• Refine Query
  • Source
  • Publication year
  • to
  • Language
  • 62
  • 17
  • 8
  • 5
  • 4
  • 2
  • 2
  • 2
  • 2
  • 2
  • 1
  • 1
  • Tagged with
  • 117
  • 20
  • 18
  • 17
  • 15
  • 14
  • 14
  • 13
  • 13
  • 13
  • 13
  • 12
  • 12
  • 12
  • 12
  • About
  • The Global ETD Search service is a free service for researchers to find electronic theses and dissertations. This service is provided by the Networked Digital Library of Theses and Dissertations.
    Our metadata is collected from universities around the world. If you manage a university/consortium/country archive and want to be added, details can be found on the NDLTD website.
101

LaMOSNet: Latent Mean-Opinion-Score Network for Non-intrusive Speech Quality Assessment : Deep Neural Network for MOS Prediction / LaMOSNet: Latent Mean-Opinion-Score Network för icke-intrusiv ljudkvalitetsbedömning : Djupt neuralt nätverk för MOS prediktion

Cumlin, Fredrik January 2022 (has links)
Objective non-intrusive speech quality assessment aimed to emulate and correlate with human judgement has received more attention over the years. It is a difficult problem due to three reasons: data scarcity, noisy human judgement, and a potential uneven distribution of bias of mean opinion scores (MOS). In this paper, we introduce the Latent Mean-Opinion-Score Network (LaMOSNet) that leverage on individual judge’s scores to increase the data size, and new ideas to deal with both noisy and biased labels. We introduce a methodology called Optimistic Judge Estimation as a way to reduce bias in MOS in a clear way. We also implement stochastic gradient noise and mean teacher, ideas from noisy image classification, to further deal with noisy and uneven bias distribution of labels. We achieve competitive results on VCC2018 modeling MOS, and state-of-the-art modeling only listener dependent scores. / Objektiv referensfri ljudkvalitétsbedömning ämnad att härma och korrelera med mänsklig bedömning har fått mer uppmärksamhet med åren. Det är ett svårt problem på grund av tre anledningar: brist på data, varians i mänsklig bedömning, och en potentiell ojämn fördelning av bias av medel bedömningsvärde (mean opinion score, MOS). I detta papper introducerar vi Latent Mean-Opinion-Score Network (LaMOSNet) som tar nytta av individuella bedömmares poäng för att öka datastorleken, och nya idéer för att handskas med både varierande och partisk märkning. Jag introducerar en metodologi som kallas Optimistisk bedömmarestimering, ett sätt att minska partiskheten i MOS på ett klart sätt. Jag implementerar också stokastisk gradient variation och medellärare, idéer från opålitlig bild igenkänning, för att ännu mer hantera opålitliga märkningar. Jag får jämförelsebara resultat på VCC2018 när jag modellerar MOS, och state-of-the-art när jag modellerar enbart beömmarnas märkning.
102

Beiträge zur Regularisierung inverser Probleme und zur bedingten Stabilität bei partiellen Differentialgleichungen

Shao, Yuanyuan 17 January 2013 (has links) (PDF)
Wir betrachten die lineare inverse Probleme mit gestörter rechter Seite und gestörtem Operator in Hilberträumen, die inkorrekt sind. Um die Auswirkung der Inkorrektheit zu verringen, müssen spezielle Lösungsmethode angewendet werden, hier nutzen wir die sogenannte Tikhonov Regularisierungsmethode. Die Regularisierungsparameter wählen wir aus das verallgemeinerte Defektprinzip. Eine typische numerische Methode zur Lösen der nichtlinearen äquivalenten Defektgleichung ist Newtonverfahren. Wir schreiben einen Algorithmus, die global und monoton konvergent für beliebige Startwerte garantiert. Um die Stabilität zu garantieren, benutzen wir die Glattheit der Lösung, dann erhalten wir eine sogenannte bedingte Stabilität. Wir demonstrieren die sogenannte Interpolationsmethode zur Herleitung von bedingten Stabilitätsabschätzungen bei inversen Problemen für partielle Differentialgleichungen.
103

Beiträge zur Regularisierung inverser Probleme und zur bedingten Stabilität bei partiellen Differentialgleichungen

Shao, Yuanyuan 14 January 2013 (has links)
Wir betrachten die lineare inverse Probleme mit gestörter rechter Seite und gestörtem Operator in Hilberträumen, die inkorrekt sind. Um die Auswirkung der Inkorrektheit zu verringen, müssen spezielle Lösungsmethode angewendet werden, hier nutzen wir die sogenannte Tikhonov Regularisierungsmethode. Die Regularisierungsparameter wählen wir aus das verallgemeinerte Defektprinzip. Eine typische numerische Methode zur Lösen der nichtlinearen äquivalenten Defektgleichung ist Newtonverfahren. Wir schreiben einen Algorithmus, die global und monoton konvergent für beliebige Startwerte garantiert. Um die Stabilität zu garantieren, benutzen wir die Glattheit der Lösung, dann erhalten wir eine sogenannte bedingte Stabilität. Wir demonstrieren die sogenannte Interpolationsmethode zur Herleitung von bedingten Stabilitätsabschätzungen bei inversen Problemen für partielle Differentialgleichungen.
104

Essays in economics of information

Gendron-Saulnier, Catherine 04 1900 (has links)
Cette thèse est une collection de trois articles en économie de l'information. Le premier chapitre sert d'introduction et les Chapitres 2 à 4 constituent le coeur de l'ouvrage. Le Chapitre 2 porte sur l’acquisition d’information sur l’Internet par le biais d'avis de consommateurs. En particulier, je détermine si les avis laissés par les acheteurs peuvent tout de même transmettre de l’information à d’autres consommateurs, lorsqu’il est connu que les vendeurs peuvent publier de faux avis à propos de leurs produits. Afin de comprendre si cette manipulation des avis est problématique, je démontre que la plateforme sur laquelle les avis sont publiés (e.g. TripAdvisor, Yelp) est un tiers important à considérer, autant que les vendeurs tentant de falsifier les avis. En effet, le design adopté par la plateforme a un effet indirect sur le niveau de manipulation des vendeurs. En particulier, je démontre que la plateforme, en cachant une partie du contenu qu'elle détient sur les avis, peut parfois améliorer la qualité de l'information obtenue par les consommateurs. Finalement, le design qui est choisi par la plateforme peut être lié à la façon dont elle génère ses revenus. Je montre qu'une plateforme générant des revenus par le biais de commissions sur les ventes peut être plus tolérante à la manipulation qu'une plateforme qui génère des revenus par le biais de publicité. Le Chapitre 3 est écrit en collaboration avec Marc Santugini. Dans ce chapitre, nous étudions les effets de la discrimination par les prix au troisième degré en présence de consommateurs non informés qui apprennent sur la qualité d'un produit par le biais de son prix. Dans un environnement stochastique avec deux segments de marché, nous démontrons que la discrimination par les prix peut nuire à la firme et être bénéfique pour les consommateurs. D'un côté, la discrimination par les prix diminue l'incertitude à laquelle font face les consommateurs, c.-à-d., la variance des croyances postérieures est plus faible avec discrimination qu'avec un prix uniforme. En effet, le fait d'observer deux prix (avec discrimination) procure plus d'information aux consommateurs, et ce, même si individuellement chacun de ces prix est moins informatif que le prix uniforme. De l'autre côté, il n'est pas toujours optimal pour la firme de faire de la discrimination par les prix puisque la présence de consommateurs non informés lui donne une incitation à s'engager dans du signaling. Si l'avantage procuré par la flexibilité de fixer deux prix différents est contrebalancé par le coût du signaling avec deux prix différents, alors il est optimal pour la firme de fixer un prix uniforme sur le marché. Finalement, le Chapitre 4 est écrit en collaboration avec Sidartha Gordon. Dans ce chapitre, nous étudions une classe de jeux où les joueurs sont contraints dans le nombre de sources d'information qu'ils peuvent choisir pour apprendre sur un paramètre du jeu, mais où ils ont une certaine liberté quant au degré de dépendance de leurs signaux, avant de prendre une action. En introduisant un nouvel ordre de dépendance entre signaux, nous démontrons qu'un joueur préfère de l'information qui est la plus dépendante possible de l'information obtenue par les joueurs pour qui les actions sont soit, compléments stratégiques et isotoniques, soit substituts stratégiques et anti-toniques, avec la sienne. De même, un joueur préfère de l'information qui est la moins dépendante possible de l'information obtenue par les joueurs pour qui les actions sont soit, substituts stratégiques et isotoniques, soit compléments stratégiques et anti-toniques, avec la sienne. Nous établissons également des conditions suffisantes pour qu'une structure d'information donnée, information publique ou privée par exemple, soit possible à l'équilibre. / This thesis is a collection of three essays in economics of information. Chapter 1 is a general introduction and Chapters 2 to 4 form the core of the thesis. Chapter 2 analyzes information dissemination on the Internet. Online platforms such as Amazon, TripAdvisor or Yelp are now key sources of information for modern consumers. The proportion of consumers consulting online reviews prior to purchasing a good or a service has grown persistently. Yet, sellers have been accused of hiring shills to post fake reviews about their products. This raises the question: Does the presence of shills make reviews less informative? I show that the answers to this question depend on the way the platform presents and summarizes reviews on its website. In particular, I find that withholding information by garbling the reviews benefits information dissemination by inducing the seller to destroy less information with manipulation. Next, I show that the platform's choice regarding how to present reviews hinges on its revenue source. Indeed, a platform that receives sales commissions optimally commits to publishing information differently from a platform that receives revenues from advertisements or from subscription fees. Incidentally, such platforms have contrasting impacts on the amount of information that is transmitted by reviews. Chapter 3 is co-authored with Marc Santugini. In this chapter, we study the impact of third-degree price discrimination in the presence of uninformed buyers who extract noisy information from observing prices. In a noisy learning environment, it is shown that price discrimination can be detrimental to the firm and beneficial to the consumers. On the one hand, discriminatory pricing reduces consumers’ uncertainty, i.e., the variance of posterior beliefs upon observing prices is reduced. Specifically, observing two prices under discriminatory pricing provides more information than one price under uniform pricing even when discriminatory pricing reduces the amount of information contained in each price. On the other hand, it is not always optimal for the firm to use discriminatory pricing since the presence of uninformed buyers provides the firm with the incentive to engage in noisy price signaling. Indeed, if the benefit from price flexibility (through discriminatory pricing) is offset by the cost of signaling quality through two distinct prices, then it is optimal to integrate markets and to use uniform pricing. Finally, Chapter 4 is co-authored with Sidartha Gordon. In this chapter, we study a class of games where players face restrictions on how much information they can obtain on a common payoff relevant state, but have some leeway in covertly choosing the dependence between their signals, before simultaneously choosing actions. Using a new stochastic dependence ordering between signals, we show that each player chooses information that is more dependent on the information of other players whose actions are either isotonic and complements with his actions or antitonic and substitutes with his actions. Similarly, each player chooses information that is less dependent on the information of other players whose actions are antitonic and complements with his actions or isotonic and substitutes with his actions. We then provide sufficient conditions for information structures such as public or private information to arise in equilibrium.
105

Uma abordagem evolutiva para geração procedural de níveis em jogos de quebra-cabeças baseados em física / An evolutionary approach for procedural generation of levels in physics-based puzzle games

Ferreira, Lucas Nascimento 15 July 2015 (has links)
Na última década diversos algoritmos baseados em busca foram desenvolvidos para a geração de níveis em diferentes tipos de jogos. O espaço de busca para geração de níveis geralmente possui restrições, uma vez que a mecânica de um jogo define regras de factibilidade para os níveis. Em alguns métodos, a avaliação de factibilidade requer uma simulação com um agente inteligente que controla o jogo. Esse processo de avaliação geralmente possui ruído, causado por componentes aleatórios no simulador ou na estratégia do agente. Diversos trabalhos têm utilizado simulação como forma de avaliação de conteúdo, no entanto, nenhum deles discutiu profundamente a presença de ruído neste tipo de abordagem. Assim, esse trabalho apresenta um algoritmo genético capaz de gerar níveis factíveis que são avaliados por um agente inteligente em uma simulação ruidosa. O algoritmo foi aplicado a jogos de quebra-cabeças baseados em física com a mecânica do Angry Birds. Uma representação dos níveis em forma de indivíduos é introduzida, a qual permite que o algoritmo genético os evolua com características diferenciadas. O ruído na função de aptidão é tratado por uma nova abordagem, baseada em uma sistema de cache, que auxilia o algoritmo genético a encontrar boas soluções candidatas. Três conjuntos de experimentos foram realizados para avaliar o algoritmo. O primeiro compara o método de cache proposto com outros métodos de redução de ruído da literatura. O segundo mede a expressividade do algoritmo genético considerando as características estruturais dos níveis gerados. O último avalia os níveis gerados considerando aspectos de design (como dificuldade, imersão e diversão), os quais são medidos por meio de questionários respondidos por jogadores humanos via Internet. Os resultados mostraram que o algoritmo genético foi capaz de gerar níveis distintos que são tão imersíveis quanto níveis produzidos manualmente. Além disso, a abordagem de cache lidou apropriadamente com o ruído nos cálculos de aptidão, permitindo uma correta evolução elitista. / In the last decade several search-based algorithms have been developed for generating levels in different types of games. The search space for level generation is typically constrained once the game mechanics define feasibility rules for the levels. In some methods, evaluating level feasibility requires a simulation with an intelligent agent which plays the game. This evaluation process usually has noise, caused by random components in the simulator or in the agent strategy. Several works have used a simulation for content evaluation, however, none of them have deeply discussed the presence of noise in this kind of approach. Thus, this paper presents a genetic algorithm capable of generating feasible levels that are evaluated by an intelligent agent in a noisy simulation. The algorithm was applied to physics-based puzzle games with the Angry Birds mechanics. A level representation in the form of individuals is introduced, which allows the genetic algorithm to evolve them with distinct characteristics. The fitness function noise is handled by a new approach, based on a cache system, which helps the genetic algorithm finding good candidate solutions. Three sets of experiments were conducted to evaluate the algorithm. The first one compares the proposed cache approach with other noise reduction methods of the literature. The second one measures the expressivity of the genetic algorithm considering the structural characteristics of the levels. The last one evaluates design aspects (such as difficulty, immersion and fun) of the generated levels using questionnaires answered by human players via Internet. Results showed the genetic algorithm was capable of generating distinct levels that are as immersive as levels manually designed. Moreover, the cache approach handled properly the noise in the fitness calculations, allowing a correct elitist evolution.
106

Cooperative Networks with Channel Uncertainty / Réseaux coopératifs avec incertitude du canal

Behboodi, Arash 13 June 2012 (has links)
Dans cette thèse, les réseaux coopératifs sont étudiés sous cette hypothèse que la source est incertain par rapport le canal en opération. Dans le premier chapitre, des stratégies coopératives sont développées pour les canaux à relais simultanés (SRC) lesquelles se composent d'un ensemble de deux canaux à relais parmi lesquels le canal en opération est choisi. Cela est équivalent au canal de diffusion à relais (BRC). Les bornes sur la région de capacité de BRC général sont dérivées. Les résultats de capacité sont obtenus pour les cas particuliers du canal à relais simultané semi-dégradé et dégradé Gaussien. Dans le deuxième chapitre, le canal à relais composite est considéré où le canal est tiré aléatoirement d'un ensemble de la distribution conditionnelle. Le débit est fixé en dépit du canal actuel et la probabilité d'erreur (EP) asymptotique est caractérisée. Une nouvelle stratégie de codage sélectif (SCS) est introduit permettant aux relais de choisir -selon leur mesurage du canal – la meilleur schéma de codage entre Décoder-et-Transmettre (DF) et Comprimer-et-Transmettre (CF). Les théorèmes de codage de réseau bruit généralisées sont démontrés pour le cas de réseau unicast général où les relais utilisent soit DF soit CF. Dans le troisième chapitre, le spectre asymptotique de EP est introduit en tant que nouvelle mesure de performance pour réseaux composites. Il est démontré que chaque code avec le débit hors de la borne cut-set, abouti à EP égal à un et le spectre asymptotique de EP coïncide avec la probabilité d'outage pour les réseaux satisfaisant la converse forte. / In this thesis, cooperative networks are studied under the assumption that the source is uncertain about the channel in operation. In the first chapter, cooperative strategies are developed for simultaneous relay channels (SRC) which consist of a set of two single relay channels out of which the channel in operation is chosen. This is equivalent to the broadcast relay channel (BRC). Bounds on the capacity region of the general BRC with two helper relays are derived. Capacity results are obtained for specific cases of semi-degraded and degraded Gaussian simultaneous relay channels. In the second chapter, the composite relay channel is considered where the channel is randomly drawn from a set of conditional distributions according to a given distribution. The transmission rate is fixed regardless of the current channel and the asymptotic error probability (EP) is characterized. A novel selective coding strategy (SCS) is introduced which enables relays to select –based on their channel measurement– the best coding scheme between Compress-and-Forward (CF) and Decode-and-Forward (DF). Generalized Noisy Network Coding theorems are shown for the case of unicast general networks where the relays use either DF or CF scheme. In the third chapter, the asymptotic behavior of EP is studied for composite multiterminal networks. The asymptotic spectrum of EP is introduced as a novel performance measure for composite networks. It is shown that every code with rate outside cut-set bound, yields EP equal to one and for the networks satisfying strong converse condition, the asymptotic spectrum of EP coincides with the outage probability.
107

A scalable evolutionary learning classifier system for knowledge discovery in stream data mining

Dam, Hai Huong, Information Technology & Electrical Engineering, Australian Defence Force Academy, UNSW January 2008 (has links)
Data mining (DM) is the process of finding patterns and relationships in databases. The breakthrough in computer technologies triggered a massive growth in data collected and maintained by organisations. In many applications, these data arrive continuously in large volumes as a sequence of instances known as a data stream. Mining these data is known as stream data mining. Due to the large amount of data arriving in a data stream, each record is normally expected to be processed only once. Moreover, this process can be carried out on different sites in the organisation simultaneously making the problem distributed in nature. Distributed stream data mining poses many challenges to the data mining community including scalability and coping with changes in the underlying concept over time. In this thesis, the author hypothesizes that learning classifier systems (LCSs) - a class of classification algorithms - have the potential to work efficiently in distributed stream data mining. LCSs are an incremental learner, and being evolutionary based they are inherently adaptive. However, they suffer from two main drawbacks that hinder their use as fast data mining algorithms. First, they require a large population size, which slows down the processing of arriving instances. Second, they require a large number of parameter settings, some of them are very sensitive to the nature of the learning problem. As a result, it becomes difficult to choose a right setup for totally unknown problems. The aim of this thesis is to attack these two problems in LCS, with a specific focus on UCS - a supervised evolutionary learning classifier system. UCS is chosen as it has been tested extensively on classification tasks and it is the supervised version of XCS, a state of the art LCS. In this thesis, the architectural design for a distributed stream data mining system will be first introduced. The problems that UCS should face in a distributed data stream task are confirmed through a large number of experiments with UCS and the proposed architectural design. To overcome the problem of large population sizes, the idea of using a Neural Network to represent the action in UCS is proposed. This new system - called NLCS { was validated experimentally using a small fixed population size and has shown a large reduction in the population size needed to learn the underlying concept in the data. An adaptive version of NLCS called ANCS is then introduced. The adaptive version dynamically controls the population size of NLCS. A comprehensive analysis of the behaviour of ANCS revealed interesting patterns in the behaviour of the parameters, which motivated an ensemble version of the algorithm with 9 nodes, each using a different parameter setting. In total they cover all patterns of behaviour noticed in the system. A voting gate is used for the ensemble. The resultant ensemble does not require any parameter setting, and showed better performance on all datasets tested. The thesis concludes with testing the ANCS system in the architectural design for distributed environments proposed earlier. The contributions of the thesis are: (1) reducing the UCS population size by an order of magnitude using a neural representation; (2) introducing a mechanism for adapting the population size; (3) proposing an ensemble method that does not require parameter setting; and primarily (4) showing that the proposed LCS can work efficiently for distributed stream data mining tasks.
108

A scalable evolutionary learning classifier system for knowledge discovery in stream data mining

Dam, Hai Huong, Information Technology & Electrical Engineering, Australian Defence Force Academy, UNSW January 2008 (has links)
Data mining (DM) is the process of finding patterns and relationships in databases. The breakthrough in computer technologies triggered a massive growth in data collected and maintained by organisations. In many applications, these data arrive continuously in large volumes as a sequence of instances known as a data stream. Mining these data is known as stream data mining. Due to the large amount of data arriving in a data stream, each record is normally expected to be processed only once. Moreover, this process can be carried out on different sites in the organisation simultaneously making the problem distributed in nature. Distributed stream data mining poses many challenges to the data mining community including scalability and coping with changes in the underlying concept over time. In this thesis, the author hypothesizes that learning classifier systems (LCSs) - a class of classification algorithms - have the potential to work efficiently in distributed stream data mining. LCSs are an incremental learner, and being evolutionary based they are inherently adaptive. However, they suffer from two main drawbacks that hinder their use as fast data mining algorithms. First, they require a large population size, which slows down the processing of arriving instances. Second, they require a large number of parameter settings, some of them are very sensitive to the nature of the learning problem. As a result, it becomes difficult to choose a right setup for totally unknown problems. The aim of this thesis is to attack these two problems in LCS, with a specific focus on UCS - a supervised evolutionary learning classifier system. UCS is chosen as it has been tested extensively on classification tasks and it is the supervised version of XCS, a state of the art LCS. In this thesis, the architectural design for a distributed stream data mining system will be first introduced. The problems that UCS should face in a distributed data stream task are confirmed through a large number of experiments with UCS and the proposed architectural design. To overcome the problem of large population sizes, the idea of using a Neural Network to represent the action in UCS is proposed. This new system - called NLCS { was validated experimentally using a small fixed population size and has shown a large reduction in the population size needed to learn the underlying concept in the data. An adaptive version of NLCS called ANCS is then introduced. The adaptive version dynamically controls the population size of NLCS. A comprehensive analysis of the behaviour of ANCS revealed interesting patterns in the behaviour of the parameters, which motivated an ensemble version of the algorithm with 9 nodes, each using a different parameter setting. In total they cover all patterns of behaviour noticed in the system. A voting gate is used for the ensemble. The resultant ensemble does not require any parameter setting, and showed better performance on all datasets tested. The thesis concludes with testing the ANCS system in the architectural design for distributed environments proposed earlier. The contributions of the thesis are: (1) reducing the UCS population size by an order of magnitude using a neural representation; (2) introducing a mechanism for adapting the population size; (3) proposing an ensemble method that does not require parameter setting; and primarily (4) showing that the proposed LCS can work efficiently for distributed stream data mining tasks.
109

A scalable evolutionary learning classifier system for knowledge discovery in stream data mining

Dam, Hai Huong, Information Technology & Electrical Engineering, Australian Defence Force Academy, UNSW January 2008 (has links)
Data mining (DM) is the process of finding patterns and relationships in databases. The breakthrough in computer technologies triggered a massive growth in data collected and maintained by organisations. In many applications, these data arrive continuously in large volumes as a sequence of instances known as a data stream. Mining these data is known as stream data mining. Due to the large amount of data arriving in a data stream, each record is normally expected to be processed only once. Moreover, this process can be carried out on different sites in the organisation simultaneously making the problem distributed in nature. Distributed stream data mining poses many challenges to the data mining community including scalability and coping with changes in the underlying concept over time. In this thesis, the author hypothesizes that learning classifier systems (LCSs) - a class of classification algorithms - have the potential to work efficiently in distributed stream data mining. LCSs are an incremental learner, and being evolutionary based they are inherently adaptive. However, they suffer from two main drawbacks that hinder their use as fast data mining algorithms. First, they require a large population size, which slows down the processing of arriving instances. Second, they require a large number of parameter settings, some of them are very sensitive to the nature of the learning problem. As a result, it becomes difficult to choose a right setup for totally unknown problems. The aim of this thesis is to attack these two problems in LCS, with a specific focus on UCS - a supervised evolutionary learning classifier system. UCS is chosen as it has been tested extensively on classification tasks and it is the supervised version of XCS, a state of the art LCS. In this thesis, the architectural design for a distributed stream data mining system will be first introduced. The problems that UCS should face in a distributed data stream task are confirmed through a large number of experiments with UCS and the proposed architectural design. To overcome the problem of large population sizes, the idea of using a Neural Network to represent the action in UCS is proposed. This new system - called NLCS { was validated experimentally using a small fixed population size and has shown a large reduction in the population size needed to learn the underlying concept in the data. An adaptive version of NLCS called ANCS is then introduced. The adaptive version dynamically controls the population size of NLCS. A comprehensive analysis of the behaviour of ANCS revealed interesting patterns in the behaviour of the parameters, which motivated an ensemble version of the algorithm with 9 nodes, each using a different parameter setting. In total they cover all patterns of behaviour noticed in the system. A voting gate is used for the ensemble. The resultant ensemble does not require any parameter setting, and showed better performance on all datasets tested. The thesis concludes with testing the ANCS system in the architectural design for distributed environments proposed earlier. The contributions of the thesis are: (1) reducing the UCS population size by an order of magnitude using a neural representation; (2) introducing a mechanism for adapting the population size; (3) proposing an ensemble method that does not require parameter setting; and primarily (4) showing that the proposed LCS can work efficiently for distributed stream data mining tasks.
110

Error-robust coding and transformation of compressed hybered hybrid video streams for packet-switched wireless networks

Halbach, Till January 2004 (has links)
<p>This dissertation considers packet-switched wireless networks for transmission of variable-rate layered hybrid video streams. Target applications are video streaming and broadcasting services. The work can be divided into two main parts.</p><p>In the first part, a novel quality-scalable scheme based on coefficient refinement and encoder quality constraints is developed as a possible extension to the video coding standard H.264. After a technical introduction to the coding tools of H.264 with the main focus on error resilience features, various quality scalability schemes in previous research are reviewed. Based on this discussion, an encoder decoder framework is designed for an arbitrary number of quality layers, hereby also enabling region-of-interest coding. After that, the performance of the new system is exhaustively tested, showing that the bit rate increase typically encountered with scalable hybrid coding schemes is, for certain coding parameters, only small to moderate. The double- and triple-layer constellations of the framework are shown to perform superior to other systems.</p><p>The second part considers layered code streams as generated by the scheme of the first part. Various error propagation issues in hybrid streams are discussed, which leads to the definition of a decoder quality constraint and a segmentation of the code stream to transmit. A packetization scheme based on successive source rate consumption is drafted, followed by the formulation of the channel code rate optimization problem for an optimum assignment of available codes to the channel packets. Proper MSE-based error metrics are derived, incorporating the properties of the source signal, a terminate-on-error decoding strategy, error concealment, inter-packet dependencies, and the channel conditions. The Viterbi algorithm is presented as a low-complexity solution to the optimization problem, showing a great adaptivity of the joint source channel coding scheme to the channel conditions. An almost constant image qualiity is achieved, also in mismatch situations, while the overall channel code rate decreases only as little as necessary as the channel quality deteriorates. It is further shown that the variance of code distributions is only small, and that the codes are assigned irregularly to all channel packets.</p><p>A double-layer constellation of the framework clearly outperforms other schemes with a substantial margin. </p><p>Keywords — Digital lossy video compression, visual communication, variable bit rate (VBR), SNR scalability, layered image processing, quality layer, hybrid code stream, predictive coding, progressive bit stream, joint source channel coding, fidelity constraint, channel error robustness, resilience, concealment, packet-switched, mobile and wireless ATM, noisy transmission, packet loss, binary symmetric channel, streaming, broadcasting, satellite and radio links, H.264, MPEG-4 AVC, Viterbi, trellis, unequal error protection</p>

Page generated in 0.051 seconds