004 Datenverarbeitung; Informatik
Refine
Has Fulltext
- yes (75)
Is part of the Bibliography
- yes (75)
Year of publication
Document Type
- Doctoral Thesis (75) (remove)
Keywords
- Leistungsbewertung (13)
- Robotik (8)
- Mobiler Roboter (7)
- Autonomer Roboter (6)
- Komplexitätstheorie (5)
- Optimierung (5)
- Simulation (5)
- Theoretische Informatik (5)
- Modellierung (4)
- Netzwerk (4)
Institute
- Institut für Informatik (75) (remove)
Schriftenreihe
Sonstige beteiligte Institutionen
The work presents a performance evaluation and optimization of so-called overlay networks for content distribution in the Internet. Chapter 1 describes the importance which have such networks in today's Internet, for example, for the transmission of video content. The focus of this work is on overlay networks based on the peer-to-peer principle. These are characterized by the fact that users who download content, also contribute to the distribution process by sharing parts of the data to other users. This enables efficient content distribution because each user not only consumes resources in the system, but also provides its own resources. Chapter 2 of the monograph contains a detailed description of the functionality of today's most popular overlay network BitTorrent. It explains the various components and their interaction. This is followed by an illustration of why such overlay networks for Internet service providers (ISPs) are problematic. The reason lies in the large amount of inter-ISP traffic that is produced by these overlay networks. Since this inter-ISP traffic leads to high costs for ISPs, they try to reduce it by improved mechanisms for overlay networks. One optimization approach is the use of topology awareness within the overlay networks. It provides users of the overlay networks with information about the underlying physical network topology. This allows them to avoid inter-ISP traffic by exchanging data preferrentially with other users that are connected to the same ISP. Another approach to save inter-ISP traffic is caching. In this case the ISP provides additional computers in its network, called caches, which store copies of popular content. The users of this ISP can then obtain such content from the cache. This prevents that the content must be retrieved from locations outside of the ISP's network, and saves costly inter-ISP traffic in this way. In the third chapter of the thesis, the results of a comprehensive measurement study of overlay networks, which can be found in today's Internet, are presented. After a short description of the measurement methodology, the results of the measurements are described. These results contain data on a variety of characteristics of current P2P overlay networks in the Internet. These include the popularity of content, i.e., how many users are interested in specific content, the evolution of the popularity and the size of the files. The distribution of users within the Internet is investigated in detail. Special attention is given to the number of users that exchange a particular file within the same ISP. On the basis of these measurement results, an estimation of the traffic savings that can achieved by topology awareness is derived. This new estimation is of scientific and practical importance, since it is not limited to individual ISPs and files, but considers the whole Internet and the total amount of data exchanged in overlay networks. Finally, the characteristics of regional content are considered, in which the popularity is limited to certain parts of the Internet. This is for example the case of videos in German, Italian or French language. Chapter 4 of the thesis is devoted to the optimization of overlay networks for content distribution through caching. It presents a deterministic flow model that describes the influence of caches. On the basis of this model, it derives an estimate of the inter-ISP traffic that is generated by an overlay network, and which part can be saved by caches. The results show that the influence of the cache depends on the structure of the overlay networks, and that caches can also lead to an increase in inter-ISP traffic under certain circumstances. The described model is thus an important tool for ISPs to decide for which overlay networks caches are useful and to dimension them. Chapter 5 summarizes the content of the work and emphasizes the importance of the findings. In addition, it explains how the findings can be applied to the optimization of future overlay networks. Special attention is given to the growing importance of video-on-demand and real-time video transmissions.
Deep learning enables enormous progress in many computer vision-related tasks. Artificial Intel- ligence (AI) steadily yields new state-of-the-art results in the field of detection and classification. Thereby AI performance equals or exceeds human performance. Those achievements impacted many domains, including medical applications.
One particular field of medical applications is gastroenterology. In gastroenterology, machine learning algorithms are used to assist examiners during interventions. One of the most critical concerns for gastroenterologists is the development of Colorectal Cancer (CRC), which is one of the leading causes of cancer-related deaths worldwide. Detecting polyps in screening colonoscopies is the essential procedure to prevent CRC. Thereby, the gastroenterologist uses an endoscope to screen the whole colon to find polyps during a colonoscopy. Polyps are mucosal growths that can vary in severity.
This thesis supports gastroenterologists in their examinations with automated detection and clas- sification systems for polyps. The main contribution is a real-time polyp detection system. This system is ready to be installed in any gastroenterology practice worldwide using open-source soft- ware. The system achieves state-of-the-art detection results and is currently evaluated in a clinical trial in four different centers in Germany.
The thesis presents two additional key contributions: One is a polyp detection system with ex- tended vision tested in an animal trial. Polyps often hide behind folds or in uninvestigated areas. Therefore, the polyp detection system with extended vision uses an endoscope assisted by two additional cameras to see behind those folds. If a polyp is detected, the endoscopist receives a vi- sual signal. While the detection system handles the additional two camera inputs, the endoscopist focuses on the main camera as usual.
The second one are two polyp classification models, one for the classification based on shape (Paris) and the other on surface and texture (NBI International Colorectal Endoscopic (NICE) classification). Both classifications help the endoscopist with the treatment of and the decisions about the detected polyp.
The key algorithms of the thesis achieve state-of-the-art performance. Outstandingly, the polyp detection system tested on a highly demanding video data set shows an F1 score of 90.25 % while working in real-time. The results exceed all real-time systems in the literature. Furthermore, the first preliminary results of the clinical trial of the polyp detection system suggest a high Adenoma Detection Rate (ADR). In the preliminary study, all polyps were detected by the polyp detection system, and the system achieved a high usability score of 96.3 (max 100). The Paris classification model achieved an F1 score of 89.35 % which is state-of-the-art. The NICE classification model achieved an F1 score of 81.13 %.
Furthermore, a large data set for polyp detection and classification was created during this thesis. Therefore a fast and robust annotation system called Fast Colonoscopy Annotation Tool (FastCAT) was developed. The system simplifies the annotation process for gastroenterologists. Thereby the
i
gastroenterologists only annotate key parts of the endoscopic video. Afterward, those video parts are pre-labeled by a polyp detection AI to speed up the process. After the AI has pre-labeled the frames, non-experts correct and finish the annotation. This annotation process is fast and ensures high quality. FastCAT reduces the overall workload of the gastroenterologist on average by a factor of 20 compared to an open-source state-of-art annotation tool.
Complexity and Partitions
(2001)
Computational complexity theory usually investigates the complexity of sets, i.e., the complexity of partitions into two parts. But often it is more appropriate to represent natural problems by partitions into more than two parts. A particularly interesting class of such problems consists of classification problems for relations. For instance, a binary relation R typically defines a partitioning of the set of all pairs (x,y) into four parts, classifiable according to the cases where R(x,y) and R(y,x) hold, only R(x,y) or only R(y,x) holds or even neither R(x,y) nor R(y,x) is true. By means of concrete classification problems such as Graph Embedding or Entailment (for propositional logic), this thesis systematically develops tools, in shape of the boolean hierarchy of NP-partitions and its refinements, for the qualitative analysis of the complexity of partitions generated by NP-relations. The Boolean hierarchy of NP-partitions is introduced as a generalization of the well-known and well-studied Boolean hierarchy (of sets) over NP. Whereas the latter hierarchy has a very simple structure, the situation is much more complicated for the case of partitions into at least three parts. To get an idea of this hierarchy, alternative descriptions of the partition classes are given in terms of finite, labeled lattices. Based on these characterizations the Embedding Conjecture is established providing the complete information on the structure of the hierarchy. This conjecture is supported by several results. A natural extension of the Boolean hierarchy of NP-partitions emerges from the lattice-characterization of its classes by considering partition classes generated by finite, labeled posets. It turns out that all significant ideas translate from the case of lattices. The induced refined Boolean hierarchy of NP-partitions enables us more accuratly capturing the complexity of certain relations (such as Graph Embedding) and a description of projectively closed partition classes.
Aktivitätsbasierte Verhaltensmodellierung und ihre Unterstützung bei Multiagentensimulationen
(2000)
Durch Zusammenführung traditioneller Methoden zur individuenbasierten Simulation und dem Konzept der Multiagentensysteme steht mit der Multiagentensimulation eine Methodik zur Verfügung, die es ermöglicht, sowohl technisch als auch konzeptionell eine neue Ebene an Detaillierung bei Modellbildung und Simulation zu erreichen. Ein Modell beruht dabei auf dem Konzept einer Gesellschaft: Es besteht aus einer Menge interagierender, aber in ihren Entscheidungen autonomen Einheiten, den Agenten. Diese ändern durch ihre Aktionen ihre Umwelt und reagieren ebenso auf die für sie wahrnehmbaren Änderungen in der Umwelt. Durch die Simulation jedes Agenten zusammen mit der Umwelt, in der er "lebt", wird die Dynamik im Gesamtsystem beobachtbar. In der vorliegenden Dissertation wurde ein Repräsentationsschema für Multiagentensimulationen entwickelt werden, das es Fachexperten, wie zum Beispiel Biologen, ermöglicht, selbständig ohne traditionelles Programmieren Multiagentenmodelle zu implementieren und mit diesen Experimente durchzuführen. Dieses deklarative Schema beruht auf zwei Basiskonzepten: Der Körper eines Agenten besteht aus Zustandsvariablen. Das Verhalten des Agenten kann mit Regeln beschrieben werden. Ausgehend davon werden verschiedene Strukturierungsansätze behandelt. Das wichtigste Konzept ist das der "Aktivität", einer Art "Verhaltenszustand": Während der Agent in einer Aktivität A verweilt, führt er die zugehörigen Aktionen aus und dies solange, bis eine Regel feuert, die diese Aktivität beendet und eine neue Aktivität auswählt. Durch Indizierung dieser Regeln bei den zugehörigen Aktivitäten und Einführung von abstrakten Aktivitäten entsteht ein Schema für eine vielfältig strukturierbare Verhaltensbeschreibung. Zu diesem Schema wurde ein Interpreter entwickelt, der ein derartig repräsentiertes Modell ausführt und so Simulationsexperimente mit dem Multiagentenmodell erlaubt. Auf dieser Basis wurde die Modellierungs- und Experimentierumgebung SeSAm ("Shell für Simulierte Agentensysteme") entwickelt. Sie verwendet vorhandene Konzepte aus dem visuellen Programmieren. Mit dieser Umgebung wurden Anwendungsmodelle aus verschiedenen Domänen realisiert: Neben abstrakten Spielbeispielen waren dies vor allem Fragestellungen zu sozialen Insekten, z.B. zum Verhalten von Ameisen, Bienen oder der Interaktion zwischen Bienenvölkern und Milbenpopulationen.
Die Realisierung einer koordinierten und effektiven Fortbewegung für einen mobilen Roboter in natürlichen, sich kontinuierlich verändernden Umgebungen unter sich ebenso bewegenden Hindernissen ist eine komplexe Aufgabe, die die Lösung einer Reihe von Unterproblemen voraussetzt. Die vorliegende Arbeit beschäftigt sich sowohl mit den Themen der Wahrnehmung und Fortbewegung in veränderlichen Umgebungen, als auch mit Methoden zur Analyse der Hindernisbewegungen in Zusammenhang mit der Roboterbewegung selbst. Die Wahrnehmung wird in erster Linie anhand von Laserscannern betrachtet, und ein entsprechendes Verfahren zur Hindernisdetektion und -verfolung wird vorgestellt. Dabei werden Verfahren der globalen Netzwerkoptimierung eingesetzt, um Korrespondenzen zwischen Objekten aus den Einzelbildern herzustellen, was sich positiv auf die Robustheit gegenüber Störungen durch sporadische kleine Objekte auswirkt. Die Navigation basiert auf einer Adaption des sog. "Velocity Obstacle" Ansatzes auf die vorhandene Fahrzeugkinematik, und eine kooperative Bewegungskoordination (Roboter begleitet Mensch) wird durch eine geeignete Auswahlregel für kollisionsfreie Geschwindigkeiten realisiert. Anschließend werden verschiedene Distanzmaße eingeführt, anhand derer sich etwa der Pfad des Roboters mit dem Pfad seiner Begleitperson vergleichen lässt. Weiter wird eine Klassifizierung von Situationen vorgenommen, in die der Roboter potentiell involviert sein kann, und nach einer Übersicht über existierende Ansätze zur automatischen Intentionserkennung wird ein praktikabler Ansatz zur Erkennung gezielter Behinderungen eines mobilen Roboters vorgestellt. Die Arbeit schließt mit einem neuen Ansatz der Bewegungsplanung in dynamischen Umgebungen, der auf rekursiven Modellen des Roboters von seinem Gegenüber basiert, d.h. der Roboter berechnet zunächst, wie er sich in der Situation des (intelligenten, beweglichen) Hindernisses fortbewegen würde, und bezieht dies in die Entscheidung über die eigene Fortbewegung mit ein. Je nach Rekursionstiefe entstehen hierdurch Verhaltensweisen unterschiedlichen Charakters für den Roboter.
Today’s Internet architecture was not designed from scratch but was driven by new services that emerged during its development. Hence, it is often described as patchwork where additional patches are applied in case new services require modifications to the existing architecture. This process however is rather slow and hinders the development of innovative network services with certain architecture or network requirements. Currently discussed technologies like Software-Defined Networking (SDN) or Network Virtualization (NV) are seen as key enabling technologies to overcome this rigid best effort legacy of the Internet. Both technologies offer the possibility to create virtual networks that accommodate the specific needs of certain services. These logical networks are operated on top of a physical substrate and facilitate flexible network resource allocation as physical resources can be added and removed depending on the current network and load situation. In addition, the clear separation and isolation of networks foster the development of application-aware networks that fulfill the special requirements of emerging applications. A prominent use case that benefits from these extended capabilities of the network is denoted with service component mobility. Services hosted on Virtual Machines (VMs) follow their consuming mobile endpoints, so that access latency as well as consumed network resources are reduced. Especially for applications like video streaming, which consume a large fraction of the available resources, is this an important means to relieve the resource constraints and eventually provide better service quality. Service and endpoint mobility both allow an adaptation of the used paths between an offered service, i.e., video streaming and the consuming users in case the service quality drops due to network problems. To make evidence-based adaptations in case of quality drops, a scalable monitoring component is required that is able to monitor the service quality for video streaming applications with reliable accuracy. This monograph details challenges that arise when deploying a certain service, i.e., video streaming, in a future virtualized network architecture and discusses possible solutions. In particular, this work evaluates the performance of mechanisms enabling service mobility and presents an optimized architecture for service mobility. Concerning endpoint mobility, improvements are developed that reduce the latency between endpoints and consumed services and ensure connectivity regardless of the used mobile access network. In the last part, a network-based video quality monitoring solution is developed and its accuracy is evaluated.
The focus of this work lies on the communication issues of Medium Access Control (MAC) and routing protocols in the context of WSNs. The communication challenges in these networks mainly result from high node density, low bandwidth, low energy constraints and the hardware limitations in terms of memory, computational power and sensing capabilities of low-power transceivers. For this reason, the structure of WSNs is always kept as simple as possible to minimize the impact of communication issues. Thus, the majority of WSNs apply a simple one hop star topology since multi-hop communication has high demands on the routing protocol since it increases the bandwidth requirements of the network. Moreover, medium access becomes a challenging problem due to the fact that low-power transceivers are very limited in their sensing capabilities. The first contribution is represented by the Backoff Preamble-based MAC Protocol with Sequential Contention Resolution (BPS-MAC) which is designed to overcome the limitations of low-power transceivers. Two communication issues, namely the Clear Channel Assessment (CCA) delay and the turnaround time, are directly addressed by the protocol. The CCA delay represents the period of time which is required by the transceiver to detect a busy radio channel while the turnaround time specifies the period of time which is required to switch between receive and transmit mode. Standard Carrier Sense Multiple Access (CSMA) protocols do not achieve high performance in terms of packet loss if the traffic is highly correlated due to the fact that the transceiver is not able to sense the medium during the switching phase. Therefore, a node may start to transmit data while another node is already transmitting since it has sensed an idle medium right before it started to switch its transceiver from receive to transmit mode. The BPS-MAC protocol uses a new sequential preamble-based medium access strategy which can be adapted to the hardware capabilities of the transceivers. The protocol achieves a very low packet loss rate even in wireless networks with high node density and event-driven traffic without the need of synchronization. This makes the protocol attractive to applications such as structural health monitoring, where event suppression is not an option. Moreover, acknowledgments or complex retransmission strategies become almost unnecessary since the sequential preamble-based contention resolution mechanism minimizes the collision probability. However, packets can still be lost as a consequence of interference or other issues which affect signal propagation. The second contribution consists of a new routing protocol which is able to quickly detect topology changes without generating a large amount of overhead. The key characteristics of the Statistic-Based Routing (SBR) protocol are high end-to-end reliability (in fixed and mobile networks), load balancing capabilities, a smooth continuous routing metric, quick adaptation to changing network conditions, low processing and memory requirements, low overhead, support of unidirectional links and simplicity. The protocol can establish routes in a hybrid or a proactive mode and uses an adaptive continuous routing metric which makes it very flexible in terms of scalability while maintaining stable routes. The hybrid mode is optimized for low-power WSNs since routes are only established on demand. The difference of the hybrid mode to reactive routing strategies is that routing messages are periodically transmitted to maintain already established routes. However, the protocol stops the transmission of routing messages if no data packets are transmitted for a certain time period in order to minimize the routing overhead and the energy consumption. The proactive mode is designed for high data rate networks which have less energy constraints. In this mode, the protocol periodically transmits routing messages to establish routes in a proactive way even in the absence of data traffic. Thus, nodes in the network can immediately transmit data since the route to the destination is already established in advance. In addition, a new delay-based routing message forwarding strategy is introduced. The forwarding strategy is part of SBR but can also be applied to many routing protocols in order to modify the established topology. The strategy can be used, e.g. in mobile networks, to decrease the packet loss by deferring routing messages with respect to the neighbor change rate. Thus, nodes with a stable neighborhood forward messages faster than nodes within a fast changing neighborhood. As a result, routes are established through nodes with correlated movement which results in fewer topology changes due to higher link durations.
Graphs are a frequently used tool to model relationships among entities. A graph is a binary relation between objects, that is, it consists of a set of objects (vertices) and a set of pairs of objects (edges).
Networks are common examples of modeling data as a graph. For example, relationships between persons in a social network, or network links between computers in a telecommunication network can be represented by a graph.
The clearest way to illustrate the modeled data is to visualize the graphs. The field of Graph Drawing deals with the problem of finding algorithms to automatically generate graph visualizations. The task is to find a "good" drawing, which can be measured by different criteria such as number of crossings between edges or the used area. In this thesis, we study Angular Schematization in Graph Drawing. By this, we mean drawings
with large angles (for example, between the edges at common vertices or at crossing points).
The thesis consists of three parts. First, we deal with the placement of boxes. Boxes are axis-parallel rectangles that can, for example, contain text.
They can be placed on a map to label important sites, or can be used to describe semantic relationships between words in a word network. In the second part of the thesis, we consider graph drawings visually guide the
viewer. These drawings generally induce large angles between edges that meet at a vertex. Furthermore, the edges are drawn crossing-free and in a way that
makes them easy to follow for the human eye. The third and final part is devoted to crossings with large angles. In drawings with crossings, it is important to have large angles between edges at their crossing point, preferably right angles.
In der Arbeit wird ein neues Konzept für Fahrsimulator-Datenbasen vorgestellt. Der Anwender entwirft eine auf seine Fragestellung zugeschnittene Datenbasis mithilfe einer einfachen Skriptsprache. Das Straßennetzwerk wird auf einer topologischen Ebene repäsentiert. In jedem Simulationsschritt wird hieraus im Sichtbarkeitsbereich des Fahrers die geometrische Repäsentation berechnet. Die für den Fahrer unsichtbaren Teile des Straßenetzwerks können während der Simulation verändert werden. Diese Veränderungen können von der Route des Fahrers oder von den in der Simulation erhobenen Messerten abhängen. Zudem kann der Anwender das Straßennetzwerk interaktiv verändern. Das vorgestellte Konzept bietet zahlreiche Möglichkeiten zur Erzeugung reproduzierbarer Szenarien für Experimente in Fahrsimulatoren.
Nowadays, robotics plays an important role in increasing fields of application. There exist many environments or situations where mobile robots instead of human beings are used, since the tasks are too hazardous, uncomfortable, repetitive, or costly for humans to perform. The autonomy and the mobility of the robot are often essential for a good solution of these problems. Thus, such a robot should at least be able to answer the question "Where am I?". This thesis investigates the problem of self-localizing a robot in an indoor environment using range measurements. That is, a robot equipped with a range sensor wakes up inside a building and has to determine its position using only its sensor data and a map of its environment. We examine this problem from an idealizing point of view (reducing it into a pure geometric one) and further investigate a method of Guibas, Motwani, and Raghavan from the field of computational geometry to solving it. Here, so-called visibility skeletons, which can be seen as coarsened representations of visibility polygons, play a decisive role. In the major part of this thesis we analyze the structures and the occurring complexities in the framework of this scheme. It turns out that the main source of complication are so-called overlapping embeddings of skeletons into the map polygon, for which we derive some restrictive visibility constraints. Based on these results we are able to improve one of the occurring complexity bounds in the sense that we can formulate it with respect to the number of reflex vertices instead of the total number of map vertices. This also affects the worst-case bound on the preprocessing complexity of the method. The second part of this thesis compares the previous idealizing assumptions with the properties of real-world environments and discusses the occurring problems. In order to circumvent these problems, we use the concept of distance functions, which model the resemblance between the sensor data and the map, and appropriately adapt the above method to the needs of realistic scenarios. In particular, we introduce a distance function, namely the polar coordinate metric, which seems to be well suited to the localization problem. Finally, we present the RoLoPro software where most of the discussed algorithms are implemented (including the polar coordinate metric).