Browsing by Subject "Network science"
Now showing 1 - 3 of 3
Results Per Page
Sort Options
Item Dynamic Learning from Time-Varying Social Networks(2016-05) Baingana, BrianOne of the foremost intellectual challenges of this century is to understand the collective behavior of complex systems. Such systems are ubiquitous, and range from ``engineered systems'' including the Internet and online social grids, to complex natural phenomena such as neural connections in the brain, and interactions between genes. Networks lie at the heart of complex systems, encoding pairwise interactions between their constituent components. In this regard, complexity captures the fact that it is difficult to derive holistic system behavior from knowledge of individual components. The key premise of network science is that despite the diversity of complex systems, the behavior of their underlying networks is driven by a common set of laws. Contemporary studies focus on models and tools to understand, predict, and control the behavior of networks. However, most of these approaches are tailored to analysis of static networks, whose node and link structure does not change with time. Cognizant of the dynamic nature of most real-world networks, analysts mostly focus on static snapshots or aggregate views of studied systems, and meaningful insights cannot be guaranteed. Indeed, the recently growing trend in analysis of dynamic networks is testament to the critical need to live up to this challenge. Moreover, issues arising from temporal network evolution are exacerbated by inherent Big Data challenges. Many large-scale networks comprise billions of nodes, which are typically associated with high-dimensional, and streaming features. Furthermore, it is often impractical to observe the entire network, and analyses must be conducted on manageable or easily accessible samples of the network. Acknowledging these limitations, this dissertation leverages recent advances in statistical signal processing, optimization, and machine learning to address the aforementioned challenges. Emphasis is placed on statistical learning approaches capable of exploiting sparsity, or low rank, attributes that have been shown useful for complexity reduction. Focusing on canonical network inference tasks such as topology identification, detection of communities, and unveiling anomalous nodes, this dissertation puts forth novel statistical models, and develops efficient algorithms for dynamic network analytics. Motivated by the need for real-time processing, online renditions of the developed algorithms are advocated for handling streaming network data. For each of the research themes considered, extensive tests are conducted on simulated and real data, while pertinent comparisons with competing approaches are drawn wherever possible.Item Line structure representation for road network analysis(Journal of Transport and Land Use, 2016) Marshall, StephenRoad hierarchy and network structure are intimately linked; however, there is not a consistent basis for representing and analyzing the particular hierarchical nature of road network structure. This paper introduces the line structure—identified mathematically as a kind of linearly ordered incidence structure—as a means of representing road network structure and demonstrates its relation to existing representations of road networks: the “primal” graph, the “dual” graph, and the route structure. In doing so, the paper shows how properties of continuity, junction type, and hierarchy relating to differential continuity and termination are necessarily absent from primal and dual graph representations but intrinsically present in line structure representations. A new property indicative of hierarchical status—“cardinality”—is introduced and illustrated with application to example networks. The paper concludes by highlighting newly explicit relationships between different kinds of road network structure representation.Item Robust Deep Learning on Graphs(2020-08) Ioannidis, VasileiosThe era of "data deluge'' has sparked the interest in graph-based learning methods and their application in a number of disciplines ranging from sociology and biology to transportation or communications. Realizing the potential of graph-based learning has never been closer, even though formidable challenges are yet there to overcome. Contemporary graphs have massive scale up to billions of nodes, and generate unceasingly "big data''. Graph edges or node attributes may be only partially available due to application specific constraints, which calls for learning approaches to impute the missing information. Graph deep learning methods model complex nonlinear functions and achieve remarkable results in various tasks but the theoretical analysis of such methods is lacking. Last but not least, approaches to learning over graph data must be also robust to adversarial behavior. These challenges have been confronted only partly and separately under different formulations and application domains. The proposed research is centered on analytical and algorithmic foundations that aspire to address the aforementioned challenges facing robust deep learning tasks over large-scale dynamic graphs. The overarching vision is to leverage and adapt state-of-the-art deep learning, optimization and networking tools for inference tasks based on limited graph data. Target applications include identifying node and edge anomalies, predicting node attributes, as well as providing graph-driven recommendations. The ultimate goal is to both analytically and numerically demonstrate how valuable insights from {modeling graph data} can lead to markedly improved learning tools. To this end, the present thesis investigates three main research thrusts: i) unveiling anomalies on graphs; ii) robust deep learning on graphs; and iii) explaining deep learning on graphs via scattering transforms.The aforementioned research thrusts introduce novel methods that aim to tackle the challenges of robust deep learning on graphs. The potential of the proposed approaches is showcased by rigorous theoretical results and extensive experiments.