Networks are models that help us understanding and thinking about relationships between entities in the real world. Many of these networks are dynamic, i.e. connectivity changes over time. Understanding changes in connectivity means to understand interactions between elements of complex systems; how people create and break up friendship relations, how signals get passed in the brain, how business collaborations evolve, or how food-webs restructure after environmental changes. However, understanding static networks is already difficult, due to size, density, attributes and particular motifs; changes over time very much increase this complexity. Quantification of change is often insufficient, but beyond an analysis that is driven by technology and algorithms, humans dispose a unique capability of understanding and interpreting information in data, based on vision and cognition. This dissertation explores ways to interactively explore dynamic networks by means of visualization. I develop and evaluate techniques to unfold the complexity of dynamic networks, making them understandable by looking at them from different angles, decomposing them into their parts and relating the parts in novel ways. While most techniques for dynamic network visualization rely on one particular type of view on the data, complementary visualizations allow for higher-level exploration and analysis. Covering three aspects Tasks, Visualization Design and Evaluation, I develop and evaluate the following unfolding techniques: (i) temporal navigation between individual time steps of a network and improved animated transitions to better understand changes, (ii) designs for the comparison of weighted graphs, (iii) the Matrix Cube, a space-time cube based on adjacency matrices, allowing to visualize dense dynamic networks by, as well as GraphCuisine, a system to (iv) generate synthetic networks with the primary focus on evaluating visualizations in user studies. In order to inform the design and evaluation of visualizations, we (v) provide a task taxonomy capturing users' tasks when exploring dynamic networks. Finally, (vi) the idea of unfolding networks with Matrix Cubes is generalized to other data sets that can be represented in space-time cubes (videos, geographical data, etc.). Visualizations in these domains can inspire visualizations for dynamic networks, and vice-versa. We propose a taxonomy of operations, describing how 3D space-time cubes are decomposed into a large variety of 2D visualizations. These operations help us exploring the design space for visualizing and interactively unfolding dynamic networks and other spatio-temporal data, as well as may serve users as a mental model of the data.

Cette thèse fournit au lecteur un recueil d'applications de la théorie des graphes ; à ce but, des outils sur mesure, adaptés aux applications considérées, ont été conçus et mis en œuvre de manière inspirée par les données.Dans la première partie, une nouvelle métrique de centralité, nommée “bridgeness”, est présentée, basée sur une décomposition de la centralité intermédiaire (“betweenness centrality”) standard. Une composante, la “connectivité locale”, correspondante approximativement au degré d'un noeud, est différenciée de l'autre, qui, en revanche, évalue les propriétés structurelles à longue distance. En effet, cette dernière fournit une mesure de l'efficacité de chaque noeud à “relayer” parties faiblement connectées d'un réseau ; une caractéristique importante de cette métrique est son agnosticisme en ce qui concerne la structure de la communauté sous jacente éventuelle.Une deuxième application vise à décrire les caractéristiques dynamiques des graphes temporels qui apparaissent au niveau mésoscopique. L'ensemble de données de choix comprend 40 ans de publications scientifiques sélectionnées. L'apparition et l'évolution dans le temps d'un domaine d'étude spécifique (les ondelettes) sont capturées, en discriminant les caractéristiques persistantes des artefacts transitoires résultants du processus de détection des communautés, intrinsèquement bruité, effectué indépendamment sur des instantanées statiques successives. La notion de “flux laminaire”, sur laquelle repose le “score de complexité” que nous cherchons à optimiser, est présentée.Dans le même ordre d'idées, un réseau d'investisseurs japonais a été construit, sur la base d'un ensemble de données qui comprend des informations (indirectes) sur les filiales étrangères en copropriété. Une question très débattue dans le domaine de l'économie industrielle, l'hypothèse de Miwa-Ramseyer, a été démontrée de manière concluante comme fausse, du moins sous sa forme forte. / This thesis provides the reader with a compendium of applications of network theory; tailor-madetools suited for the purpose have been devised and implemented in a data-driven fashion. In the first part, a novel centrality metric, aptly named “bridgeness”, is presented, based on adecomposition of the standard betweenness centrality. One component, local connectivity, roughlycorresponding to the degree of a node, is set apart from the other, which evaluates longer-rangestructural properties. Indeed, the latter provides a measure of the relevance of each node in“bridging” weakly connected parts of a network; a prominent feature of the metric is its agnosticism with regard to the eventual ground truth community structure.A second application is aimed at describing dynamic features of temporal graphs which are apparent at the mesoscopic level. The dataset of choice includes 40 years of selected scientific publications.The appearance and evolution in time of a specific field of study (“wavelets”) is captured,discriminating persistent features from transient artifacts, which result from the intrinsically noisy community detection process, independently performed on successive static snapshots. The concept of “laminar stream”, on which the “complexity score” we seek to optimize is based, is introduced.In a similar vein, a network of Japanese investors has been constructed, based on a dataset which includes (indirect) information on co-owned overseas subsidiaries. A hotly debated issue in the field of industrial economics, the Miwa-Ramseyer hypothesis, has been conclusively shown to be false, at least in its strong form.

Les réseaux sont des modèles qui nous permettent de comprendre les relations entre éléments du monde réel. Une grande quantité de réseaux sont dynamiques, c'est-à-dire que leur connexité change au cours du temps. Comprendre les changements de connexité signifie comprendre les interactions entre les éléments de systèmes complexes: comment se forment les relations sociales et commerciales, comment sont transmis les signaux entre les régions du cerveau, comment s'organisent les réseaux trophiques après des catastrophes environnementales. Au-delà de ce que nous permettent la technologie et les algorithmes d'analyses, l'homme dispose d'une capacité unique pour comprendre et interpréter des informations : la vision et la cognition. Cette thèse développe et examine des moyens pour explorer les réseaux dynamiques d'une manière interactive et visuelle. Je propose des techniques pour déplier la complexité des réseaux, avec le but de les rendre compréhensibles, de les voir à partir de perspectives différentes, d'examiner leurs composantes. Déplier des réseaux est une métaphore, comme la création des cartes bidimensionelles d'objets tridimensionnels comme la Terre: chaque méthode de projection a comme résultat une carte différente qui permet de voir des relations différentes entre la taille des continents et des océans, des distances, etc. Je propose les techniques de dépliage suivantes, implémentées et évaluées dans des systèmes interactifs : (i) une navigation temporelle qui permet de naviguer plus efficacement entre des différents instants, ainsi qu'un feedback visuel qui permet de mieux comprendre les changements dans les réseaux entre deux instants arbitraires. (ii) Des designs permettant la comparaison directe de deux réseaux avec des liens pondérés. (iii) Un modèle de visualisation pour des réseaux denses avec des liens pondérés, ainsi que (iv) la génération de réseaux synthétiques utilisés pour l'évaluation des visualisations. Afin de mieux créer et évaluer des visualisations, nous (v) proposons une taxonomie de tâche pour décrire des tâches accomplies par des analystes des réseaux. Pour compléter, (vi) nous généralisons l'idée de dépliage pour décrire d'autres genres de données temporelles, représentable dans des cubes espace-temps. Cela concerne la visualisation de vidéos, des données multi-variées, ainsi que la géographique. Une telle généralisation a pour but de fournir une base commune pour échanger des techniques de visualisation et de mieux comprendre l'espace de design pour les réseaux dynamiques. Dans cette optique, nous proposons une taxonomie d'opérations génériques qui nous permet de transformer un cube espace-temps en visualisation bidimensionelle, ainsi qu'une description des formes évoquées par les données dans le cube espace-temps. / Networks are models that help us understanding and thinking about relationships between entities in the real world. Many of these networks are dynamic, i.e. connectivity changes over time. Understanding changes in connectivity means to understand interactions between elements of complex systems; how people create and break up friendship relations, how signals get passed in the brain, how business collaborations evolve, or how food-webs restructure after environmental changes. However, understanding static networks is already difficult, due to size, density, attributes and particular motifs; changes over time very much increase this complexity. Quantification of change is often insufficient, but beyond an analysis that is driven by technology and algorithms, humans dispose a unique capability of understanding and interpreting information in data, based on vision and cognition. This dissertation explores ways to interactively explore dynamic networks by means of visualization. I develop and evaluate techniques to unfold the complexity of dynamic networks, making them understandable by looking at them from different angles, decomposing them into their parts and relating the parts in novel ways. While most techniques for dynamic network visualization rely on one particular type of view on the data, complementary visualizations allow for higher-level exploration and analysis. Covering three aspects Tasks, Visualization Design and Evaluation, I develop and evaluate the following unfolding techniques: (i) temporal navigation between individual time steps of a network and improved animated transitions to better understand changes, (ii) designs for the comparison of weighted graphs, (iii) the Matrix Cube, a space-time cube based on adjacency matrices, allowing to visualize dense dynamic networks by, as well as GraphCuisine, a system to (iv) generate synthetic networks with the primary focus on evaluating visualizations in user studies. In order to inform the design and evaluation of visualizations, we (v) provide a task taxonomy capturing users' tasks when exploring dynamic networks. Finally, (vi) the idea of unfolding networks with Matrix Cubes is generalized to other data sets that can be represented in space-time cubes (videos, geographical data, etc.). Visualizations in these domains can inspire visualizations for dynamic networks, and vice-versa. We propose a taxonomy of operations, describing how 3D space-time cubes are decomposed into a large variety of 2D visualizations. These operations help us exploring the design space for visualizing and interactively unfolding dynamic networks and other spatio-temporal data, as well as may serve users as a mental model of the data.

Wireless networks have potential applications in wireless Internet connectivity, battlefields, disaster relief, and cyber-physical systems. While the nodes in these networks communicate with each other over the air, the challenges faced by and the subsequent design criteria of these networks are diverse. In this thesis, we study and discuss a few design requirements of these networks, such as efficient utilization of the network bandwidth in IEEE 802.11 infrastructure networks, evaluating utility of sensor node deployments, and security from eavesdroppers. The presence of infrastructure IEEE 802.11 based Wireless Local Area Networks (WLANs) allows mobile users to seamlessly transfer huge volumes of data. While these networks accommodate mobility, and are a cost-effective alternative to cellular networks, they are well known to display several performance anomalies. We study a few such anomalies, and provide a performance management solution for IEEE 802.11 based WLANs. On the other hand, in sensor networks, the absence of infrastructure mandates the use of adhoc network architectures. In these architectures, nodes are required to route data to gateway nodes over a multi-hop network. These gateway nodes are larger in size, and costlier in comparison with the regular nodes. In this context, we propose a unified framework that can be used to compare different deployment scenarios, and provide a means to design efficient large-scale adhoc networks. In modern times, security has become an additional design criterion in wireless networks. Traditionally, secure transmissions were enabled using cryptographic schemes. However, in recent years, researchers have explored physical layer security as an alternative to these traditional cryptographic schemes. Physical layer security enables secure transmissions at non-zero data rate between two communicating nodes, by exploiting the degraded nature of the eavesdropper channel and the inherent randomness of the wireless medium. Also, in many practical scenarios, several nodes cooperate to improve their individual secrecy rates. Therefore, in this thesis, we also study scenarios, where cooperative schemes can improve secure end-to-end data transmission rates, while adhering to an overall power budget. In spite of the presence of voluminous reservoirs of information such as digital libraries and the Internet, asking around still remains a popular means of seeking information. In scenarios where the person is interested in communal, or location-specific information, such kind of retrieval may yield better results than a global search. Hence, wireless networks should be designed, analyzed and controlled by taking into account the evolution of the underlying social networks. This alliance between social network analysis and adhoc network architectures can greatly advance the design of network protocols, especially in environments with opportunistic communications. Therefore, in addition to the above mentioned problem, in this thesis, we have also presented and studied a model that captures the temporal evolution of information in social networks with memory.

Menschliche Systeme werden seit einiger Zeit modelliert und analysiert auf der Basis der Theorie komplexer Netzwerke. Dies erlaubt es quantitativ zu untersuchen, welche strukturellen und zeitlichen Merkmale eines Systems Ausbreitungsprozesse beeinflussen, z.B. von Informationen oder von Infektionskrankheiten. Im ersten Teil der Arbeit wird untersucht, wie eine modular-hierarchische Struktur von statischen Netzwerken eine schnelle Verbreitung von Signalen ermöglicht. Es werden neue Heuristiken entwickelt um die Random-Walk-Observablen “First Passage Time” und “Cover Time” auf lokal geclusterten Netzwerken zu ermitteln. Vergleiche mit der Approximation eines gemittelten Mediums zeigen, dass das Auftreten der beobachteten Minima der Observablen ein reiner Netzwerkeffekt ist. Es wird weiterhin dargelegt, dass nicht alle modular-hierarchischen Netzwerkmodelle dieses Phänomen aufweisen. Im zweiten Teil werden zeitlich veränderliche face-to-face Kontaktnetzwerke auf ihre Anfälligkeit für Infektionskrankheiten untersucht. Mehrere Studien belegen, dass Menschen vornehmlich Zeit in Isolation oder kleinen, stark verbundenen Gruppen verbringen, und dass ihre Kontaktaktivität einem zirkadianen Rhythmus folgt. Inwieweit diese beiden Merkmale die Ausbreitung von Krankheiten beeinflussen, ist noch unklar. Basierend auf einem neuen Modell wird erstmals gezeigt, dass zirkadian variierende Netzwerke Trajektorien folgen in einem Zustandsraum mit einer strukturellen und einer zeitlichen Dimension. Weiterhin wird dargelegt, dass mit zunehmender Annäherung der zeitlichen Dimension von System und Krankheit die systemische Infektionsanfälligkeit sinkt. Dies steht in direktem Widerspruch zu Ergebnissen anderer Studien, die eine zunehmende Anfälligkeit vorhersagen, eine Diskrepanz, die auf die Ungültigkeit einer weit verbreiteten Approximation zurückzuführen ist. Die hier vorgestellten Ergebnisse implizieren, dass auf dem Gebiet die Entwicklung neuer theoretischer Methoden notwendig ist. / Human systems have been modeled and analyzed on the basis of complex networks theory in recent time. This abstraction allows for thorough quantitative analyses to investigate which structural and temporal features of a system influence the evolution of spreading processes, such as the passage of information or of infectious diseases. The first part of this work investigates how the ubiquitous modular hierarchical structure of static real-world networks allows for fast delivery of messages. New heuristics are developed to evaluate random walk mean first passage times and cover times on locally clustered networks. A comparison to average medium approximations shows that the emergence of these minima are pure network phenomena. It is further found that not all modular hierarchical network models provide optimal message delivery structure. In the second part, temporally varying face-to-face contact networks are investigated for their susceptibility to infection. Several studies have shown that people tend to spend time in small, densely-connected groups or in isolation, and that their connection behavior follows a circadian rhythm. To what extent both of these features influence the spread of diseases is as yet unclear. Therefore, a new temporal network model is devised here. Based on this model, circadially varying networks can for the first time be interpreted as following trajectories through a newly defined systemic state space. It is further revealed that in many temporally varying networks the system becomes less susceptible to infection when the time-scale of the disease approaches the time-scale of the network variation. This is in direct conflict with findings of other studies that predict increasing susceptibility of temporal networks, a discrepancy which is attributed to the invalidity of a widely applied approximation. The results presented here imply that new theoretical advances are necessary to study the spread of diseases in temporally varying networks.

The ability of human beings to perceive the environment around them with their visual system is called motion perception. This means that the attention of our visual system is primarily focused on those objects that are moving. The property of human motion perception is used in this dissertation to infer human activity from data using artificial neural networks. One of the main aims of this thesis is to discover which modalities, namely RGB images, optical flow and human keypoints, are best suited for HAR in omnidirectional data. Since these modalities are not yet available for omnidirectional cameras, they are synthetically generated and captured with an omnidirectional camera. During data generation, a distinction is made between synthetically generated omnidirectional data and a real omnidirectional dataset that was recorded in a Living Lab at Chemnitz University of Technology and subsequently annotated by hand. The synthetically generated dataset, called OmniFlow, consists of RGB images, optical flow in forward and backward directions, segmentation masks, bounding boxes for the class people, as well as human keypoints. The real-world dataset, OmniLab, contains RGB images from two top-view scenes as well as manually annotated human keypoints and estimated forward optical flow. In this thesis, the generation of the synthetic and real-world datasets is explained. The OmniFlow dataset is generated using the 3D rendering engine Blender, in which a fully configurable 3D indoor environment is created with artificially textured rooms, human activities, objects and different lighting scenarios. A randomly placed virtual camera following the omnidirectional camera model renders the RGB images, all other modalities and 15 predefined activities. The result of modelling the 3D indoor environment is the OmniFlow dataset. Due to the lack of omnidirectional optical flow data, the OmniFlow dataset is validated using Test-Time Augmentation (TTA). Compared to the baseline, which contains Recurrent All-Pairs Field Transforms (RAFT) trained on the FlyingChairs and FlyingThings3D datasets, it was found that only about 1000 images need to be used for fine-tuning to obtain a very low End-point Error (EE). Furthermore, it was shown that the influence of TTA on the test dataset of OmniFlow affects EE by about a factor of three. As a basis for generating artificial keypoints on OmniFlow with action labels, the Carnegie Mellon University motion capture database is used with a large number of sports and household activities as skeletal data defined in the BVH format. From the BVH-skeletal data, the skeletal points of the people performing the activities can be directly derived or extrapolated by projecting these points from the 3D world into an omnidirectional 2D image. The real-world dataset, OmniLab, was recorded in two rooms of the Living Lab with five different people mimicking the 15 actions of OmniFlow. Human keypoint annotations were added manually in two iterations to reduce the error rate of incorrect annotations. The activity-level evaluation was investigated using a TSN and a PoseC3D network. The TSN consists of two CNNs, a spatial component trained on RGB images and a temporal component trained on the dense optical flow fields of OmniFlow. The PoseC3D network, an approach to skeleton-based activity recognition, uses a heatmap stack of keypoints in combination with 3D convolution, making the network more effective at learning spatio-temporal features than methods based on 2D convolution. In the first step, the networks were trained and validated on the synthetically generated dataset OmniFlow. In the second step, the training was performed on OmniFlow and the validation on the real-world dataset OmniLab. For both networks, TSN and PoseC3D, three hyperparameters were varied and the top-1, top-5 and mean accuracy given. First, the learning rate of the stochastic gradient descent (Stochastic Gradient Descent (SGD)) was varied. Secondly, the clip length, which indicates the number of consecutive frames for learning the network, was varied, and thirdly, the spatial resolution of the input data was varied. For the spatial resolution variation, five different image sizes were generated from the original dataset by cropping from the original dataset of OmniFlow and OmniLab. It was found that keypoint-based HAR with PoseC3D performed best compared to human activity classification based on optical flow and RGB images. This means that the top-1 accuracy was 0.3636, the top-5 accuracy was 0.7273 and the mean accuracy was 0.3750, showing that the most appropriate output resolution is 128px × 128px and the clip length is at least 24 consecutive frames. The best results could be achieved with a learning rate of PoseC3D of 10-3. In addition, confusion matrices indicating the class-wise accuracy of the 15 activity classes have been given for the modalities RGB images, optical flow and human keypoints. The confusion matrix for the modality RGB images shows the best classification result of the TSN for the action walk with an accuracy of 1.00, but almost all other actions are also classified as walking in real-world data. The classification of human actions based on optical flow works best on the action sit in chair and stand up with an accuracy of 1.00 and walk with 0.50. Furthermore, it is noticeable that almost all actions are classified as sit in chair and stand up, which indicates that the intra-class variance is low, so that the TSN is not able to distinguish between the selected action classes. Validated on real-world data for the modality keypoint the actions rugpull (1.00) and cleaning windows (0.75) performs best. Therefore, the PoseC3D network on a time-series of human keypoints is less sensitive to variations in the image angle between the synthetic and real-world data than for the modalities RGB images and optical flow. The pipeline for the generation of synthetic data with regard to a more uniform distribution of the motion magnitudes needs to be investigated in future work. Random placement of the person and other objects is not sufficient for a complete coverage of all movement magnitudes. An additional improvement of the synthetic data could be the rotation of the person around their own axis, so that the person moves in a different direction while performing the activity and thus the movement magnitudes contain more variance. Furthermore, the domain transition between synthetic and real-world data should be considered further in terms of viewpoint invariance and augmentation methods. It may be necessary to generate a new synthetic dataset with only top-view data and re-train the TSN and PoseC3D. As an augmentation method, for example, the Fourier Domain Adaption (FDA) could reduce the domain gap between the synthetically generated and the real-world dataset.:1 Introduction 2 Theoretical Background 3 Related Work 4 Omnidirectional Synthetic Human Optical Flow 5 Human Keypoints for Pose in Omnidirectional Images 6 Human Activity Recognition in Indoor Scenarios 7 Conclusion and Future Work A Chapter 4: Flow Dataset Statistics B Chapter 5: 3D Rotation Matrices C Chapter 6: Network Training Parameters