novel model of team synergy — the Synergy Graph model — where the performance of a team depends on each robot’s individual capabilities and a task-based relationship among them. Robots are capable of learning to collaborate and improving team performance over time, and this thesis explores how such robots are represented in the Synergy Graph Model. This thesis contributes a novel algorithm that allocates training instances for the robots to improve, so as to form an effective multi-robot team. The goal of team formation is the optimal selection of a subset of robots to perform the task, and this thesis contributes team formation algorithms that use a Synergy Graph to form an effective multi-robot team with high performance. In particular, the performance of a team is modeled with a Normal distribution to represent the nondeterminism of the robots’ actions in a dynamic world, and this thesis introduces the concept of a δ-optimal team that trades off risk versus reward. Further, robots may fail from time to time, and this thesis considers the formation of a robust multi-robot team that attains high performance even if failures occur. This thesis considers ad hoc teams, where the robots of the team have not collaborated together, and so their capabilities and synergy are initially unknown. This thesis contributes a novel learning algorithm that uses observations of team performance to learn a Synergy Graph that models the capabilities and synergy of the team. Further, new robots may become available, and this thesis introduces an algorithm that iteratively updates a Synergy Graph with new robots.

Intelligent agents are designed to interact with, and learn about, their environment so that they can act purposefully towards a goal. One class of problems encountered in building such agents is learning how to respond to dynamic systems with a continuous state space. The goals of this dissertation are to develop a framework for understanding the behaviour of partitioned dynamic systems with continuous underlying state and to translate this framework into algorithms which adaptively form a partition of the continuous space such that the partitioned system is more easily learned and controlled, and such that the control law may be easily explained in intuitive ways. Currently, algorithms which learn a control policy for partitioned continuous state space systems treat the partitioned system as an approximation to a Markov chain. I give conditions for the partitioned system to be a Markov chain, a semi-Markov process and a new class of system, a weak-semi-Markov process. The weak-semi-Markov model is shown to model partitioned dynamic systems with greater economy than other surveyed models. The behaviour of a partitioned state space system in the area around the region boundaries is also considered. I use the theory of sliding surfaces, and some heuristic arguments to recommend region boundary shape and position. The concept of 'staying on the boundary' then becomes a robust and relatively easy subgoal within the control algorithm. The concept of 'reaching the sliding surface' as a subgoal is used as the basis for an intuitive explanation of the learnt controller. I present an algorithm based on this concept which explains the behaviour of a learnt controller in ways not previously available to a machine learning algorithms. Finally, the Markov Property and the theory of Sliding Mode Control are used as the basis of a class of recursive algorithms. These algorithms adaptively find a partition, and simultaneously use this partition in conjunction with one of five reinforcement learning algorithms to find a control policy based on that partition. This technique is shown to work very well in learning, controlling and explaining a variety of physical systems, from a monorail to a container crane.

Network reconstruction is the process of recovering a unique structured representation of some dynamic system using input-output data and some additional knowledge about the structure of the system. Many network reconstruction algorithms have been proposed in recent years, most dealing with the reconstruction of strictly proper networks (i.e., networks that require delays in all dynamics between measured variables). However, no reconstruction technique presently exists capable of recovering both the structure and dynamics of networks where links are proper (delays in dynamics are not required) and not necessarily strictly proper.The ultimate objective of this dissertation is to develop algorithms capable of reconstructing proper networks, and this objective will be addressed in three parts. The first part lays the foundation for the theory of mathematical representations of proper networks, including an exposition on when such networks are well-posed (i.e., physically realizable). The second part studies the notions of abstractions of a network, which are other networks that preserve certain properties of the original network but contain less structural information. As such, abstractions require less a priori information to reconstruct from data than the original network, which allows previously-unsolvable problems to become solvable. The third part addresses our original objective and presents reconstruction algorithms to recover proper networks in both the time domain and in the frequency domain.

Thesis (S.M.)--Massachusetts Institute of Technology, School of Architecture and Planning, Program in Media Arts and Sciences, 2006.
Includes bibliographical references (p. 58-60).
Graphs are often used in artificial intelligence as means for symbolic knowledge representation. A graph is nothing more than a collection of symbols connected to each other in some fashion. For example, in computer vision a graph with five nodes and some edges can represent a table - where nodes correspond to particular shape descriptors for legs and a top, and edges to particular spatial relations. As a framework for representation, graphs invite us to simplify and view the world as objects of pure structure whose properties are fixed in time, while the phenomena they are supposed to model are actually often changing. A node alone cannot represent a table leg, for example, because a table leg is not one structure (it can have many different shapes, colors, or it can be seen in many different settings, lighting conditions, etc.) Theories of knowledge representation have in general concentrated on the stability of symbols - on the fact that people often use properties that remain unchanged across different contexts to represent an object (in vision, these properties are called invariants). However, on closer inspection, objects are variable as well as stable. How are we to understand such problems? How is that assembling a large collection of changing components into a system results in something that is an altogether stable collection of parts?
(cont.) The work here presents one approach that we came to encompass by the phrase "graph dynamics". Roughly speaking, dynamical systems are systems with states that evolve over time according to some lawful "motion". In graph dynamics, states are graphical structures, corresponding to different hypothesis for representation, and motion is the correction or repair of an antecedent structure. The adapted structure is an end product on a path of test and repair. In this way, a graph is not an exact record of the environment but a malleable construct that is gradually tightened to fit the form it is to reproduce. In particular, we explore the concept of attractors for the graph dynamical system. In dynamical systems theory, attractor states are states into which the system settles with the passage of time, and in graph dynamics they correspond to graphical states with many repairs (states that can cope with many different contingencies). In parallel with introducing the basic mathematical framework for graph dynamics, we define a game for its control, its attractor states and a method to find the attractors. From these insights, we work out two new algorithms, one for Bayesian network discovery and one for active learning, which in combination we use to undertake the object recognition problem in computer vision. To conclude, we report competitive results in standard and custom-made object recognition datasets.
by Andre Figueiredo Ribeiro.
S.M.

La démocratisation récente et massive des outils numériques a donné à tous le moyen de produire de l'information et de la partager sur le web, que ce soit à travers des blogs, des réseaux sociaux, des plateformes de partage, ... La croissance exponentielle de cette masse d'information disponible, en grande partie textuelle, nécessite le développement de modèles de traitement automatique du langage naturel (TAL), afin de la représenter mathématiquement pour ensuite la classer, la trier ou la recommander. C'est l'apprentissage de représentation. Il vise à construire un espace de faible dimension où les distances entre les objets projetées (mots, textes) reflètent les distances constatées dans le monde réel, qu'elles soient sémantique, stylistique, ...La multiplication des données disponibles, combinée à l'explosion des moyens de calculs et l'essor de l'apprentissage profond à permis de créer des modèles de langue extrêmement performant pour le plongement des mots et des documents. Ils assimilent des notions sémantiques et de langue complexes, en restant accessibles à tous et facilement spécialisables sur des tâches ou des corpus plus spécifiques. Il est possible de les utiliser pour construire des plongements d'auteurices. Seulement il est difficile de savoir sur quels aspects un modèle va se focaliser pour les rapprocher ou les éloigner. Dans un cadre littéraire, il serait préférable que les similarités se rapportent principalement au style écrit. Plusieurs problèmes se posent alors. La définition du style littéraire est floue, il est difficile d'évaluer l'écart stylistique entre deux textes et donc entre leurs plongements. En linguistique computationnelle, les approches visant à le caractériser sont principalement statistiques, s'appuyant sur des marqueurs du langage. Fort de ces constats, notre première contribution propose une méthode d'évaluation de la capacité des modèles de langue à appréhender le style écrit. Nous aurons au préalable détaillé comment le texte est représenté en apprentissage automatique puis en apprentissage profond, au niveau du mot, du document puis des auteurices. Nous aurons aussi présenté le traitement de la notion de style littéraire en TAL, base de notre méthode. Le transfert de connaissances entre les boîtes noires que sont les grands modèles de langue et ces méthodes issues de la linguistique n'en demeure pas moins complexe. Notre seconde contribution vise à réconcilier ces approches via un modèle d'apprentissage de représentations d'auteurices se focalisant sur le style, VADES (Variational Author and Document Embedding with Style). Nous nous comparons aux méthodes existantes et analysons leurs limites dans cette optique-là. Enfin, nous nous intéressons à l'apprentissage de plongements dynamiques d'auteurices et de documents. En effet, l'information temporelle est cruciale et permet une représentation plus fine des dynamiques d'écriture. Après une présentation de l'état de l'art, nous détaillons notre dernière contribution, B²ADE (Brownian Bridge for Author and Document Embedding), modélisant les auteurices comme des trajectoires. Nous finissons en décrivant plusieurs axes d'améliorations de nos méthodes ainsi que quelques problématiques pour de futurs travaux
The recent and massive democratization of digital tools has empowered individuals to generate and share information on the web through various means such as blogs, social networks, sharing platforms, and more. The exponential growth of available information, mostly textual data, requires the development of Natural Language Processing (NLP) models to mathematically represent it and subsequently classify, sort, or recommend it. This is the essence of representation learning. It aims to construct a low-dimensional space where the distances between projected objects (words, texts) reflect real-world distances, whether semantic, stylistic, and so on.The proliferation of available data, coupled with the rise in computing power and deep learning, has led to the creation of highly effective language models for word and document embeddings. These models incorporate complex semantic and linguistic concepts while remaining accessible to everyone and easily adaptable to specific tasks or corpora. One can use them to create author embeddings. However, it is challenging to determine the aspects on which a model will focus to bring authors closer or move them apart. In a literary context, it is preferable for similarities to primarily relate to writing style, which raises several issues. The definition of literary style is vague, assessing the stylistic difference between two texts and their embeddings is complex. In computational linguistics, approaches aiming to characterize it are mainly statistical, relying on language markers. In light of this, our first contribution is a framework to evaluate the ability of language models to grasp writing style. We will have previously elaborated on text embedding models in machine learning and deep learning, at the word, document, and author levels. We will also have presented the treatment of the notion of literary style in Natural Language Processing, which forms the basis of our method. Transferring knowledge between black-box large language models and these methods derived from linguistics remains a complex task. Our second contribution aims to reconcile these approaches through a representation learning model focusing on style, VADES (Variational Author and Document Embedding with Style). We compare our model to state-of-the-art ones and analyze their limitations in this context.Finally, we delve into dynamic author and document embeddings. Temporal information is crucial, allowing for a more fine-grained representation of writing dynamics. After presenting the state of the art, we elaborate on our last contribution, B²ADE (Brownian Bridge Author and Document Embedding), which models authors as trajectories. We conclude by outlining several leads for improving our methods and highlighting potential research directions for the future

The overarching aim of this thesis was to investigate how processes of perception and action emerge under changing informational constraints during performance of multi-articular interceptive actions. Interceptive actions provide unique opportunities to study processes of perception and action in dynamic performance environments. The movement model used to exemplify the functionally coupled relationship between perception and action, from an ecological dynamics perspective, was cricket batting. Ecological dynamics conceptualises the human body as a complex system composed of many interacting sub-systems, and perceptual and motor system degrees of freedom, which leads to the emergence of patterns of behaviour under changing task constraints during performance. The series of studies reported in the Chapters of this doctoral thesis contributed to understanding of human behaviour by providing evidence of key properties of complex systems in human movement systems including self-organisation under constraints and meta-stability. Specifically, the studies: i) demonstrated how movement organisation (action) and visual strategies (perception) of dynamic human behaviour are constrained by changing ecological (especially informational) task constraints; (ii) provided evidence for the importance of representative design in experiments on perception and action; and iii), provided a principled theoretical framework to guide learning design in acquisition of skill in interceptive actions like cricket batting.

Published Article
This article reviews research on learning with external representations and provides a theoretical background on how to optimize learning from external representations. General factors, such as the type of material to be learned, learner characteristics and the testing method, are some of the variables that can determine if graphic medium can increase a subject's comprehension and if such comprehension can be accurately measured. These factors are discussed and represented by a model to suggest how external representations can be effectively used in a learning environment. Two key conclusions are drawn from the observation made in these studies. Firstly, the proper design of a particular external representation and supporting text can promote relevant activities that ultimately contribute to fuller understanding of the content. Secondly, external representations must be developed to address the size complexity and variety of the content that must be analysed in order to extract knowledge for scientific discovery.

Les données temporelles constituent une partie importante des données digitales. Prévoir leurs prochaines valeurs est une tâche importante et difficile. Les méthodes statistiques standard sont fondées sur des modèles linéaires souvent limitées aux données de faible dimension. Ici, nous utilisons plutôt des méthodes d'apprentissage profond capables de traiter des données structurées en haute dimension. Dans cette thèse, nous nous intéressons aux modèles à variables latentes. Contrairement aux modèles auto-régressifs qui utilisent directement des données passées pour effectuer des prédictions, les modèles latents infèrent des représentations vectorielles qui sont ensuite prédites. Nous proposons d'abord un modèle latent structuré pour la prévision de données spatio-temporelles. Des variables latentes sont inférés à partir d'un ensemble de points dans l'espace où des données sont collectées (météo, trafic). Ensuite, la structure spatiale est utilisée dans la fonction dynamique. Le modèle est également capable de découvrir des corrélations entre des séries sans information spatiale préalable. Ensuite, nous nous intéressons à la prédiction des distributions de données. Nous proposons un modèle qui génère des variables latentes utilisées pour conditionner un modèle génératif. Les données textuelles sont utilisées pour évaluer le modèle sur la modélisation diachronique du langage. Enfin, nous proposons un modèle de prédiction stochastique. Il utilise les premières valeurs des séquences pour générer plusieurs futurs possibles. Ici, le modèle génératif n'est pas conditionné à une époque absolue, mais à une séquence. Le modèle est appliqué à la prédiction vidéo stochastique
Temporal data constitute a large part of data collected digitally. Predicting their next values is an important and challenging task in domains such as climatology, optimal control, or natural language processing. Standard statistical methods are based on linear models and are often limited to low dimensional data. We instead use deep learning methods capable of handling high dimensional structured data and leverage large quantities of examples. In this thesis, we are interested in latent variable models. Contrary to autoregressive models that directly use past data to perform prediction, latent models infer low dimensional vectorial representations of data on which prediction is performed. Latent vectorial spaces allow us to learn dynamic models that are able to generate high-dimensional and structured data. First, we propose a structured latent model for spatio-temporal data forecasting. Given a set of spatial locations where data such as weather or traffic are collected, we infer latent variables for each location and use spatial structure in the dynamic function. The model is also able to discover correlations between series without prior spatial information. Next, we focus on predicting data distributions, rather than point estimates. We propose a model that generates latent variables used to condition a generative model. Text data are used to evaluate the model on diachronic language modeling. Finally, we propose a stochastic prediction model. It uses the first values of sequences to generate several possible futures. Here, the generative model is not conditioned to an absolute epoch, but to a sequence. The model is applied to stochastic video prediction

This thesis details a series of studies investigating long jump locomotor pointing tasks through a theoretical framework of ecological dynamics. A multi-methods approach was used to investigate athlete behaviours in competition environments revealing the complexities of field-based locomotor pointing tasks, providing important information for the design of practice environments for practitioners. The development of a new substantive theoretical model of perform, respond and manage, presents as a novel tool for practitioners in understanding emergent behaviour in long jump and can be used to better frame representative learning designs. Importantly, athletes need to prepare for more than just a technical performance.

This work presents an investigation of tailoring Network Representation Learning (NRL) for an application in the Financial Industry. NRL approaches are data-driven models that learn how to encode graph structures into low-dimensional vector spaces, which can be further exploited by downstream Machine Learning applications. They can potentially bring a lot of benefits in the Financial Industry since they extract in an automatic way features that can provide useful input regarding graph structures, called embeddings. Financial transactions can be represented as a network, and through NRL, it is possible to extract embeddings that reflect the intrinsic inter-connected nature of economic relationships. Such embeddings can be used for several purposes, among which Anomaly Detection to fight financial crime.This work provides a qualitative analysis over state-of-the-art NRL models, which identifies Graph Convolutional Network (ConvGNN) as the most suitable category of approaches for Financial Industry but with a certain need for further improvement. Financial Industry poses additional challenges when modelling a NRL solution. Despite the need of having a scalable solution to handle real-world graph with considerable dimensions, it is necessary to take into consideration several characteristics: transactions graphs are inherently dynamic since every day new transactions are executed and nodes can be heterogeneous. Besides, everything is further complicated by the need to have updated information in (near) real-time due to the sensitivity of the application domain. For these reasons, GraphSAGE has been considered as a base for the experiments, which is an inductive ConvGNN model. Two variants of GraphSAGE are presented: a dynamic variant whose weights evolve accordingly with the input sequence of graph snapshots, and a variant specifically meant to handle bipartite graphs. These variants have been evaluated by applying them to real-world data and leveraging the generated embeddings to perform Anomaly Detection. The experiments demonstrate that leveraging these variants leads toimagecomparable results with other state-of-the-art approaches, but having the advantage of being suitable to handle real-world financial data sets.
Detta arbete presenterar en undersökning av tillämpningar av Network Representation Learning (NRL) inom den finansiella industrin. Metoder inom NRL möjliggör datadriven kondensering av grafstrukturer till lågdimensionella och lätthanterliga vektorer.Dessa vektorer kan sedan användas i andra maskininlärningsuppgifter. Närmare bestämt, kan metoder inom NRL underlätta hantering av och informantionsutvinning ur beräkningsintensiva och storskaliga grafer inom den finansiella sektorn, till exempel avvikelsehantering bland finansiella transaktioner. Arbetet med data av denna typ försvåras av det faktum att transaktionsgrafer är dynamiska och i konstant förändring. Utöver detta kan noderna, dvs transaktionspunkterna, vara vitt skilda eller med andra ord härstamma från olika fördelningar.I detta arbete har Graph Convolutional Network (ConvGNN) ansetts till den mest lämpliga lösningen för nämnda tillämpningar riktade mot upptäckt av avvikelser i transaktioner. GraphSAGE har använts som utgångspunkt för experimenten i två olika varianter: en dynamisk version där vikterna uppdateras allteftersom nya transaktionssekvenser matas in, och en variant avsedd särskilt för bipartita (tvådelade) grafer. Dessa varianter har utvärderats genom användning av faktiska datamängder med avvikelsehantering som slutmål.

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

This thesis presents data supporting a re-definition of the concept of learning and transfer of skills in physical activity and sport performance contexts. The key findings indicate that during practice, environmental designs that promote exploratory behaviour facilitate improved performance in modified conditions.

The overarching aim of this programme of work was to evaluate the effectiveness of the existing learning environment within the Australian Institute of Sport (AIS) elite springboard diving programme. Unique to the current research programme, is the application of ideas from an established theory of motor learning, specifically ecological dynamics, to an applied high performance training environment. In this research programme springboard diving is examined as a complex system, where individual, task, and environmental constraints are continually interacting to shape performance. As a consequence, this thesis presents some necessary and unique insights into representative learning design and movement adaptations in a sample of elite athletes. The questions examined in this programme of work relate to how best to structure practice, which is central to developing an effective learning environment in a high performance setting. Specifically, the series of studies reported in the chapters of this doctoral thesis: (i) provide evidence for the importance of designing representative practice tasks in training; (ii) establish that completed and baulked (prematurely terminated) take-offs are not different enough to justify the abortion of a planned dive; and (iii), confirm that elite athletes performing complex skills are able to adapt their movement patterns to achieve consistent performance outcomes from variable dive take-off conditions. Chapters One and Two of the thesis provide an overview of the theoretical ideas framing the programme of work, and include a review of literature pertinent to the research aims and subsequent empirical chapters. Chapter Three examined the representativeness of take-off tasks completed in the two AIS diving training facilities routinely used in springboard diving. Results highlighted differences in the preparatory phase of reverse dive take-offs completed by elite divers during normal training tasks in the dry-land and aquatic training environments. The most noticeable differences in dive take-off between environments began during the hurdle (step, jump, height and flight) where the diver generates the necessary momentum to complete the dive. Consequently, greater step lengths, jump heights and flight times, resulted in greater board depression prior to take-off in the aquatic environment where the dives required greater amounts of rotation. The differences observed between the preparatory phases of reverse dive take-offs completed in the dry-land and aquatic training environments are arguably a consequence of the constraints of the training environment. Specifically, differences in the environmental information available to the athletes, and the need to alter the landing (feet first vs. wrist first landing) from the take-off, resulted in a decoupling of important perception and action information and a decomposition of the dive take-off task. In attempting to only practise high quality dives, many athletes have followed a traditional motor learning approach (Schmidt, 1975) and tried to eliminate take-off variations during training. Chapter Four examined whether observable differences existed between the movement kinematics of elite divers in the preparation phases of baulked (prematurely terminated) and completed take-offs that might justify this approach to training. Qualitative and quantitative analyses of variability within conditions revealed greater consistency and less variability when dives were completed, and greater variability amongst baulked take-offs for all participants. Based on these findings, it is probable that athletes choose to abort a planned take-off when they detect small variations from the movement patterns (e.g., step lengths, jump height, springboard depression) of highly practiced comfortable dives. However, with no major differences in coordination patterns (topology of the angle-angle plots), and the potential for negative performance outcomes in competition, there appears to be no training advantage in baulking on unsatisfactory take-offs during training, except when a threat of injury is perceived by the athlete. Instead, it was considered that enhancing the athletes' movement adaptability would be a more functional motor learning strategy. In Chapter Five, a twelve-week training programme was conducted to determine whether a sample of elite divers were able to adapt their movement patterns and complete dives successfully, regardless of the perceived quality of their preparatory movements on the springboard. The data indeed suggested that elite divers were able to adapt their movements during the preparatory phase of the take-off and complete good quality dives under more varied take-off conditions; displaying greater consistency and stability in the key performance outcome (dive entry). These findings are in line with previous research findings from other sports (e.g., shooting, triple jump and basketball) and demonstrate how functional or compensatory movement variability can afford greater flexibility in task execution. By previously only practising dives with good quality take-offs, it can be argued that divers only developed strong couplings between information and movement under very specific performance circumstances. As a result, this sample was sometimes characterised by poor performance in competition when the athletes experienced a suboptimal take-off. Throughout this training programme, where divers were encouraged to minimise baulking and attempt to complete every dive, they demonstrated that it was possible to strengthen the information and movement coupling in a variety of performance circumstances, widening of the basin of performance solutions and providing alternative couplings to solve a performance problem even when the take-off was not ideal. The results of this programme of research provide theoretical and experimental implications for understanding representative learning design and movement pattern variability in applied sports science research. Theoretically, this PhD programme contributes empirical evidence to demonstrate the importance of representative design in the training environments of high performance sports programmes. Specifically, this thesis advocates for the design of learning environments that effectively capture and enhance functional and flexible movement responses representative of performance contexts. Further, data from this thesis showed that elite athletes performing complex tasks were able to adapt their movements in the preparatory phase and complete good quality dives under more varied take-off conditions. This finding signals some significant practical implications for athletes, coaches and sports scientists. As such, it is recommended that care should be taken by coaches when designing practice tasks since the clear implication is that athletes need to practice adapting movement patterns during ongoing regulation of multi-articular coordination tasks. For example, volleyball servers can adapt to small variations in the ball toss phase, long jumpers can visually regulate gait as they prepare for the take-off, and springboard divers need to continue to practice adapting their take-off from the hurdle step. In summary, the studies of this programme of work have confirmed that the task constraints of training environments in elite sport performance programmes need to provide a faithful simulation of a competitive performance environment in order that performance outcomes may be stabilised with practice. Further, it is apparent that training environments can be enhanced by ensuring the representative design of task constraints, which have high action fidelity with the performance context. Ultimately, this study recommends that the traditional coaching adage 'perfect practice makes perfect", be reconsidered; instead advocating that practice should be, as Bernstein (1967) suggested, "repetition without repetition".

This thesis proposes a general approach to solve the offline flight-maneuver identification problem using machine learning methods. The purpose of the study was to provide means for the aircraft professionals at the flight test and verification department of Saab Aeronautics to automate the procedure of analyzing flight test data. The suggested approach succeeded in generating binary classifiers and multiclass classifiers that identified six flight maneuvers of different complexity from real flight test data. The binary classifiers solved the problem of identifying one maneuver from flight test data at a time, while the multiclass classifiers solved the problem of identifying several maneuvers from flight test data simultaneously. To achieve these results, the difficulties that this time series classification problem entailed were simplified by using different strategies. One strategy was to develop a maneuver extraction algorithm that used handcrafted rules. Another strategy was to represent the time series data by statistical measures. There was also an issue of an imbalanced dataset, where one class far outweighed others in number of samples. This was solved by using a modified oversampling method on the dataset that was used for training. Logistic Regression, Support Vector Machines with both linear and nonlinear kernels, and Artifical Neural Networks were explored, where the hyperparameters for each machine learning algorithm were chosen during model estimation by 4-fold cross-validation and solving an optimization problem based on important performance metrics. A feature selection algorithm was also used during model estimation to evaluate how the performance changes depending on how many features were used. The machine learning models were then evaluated on test data consisting of 24 flight tests. The results given by the test data set showed that the simplifications done were reasonable, but the maneuver extraction algorithm could sometimes fail. Some maneuvers were easier to identify than others and the linear machine learning models resulted in a poor fit to the more complex classes. In conclusion, both binary classifiers and multiclass classifiers could be used to solve the flight maneuver identification problem, and solving a hyperparameter optimization problem boosted the performance of the finalized models. Nonlinear classifiers performed the best on average across all explored maneuvers.

This study attempted to combine the benefits of multimedia learning, adaptive interfaces, and learning style theory by constructing a novel e-learning environment. The environment was designed to accommodate individual learning styles while students progressed through a computer programming course. Despite the benefits of individualised instruction and a growing worldwide e-learning market, there is a paucity of guidance on how to effectively accommodate learning styles in an online environment. Several existing learning-style adaptive environments base their behaviour on an initial assessment of the learner's profile, which is then assumed to remain stable. Consequently, these environments rarely offer the learner choices between different versions of content. However, these choices could cater for flexible learning styles, promote cognitive flexibility, and increase learner control. The first research question underlying the project asked how learning styles could be accommodated in an adaptive e-learning environment. The second question asked whether a dynamically adaptive environment that provides the learner with a choice of media experiences is more beneficial than a statically adapted environment. To answer these questions, an adaptive e-learning environment named iWeaver was created and experimentally evaluated. iWeaver was based on an introductory course in Java programming and offered learning content as style-specific media experiences, assisted by additional learning tools. These experiences and tools were based on the perceptual and information processing dimension of an adapted version of the Dunn and Dunn learning styles model. An experimental evaluation of iWeaver was conducted with 63 multimedia students. The analysis investigated the effect of having a choice of multiple media experiences (compared to having just one static media experience) on learning gain, enjoyment, perceived progress, and motivation. In addition to these quantitative measurements, learners provided qualitative feedback at the end of each lesson. Data from 27 participants were sufficiently complete to be analysed. For the data analysis, participants were divided into two groups of high and low interest in programming and Java, then into two groups of high and low experience with computers and the Internet. Both group comparisons revealed statistically significant differences for the effect of choice. Having a choice of media experiences proved beneficial for learners with low experience but detrimental for learners with high experience or interest. These findings suggest that the effect of choice appears to be strongly influenced by the learner's background. It is hypothesised that encouraging a more active learner role in educational systems would expand the positive influence of choice to a wider range of learners. The study has contributed some weight to the argument that for certain groups of learners, it is more beneficial to view learning style as a flexible, rather than a stable construct. As a practical implication, it seems advisable to collect data on prior experience, interest, and the initial learning style distribution of the target audience before developing environments comparable to iWeaver. [See http://www.adaptive-learning.net/research/media.htm for media files associated with this thesis.]

The neural net computer simulations which will be presented here are based on the acceptance of a set of assumptions that for the last twenty years have been expressed in the fields of information processing, neurophysiology and cognitive sciences. First of all, neural networks and their dynamical behaviors in terms of attractors is the natural way adopted by the brain to encode information. Any information item to be stored in the neural net should be coded in some way or another in one of the dynamical attractors of the brain and retrieved by stimulating the net so as to trap its dynamics in the desired item's basin of attraction. The second view shared by neural net researchers is to base the learning of the synaptic matrix on a local Hebbian mechanism. The last assumption is the presence of chaos and the benefit gained by its presence. Chaos, although very simply produced, inherently possesses an infinite amount of cyclic regimes that can be exploited for coding information. Moreover, the network randomly wanders around these unstable regimes in a spontaneous way, thus rapidly proposing alternative responses to external stimuli and being able to easily switch from one of these potential attractors to another in response to any coming stimulus.

In this thesis, it is shown experimentally that the more information is to be stored in robust cyclic attractors, the more chaos appears as a regime in the back, erratically itinerating among brief appearances of these attractors. Chaos does not appear to be the cause but the consequence of the learning. However, it appears as an helpful consequence that widens the net's encoding capacity. To learn the information to be stored, an unsupervised Hebbian learning algorithm is introduced. By leaving the semantics of the attractors to be associated with the feeding data unprescribed, promising results have been obtained in term of storing capacity.
Doctorat en sciences appliquées
info:eu-repo/semantics/nonPublished

Orientador: Antonio Carlos Zambon
Dissertação (mestrado) - Universidade Estadual de Campinas, Faculdade de Tecnologia
Made available in DSpace on 2018-08-19T23:02:19Z (GMT). No. of bitstreams: 1 Costa_FabianaMarques_M.pdf: 2422520 bytes, checksum: 966c5b82f59168e9a1400b9b58760301 (MD5) Previous issue date: 2012
Resumo: Esta pesquisa fundamenta-se no desenvolvimento de um método de aquisição de conhecimento de agentes textuais baseada em MORPH - Modelo Orientado à Representação do Pensamento Humano - que permite que se extraia o modelo mental de agentes textuais. O objetivo é evidenciar o conhecimento contido no agente textual, representá-lo graficamente para compreendê-lo, facilitando o processo de aprendizagem e refinando o estudo dos conteúdos de um texto. Pois considera-se que nem sempre autores deixam as ideias explícitas (suas estruturas mentais) em artigos científicos, de forma clara e objetiva. O MACAT é um processo composto por três etapas, estruturadas em diretrizes para a extração de objetos de agentes textuais diversos. Apresenta-se além do desenvolvimento do método, a aplicação do MACAT baseado em MORPH, para investigação de artigos científicos, visando à exemplificação de sua utilização e demonstrando sua utilidade na explicitação de conhecimento. Com isso, é póssível evidenciar a dinâmica dos processos contidos nos sistemas organizacionais, que apresentam dificuldades de construir o aprendizado, em razão da ausência de instrumentos pelos quais se possa avaliar a progressão do conhecimento. Como resultado, demonstra-se que o método torna possível a extração e representação do conhecimento de agentes humanos externalizados em agentes textuais, permitindo a compreensão de modelos mentais, alavancando a tomada de decisão em situações complexas
Abstract: This research is based on developing on a method for the Knowledge Acquisition of Textual Agents based on MORPH - Oriented Model to the Human Thought Representation - which allows you to extraction of a textual agent's mental model. The goal is to demonstrate the knowledge present in textual agent, representing it graphically by facilitating its understanding and the learning process and refining the study of the contents of a text. Because it is considered that the authors don't always make explicit ideas (mental structures) of scientific articles, clearly and objectively. The MACAT is a process composed of three steps, structured guidelines for the extraction of objects of various textual agents. It is presented in addition to method development, the application of MACAT based on MORPH for research papers, aimed at the exemplification of its use and demonstrating its usefulness in explicit knowledge. This makes it possible to demonstrate the dynamics organizational processes in computer systems in those which have difficulty in learning to build, due to the lack of instruments that can evaluate the evolution of knowledge. As a result, it is shown that the method makes possible the extraction and representation of knowledge into human agents that externalized in textual agents, able to understanding the mental models, leveraging the decision-making in complex situations
Mestrado
Tecnologia e Inovação
Mestre em Tecnologia

碩士
國立彰化師範大學
數學系所
98
This study aims to develop a learning module, that learners studied by themselves. This module designed by multiple representation、dynamic assessment and precept theory and used the interaction of multimedia computer and the properties of multiple representation. This module not only supports the learning environment with images and voice for learners, and promptly gives learners the feedback by interactive technology. The feedback contains different degree of guidance and explaining the puzzles. By these ways develop the learners’ procedural and conceptual knowledge for square root. Except for supporting feedback, this module will record of the learners’ learning process. Researchers can analyze the process to understand the learners’ learning state and improve the design of this module. The foundation of this module is dynamic assessment to design interactive function and feedback. And the foundation of the feedback is multiple representation. This module cut four parts by precept theory. By analyzing the learners’ process, 30% of 7th grade students could find out the square root approximation before using this module. After using this module, 52% of 7th grade students could find out square root approximation by themselves. Through the module taught the applied problem of square root approximation, 40.4% of 7th grade students could find out cube root approximation.

碩士
國立交通大學
理學院科技與數位學習學程
100
This study aimed to explore influence on outcomes of scientific conceptual construction through teaching principles of wave superposition with “dynamic representation” or “static representation”.To achieve the purpose of research, a quasi-experimental design was conduced, high school students in the Northen Taiwan. Dynamic or static representation was incorporated into a design of instruction on wave superposition. Students in the experimental group Ⅰ received instruction incorporating dynamic representations. Students in the experimental group Ⅱreceived instruction incorporating static representations. Students’ performance on a conceptual diagnostic exam and on their fearning worksheek were analyzed to determine effectiveness dynamic and static representation on assisting high school students’ learning concept of wave superposition. The results showed that students who received instruction with dynamic representations outperformed those who received instruction with static representations on their post and postpond test’s scores of conceptual diagnostic exams well as on their scores of learning worksheet. That the use of dynamic characterization of the concept of teaching can help students learning and retention enhance the effectiveness. Another record for the effectiveness of the learning process analysis, dynamic characterization results presented to accept the teaching of learners are also better teaching and learning, and significant differences . This study suggest that instructions incorporating dynamic representation is more effective in assisting learning of wave superposition, in comparison to instructions combining with static representation.

碩士
國立交通大學
理學院科技與數位學習學程
103
This purpose of this study is to explore the influence of “dynamic representation”or“static representation”on teaching.By supplementing or without sequential scaffolding problem to discover the influence on studnets’learning performance and concept profile. To achieve the purpose of research, quasi-experimental design was conduced. Some first-grade students in high school in central Taiwan are divided into three groups. And those students were given three lessons as a teaching experiment.Three groups were given different experiemental design individually.The experiemental design of dynamic and sequential scaffolding problem was given in first group. The experiemental design of dynamic representation and non-sequential scaffolding problem was given in the second group. Eventually, the experiemental design of static graphs (captured from dynamic representation) and sequential scaffolding problem was given in the last group. The result reveals that,no matter with or without sequential scaffolding problem,the learning process in the experimental design with dynamic representation is better than the group in experimental design with static representation and sequential scaffolding problem.Nevertheless, after teaching, compared to the other two groups, the group in the experimental design with dynamic and sequential scaffolding problem has better performance at Projectile Motion concept posttest. This outcome shows that if teaching process is supplemented with dynamic representation it will contribute to conceptual learning. Furthemore,when the experimental design with dynamic representation is supplemented with sequential scaffolding problem, the effectiveness of learning will be able to be continue to posttest performance of Projectile Motion concept. Analysis result reveals that the reply of difficult question situations during each group is insignificant. Most of learners link to inappropriate concept profile only by the single feature of problem, or learners do not have science-based concept profile.

Increasingly, researchers and developers of knowledge based systems (KBS) have been attempting to incorporate the notion of context. For instance, Repertory Grids, Formal Concept Analysis (FCA) and Ripple-Down Rules (RDR) all integrate either implicit or explicit contextual information. However, these methodologies treat context as a static entity, neglecting many connectionists’ work in learning hidden and dynamic contexts. This thesis argues that the omission of these higher forms of context, which allow connectionist systems to generalise effectively, is one of the fundamental problems in the application and interpretation of symbolic knowledge. This thesis tackles the problems of KBSs by addressing these contextual inadequacies over a three stage approach: philosophically, methodologically and through the application of prudence analysis. Firstly, it challenges existing notions of knowledge by introducing a new philosophical view referred to as Intermediate Situation Cognition. This new position builds on the existing SC premise, that knowledge and memory is re-constructed at the moment required, by allowing for the inclusion of hidden and dynamic contexts in symbolic reasoning. This philosophical position has been incorporated into the development of a hybridised methodology, combining Multiple Classification Ripple-Down Rules (MCRDR) with a function-fitting technique. This approach, referred to as Rated MCRDR (RM), retains a symbolic core acting as a contextually static memory, while using a connection based approach to learn a deeper understanding of the knowledge captured. This analysis of the knowledge map is performed dynamically, providing constant online information. Results indicate that the method developed can learn the information that experts have difficulty providing. This supplies the information required to allow for generalisation of the knowledge captured. In order to show that hidden and dynamic contextual information can improve the robustness of a KBS, RM must reduce brittleness. Brittleness, which is widely recognised as the primary impediment in KBS performance, is caused by a system’s inability to realise when its knowledge base is inadequate for a particular situation. RM partly addresses this through providing better generalisation; however, brittleness can be more directly addressed by detecting when such inadequacies occur. This process is commonly referred to as prudence analysis. The final part of this thesis proves the methods philosophical and methodological approach by illustrating how RM’s use of hidden and dynamic contextual information, allows the system to perform this analysis. Results show how experts can confidently leave the verification of cases when not warned, reducing brittleness and the knowledge acquisition effort. This thesis shows that the idea of incorporating higher forms of context in symbolic reasoning domains is both possible and highly effective, vastly improving the robustness of the KBS approach. Not only does this facilitate improved classification through better generalisation, but also reduces the KA effort required by experts. Additionally, the methodology developed has further potential for many possible applications across numerous domains, such as Information Filtering, Data Mining, incremental induction and even reinforcement learning.

碩士
國立臺北科技大學
技術及職業教育研究所
102
This study used dynamic representation on four arithmetic operations of integer for fourth grade students. There were two main purpose of this study. One was developing a dynamic representation design teaching material. The other was exploring dynamic representation’s influence on different levels students’ learning outcomes and cognitive load. The research method was a quasi-experimental research design. There were 50 fourth grade students selected from one elementary school to participate four lessons long experiment. One of the two classes was assigned as the experimental group, which uses dynamic representation design, and the other as the comparison group, which uses a design without dynamic representation. The results showed that using dynamic representation could effectively improve students’ learning outcomes, and reduce extraneous cognitive load.

The Convolutional Neural Network (CNN) have shown a substantial improvement in the field of Machine Learning. But they do come with their own set of drawbacks. Capsule Networks have addressed the limitations of CNNs and have shown a great improvement by calculating the pose and transformation of the image. Deeper networks are more powerful than shallow networks but at the same time, more difficult to train. Residual Networks ease the training and have shown evidence that they can give good accuracy with considerable depth. Putting the best of Capsule Network and Residual Network together, we present Residual Capsule Network and 3-Level Residual Capsule Network, a framework that uses the best of Residual Networks and Capsule Networks. The conventional Convolutional layer in Capsule Network is replaced by skip connections like the Residual Networks to decrease the complexity of the Baseline Capsule Network and seven ensemble Capsule Network. We trained our models on MNIST and CIFAR-10 datasets and have seen a significant decrease in the number of parameters when compared to the Baseline models.

碩士
國立臺灣師範大學
數學研究所
88
This study initially investigated the learning processes of high-school students and how they represent elliptical concepts and interlink them. The results were used to develop computer based instructional tools. The way in which these tools affected the student’s representations was then studied. The research was in two parts: Initially nine student volunteers participated in the study; each was given diagnostic teaching. It was found that the main external representations of elliptical concepts are verbal, graphical, structural, formulaic, and locus representation. Students were found to have different ways to develop their elliptical concepts, according to their initial knowledge. When students could represent the problem and their related ideas in terms of graphical representation, then they could develop links between different representations, and could thus solve problems more easily. In the second part of the research two high school classes were compared. A computer-based dynamic linked multiple representation Windows learning environment was used to teach one class. The results showed that students could make significant progress when taught in this way. The progress scores of the experimental group were higher than the control group, but this was not statistically significant. It was found that the experimental group tended to employ graphical representation as their main representation and use it to integrate the other representations. This paper analyses high school instruction for teaching elliptical concepts, looking in particular at the process by which students construct multiple elliptical representations. It also gives insights into the design of computer-based dynamic linked multiple representation Windows learning environments

Programa Doutoral em Engenharia Eletrónica e de Computadores
Due to the increasing demand for adaptive robots able to assist humans in their everyday tasks, furnishing robots with learning abilities is one of the most important goals of current robotics research. The work reported in this thesis is focused on the integration of learning capacities in an existing Dynamic Field based control architecture developed for natural human-robot collaboration. Specifically, it addresses two important serial order problems that appear in the architecture at distinct but closely coupled levels of abstraction: 1) the learning of the sequential order of sub-goals that has to be followed to accomplish a certain task, and 2) the learning of representations of motor primitives that can be chained to achieve a certain sub-goal. A model based on the theoretical framework of Dynamic Neural Fields (DNFs) is developed that allows the robot to acquire a multi-order sequential plan of a task from demonstration by human tutors. The model is inspired by known processing principles of human serial order learning. Specifically, it implements the idea of two complementary learning systems. A fast system encodes the sequential order of a single demonstration. During periods of internal rehearsal, it acts as a teacher for a slow system that is responsible for extracting generalized task knowledge from memorized demonstrations of different users. The efficiency of the learning model is tested in a real world experiment in which the humanoid robot ARoS learns the plan of an assembly task by observing human tutors executing possible sequential orders of sub-goals. An extension of the basic model is also proposed and tested in a real-world experiment. It addresses the fundamental problem of a hierarchical encoding of complex sequential tasks. It is shown how verbal feedback by the tutor about a serial order error may lead to the autonomous development of a neural representation of a group of sub-goals forming a sub-task. The second serial order problem of learning goal-directed chains of motor primitives is addressed by combining the associative learning mechanism of the dynamic field model with self-organizing properties. Inspired by the basic idea of the Kohonen's map algorithm, it is shown how self-organizing principles can be exploited to develop field representations of motor primitives, like for instance, a specific grasping behaviour, from observed motion trajectories. Moreover, the integration of additional contextual cues (e.g. object properties) in the learning process may cause the splitting of an existing motor primitive representation into two new representations that are context sensitive. In model simulations, it is shown that the learning mechanisms for representing sequential task knowledge in the DNF model can be also applied to establish chains of motor primitives directed towards a final goal (e.g. reach-grasp-place). Such a chained organization has been discussed in the neurophysiological literature to support not only a fluent execution of known action sequences but also the cognitive capacity of inferring the goal of observed motor behaviour of another individual.
Devido à procura crescente por robôs adaptativos capazes de auxiliar humanos nas suas tarefas diárias, um dos mais importantes objetivos da investigação em robótica atual é o de dotar robôs com a capacidade de aprender. O trabalho apresentado nesta tese foca-se na integração de capacidades de aprendizagem numa arquitetura de controlo, baseada em campos dinâmicos, para colaboração fluente entre Humano e Robô. Especificamente são abordados dois problemas importantes relacionados com ordem sequencial que surgem na arquitetura em níveis de abstração distintos embora relacionados: 1) a aprendizagem da ordem sequencial de sub-objetivos que tem de ser seguida para completar uma determinada tarefa e 2) a aprendizagem de representações de primitivas motoras que podem ser encadeadas para atingir um certo sub-objetivo. Foi desenvolvido um modelo baseado em teoria de Campos Dinâmicos Neuronais (CDNs) que permite ao robô adquirir um plano sequencial multi-ordem de uma tarefa a partir de demonstrações de tutores humanos. O modelo é inspirado em princípios conhecidos da aprendizagem da ordem sequencial por humanos. Especificamente, implementa a ideia de dois sistemas de aprendizagem complementares. Um mecanismo rápido codifica a ordem sequencial de uma demonstração única. Durante períodos de repetição interna, este mecanismo age como professor de um sistema mais lento responsável por extrair conhecimento generalizado da tarefa a partir das demonstrações de diferentes utilizadores. A eficiência do modelo de aprendizagem é testada numa experiência em cenário real na qual o robô humanoide ARoS aprende o plano de uma tarefa de montagem através da observação de tutores humanos que executam possíveis ordens sequenciais de execução dos sub-objetivos. Uma extensão do modelo básico é também proposta e testada em ambiente real. A extensão aborda o problema fundamental da codificação hierárquica de tarefas sequenciais complexas. É mostrado como feedback verbal fornecido pelo tutor acerca de erros na sequência pode levar ao desenvolvimento autónomo de uma representação neuronal de um grupo de sub-objetivos que formam uma sub-tarefa. O segundo problema de ordem sequencial, que consiste na aprendizagem de cadeias de primitivas motoras direcionadas a um objetivo, é abordado através da combinação do mecanismo de aprendizagem associativa do modelo baseado em Campos Dinâmicos com propriedades de auto-organização. Partindo da ideia fundamental do algoritmo do mapa de Kohonen, é mostrado como princípios de auto-organização podem ser explorados para desenvolver representações, em campos dinâmicos, de primitivas motoras, como por exemplo, um gesto especifico de agarrar, a partir de trajetórias de movimentos observados. Além disso, a integração de informação contextual adicional (e.g. propriedades do objeto) no processo de aprendizagem pode causar a divisão de uma representação de uma primitiva motora em duas novas representações específicas de cada contexto. É mostrado em simulação que os modelos de aprendizagem para representação do conhecimento da tarefa sequencial no modelo baseado em CDNs podem ser também aplicados na formação de cadeias de primitivas motoras direcionadas a um objetivo final (e.g. aproximar-agarrar-colocar). A organização em cadeia tem sido discutida na literatura sobre neurofisiologia, como sendo o suporte não só da execução fluente de ações sequenciais conhecidas, mas também da capacidade cognitiva de inferir o objetivo do comportamento motor observado num outro individuo.
Fundação para a Ciência e a Tecnologia SFRH/BD/48529/2008

The generation of adaptive and reflexive behavior is a challenging task in artificial intelligence and robotics. In this thesis, we develop a framework for knowledge representation, acquisition, and behavior generation that explicitly incorporates semantics, adaptive reasoning and knowledge revision. By using our model, semantic information can be exploited by traditional planning and decision making frameworks to generate empirically effective and adaptive robot behaviors, as well as to enable complex but natural human-robot interactions. In our work, we introduce a model of semantic mapping, we connect it with the notion of affordances, and we use those concepts to develop semantic-driven algorithms for knowledge acquisition, update, learning and robot behavior generation. In particular, we apply such models within existing planning and decision making frameworks to achieve semantic-driven and adaptive robot behaviors in a generic environment. On the one hand, this work generalizes existing semantic mapping models and extends them to include the notion of affordances. On the other hand, this work integrates semantic information within well-defined long-term planning and situated action frameworks to effectively generate adaptive robot behaviors. We validate our approach by evaluating it on a number of problems and robot tasks. In particular, we consider service robots deployed in interactive and social domains, such as offices and domestic environments. To this end, we also develop prototype applications that are useful for evaluation purposes.

碩士
國立嘉義大學
數學教育研究所
95
The purpose of this study is to probe into the influence of manipulating dynamic representations on the performance of senior vocational school students. A total of 319 senior vocational school students participated in this survey process. The respondents’ background misconception were collected and measured by using the researcher-designed questionnaire. Then 76 second-grade students taught by the researcher were taken as the mainly study object. The collected data were statistically analyzed by using ANCOVA and qualitative observation models. Results and evidences indicated that the 0/0 atypical limit and juding increasing or decreasing intervals by one-degree deritivatives were two out of seven selected sections in which empiricists learned better than those who in traditional course. On the other hand, substitution was the section in which traditional course learners did better. Generally speaking, teaching with manipulating dynamic representations had the positive effect for vocational school students on learning these units. However, traditional teaching methods were still recommendable. Suggestions for teachers, schools and researchers are provided in this paper based on the research findings.

abstract: Autonomic closure is a recently-proposed subgrid closure methodology for large eddy simulation (LES) that replaces the prescribed subgrid models used in traditional LES closure with highly generalized representations of subgrid terms and solution of a local system identification problem that allows the simulation itself to determine the local relation between each subgrid term and the resolved variables at every point and time. The present study demonstrates, for the first time, practical LES based on fully dynamic implementation of autonomic closure for the subgrid stress and the subgrid scalar flux. It leverages the inherent computational efficiency of tensorally-correct generalized representations in terms of parametric quantities, and uses the fundamental representation theory of Smith (1971) to develop complete and minimal tensorally-correct representations for the subgrid stress and scalar flux. It then assesses the accuracy of these representations via a priori tests, and compares with the corresponding accuracy from nonparametric representations and from traditional prescribed subgrid models. It then assesses the computational stability of autonomic closure with these tensorally-correct parametric representations, via forward simulations with a high-order pseudo-spectral code, including the extent to which any added stabilization is needed to ensure computational stability, and compares with the added stabilization needed in traditional closure with prescribed subgrid models. Further, it conducts a posteriori tests based on forward simulations of turbulent conserved scalar mixing with the same pseudo-spectral code, in which velocity and scalar statistics from autonomic closure with these representations are compared with corresponding statistics from traditional closure using prescribed models, and with corresponding statistics of filtered fields from direct numerical simulation (DNS). These comparisons show substantially greater accuracy from autonomic closure than from traditional closure. This study demonstrates that fully dynamic autonomic closure is a practical approach for LES that requires accuracy even at the smallest resolved scales.
Dissertation/Thesis
Doctoral Dissertation Aerospace Engineering 2020

A key component of research in the aerospace industry constitutes hypersonic flights (M>5) which includes the design of commercial high-speed aircrafts and development of rockets. Computational analysis becomes more important due to the difficulty in performing experiments and reliability of its results at these harsh operating conditions. There is an increasing demand from the industry for the accurate prediction of wall-shear and heat transfer with a low computational cost. Direct Numerical Simulations (DNS) create the standard for accuracy, but its practical usage is difficult and limited because of its high cost of computation. The usage of Reynold's Averaged Navier Stokes (RANS) simulations provide an affordable gateway for industry to capitalize its lower computational time for practical applications. However, the presence of existing RANS turbulence closure models and associated wall functions result in poor prediction of wall fluxes and inaccurate solutions in comparison with high fidelity DNS data. In recent years, machine learning emerged as a new approach for physical modeling. This thesis explores the potential of employing Machine Learning (ML) to improve the predictions of wall fluxes for hypersonic turbulent boundary layers. Fine-grid RANS simulations are used as training data to construct a suitable machine learning model to improve the solutions and predictions of wall quantities for coarser meshes. This strategy eliminates the usage of wall models and extends the range of applicability of grid sizes without a significant drop in accuracy of solutions. Random forest methodology coupled with a bagged aggregation algorithm helps in modeling a correction factor for the velocity gradient at the first grid points. The training data set for the ML model extracted from fine-grid RANS, includes neighbor cell information to address the memory effect of turbulence, and an optimal set of parameters to model the gradient correction factor. The successful demonstration of accurate predictions of wall-shear for coarse grids using this methodology, provides the confidence to build machine learning models to use DNS or high-fidelity modeling results as training data for reduced-order turbulence model development. This paves the way to integrate machine learning with RANS to produce accurate solutions with significantly lesser computational costs for hypersonic boundary layer problems.