Theses and Dissertations from UMD

Permanent URI for this communityhttp://hdl.handle.net/1903/2

New submissions to the thesis/dissertation collections are added automatically as they are received from the Graduate School. Currently, the Graduate School deposits all theses and dissertations from a given semester after the official graduation date. This means that there may be up to a 4 month delay in the appearance of a give thesis/dissertation in DRUM

More information is available at Theses and Dissertations at University of Maryland Libraries.

Browse

Subject: search.filters.subject.Network Science

Search Results

Now showing 1 - 5 of 5
  • Thumbnail Image
    Item
    Developing Methods and Theories for Modeling Student Leadership and Students' Experiences of Academic Support
    (2024) Dalka, Robert Paul; Turpen, Chandra; Physics; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
    This dissertation brings together two research strands that study: (a) the ways in which physics and STEM students contribute to growing capacity for institutional change within collaborative teams and (b) the support structures of graduate programs through an innovative methodology grounded in network science. The first research strand is explored within two different team environments, one of a student-centric interinstitutional team and a second of departmental change teams. Across both of these contexts, I identify how by engaging in an interaction-based agency, students are able to jointly define their own roles and the projects they pursue. In comparing across these contexts, we identify how students navigate different leadership structures and how this can support or limit student contributions in these teams. A central contribution of this work is a model for cultivating capacity for change through shared leadership and relational agency. This model captures how capacity can be built in different domains tied to the activity systems of the work. We show how this model can help practitioners and facilitators better partner with students as well as how researchers can use this model to capture how students contribute to the work of the team. The second research strand focuses on developing and applying an innovative methodology for network analysis of Likert-style surveys. This methodology generates a meaningful network based on survey item response similarity. I show how researchers can use modular analysis of the network to identify larger themes built from the connections of particular aspects. Additionally, I apply this methodology to provide a unique interpretation of responses to the Aspects of Student Experience Scale instrument for well-defined demographic groups to show how thematic clusters identified in the full data set re-emerge or change for different groups of respondents. These results are important for practitioners who seek to make targeted changes to their physics graduate programs in hopes of seeing particular benefits for particular groups. This dissertation opens up lines for future work within both strands. The model for building capacity for change draws attention to the mediating processes that emerge on a team and in students’ interactions with others. This model can be developed further to include additional constructs and leadership structures, as well as explore the relevance to other academic contexts. For quantitative researchers, the network analysis for Likert-style surveys methodology is widely applicable and provides a new way to investigate the wide range of phenomena assessed by Likert-style surveys.
  • Thumbnail Image
    Item
    Understanding Allosteric Communication in Biological Systems using Molecular Dynamics Simulations
    (2024) Samanta, Riya; Matysiak, Silvina; Biophysics (BIPH); Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
    Allostery is critical to survival in living organisms due to its biological relevance in signal transduction, metabolism, and drug discovery. However, the molecular details of this phenomenon remain unclear. In this thesis, I present my work on two allosteric protein systems, each representative of structure-based (E. coli Biotin Protein Ligase) and dynamics-based (B. taurus S100B) allostery. I examined the structural and dynamic features of the proteins and associated variants subjected to various allosteric triggers (ligand/salt/mutations) to study how external/internal perturbations transmit across large distances using Molecular Dyanmic simulations in conjunction with the experiments carried out by our collaborators. Additionally, I carried out Network analyses on the two systems to characterize communication pathways on the protein/ protein family levels. Together, the structural and dynamic features would help us elucidate the underlying mechanism of allostery. The first chapter introduces the two systems with a brief dive into the history of allostery. In the second chapter, my work on E. coli Biotin Protein Ligase and its variants reveal one possible mechanism by which disorder-to-order transitions at the functional surfaces transmit via local changes around the critical residues in the allosteric network. The third chapter explores how the protein network reconfigures to adopt a new allosteric function by studying the allosteric and non-allosteric Biotin Protein Ligases. The fourth chapter elucidates the structural and dynamical markers in bovine S100B, which help to relay information about an allosteric signal by varying two allosteric triggers - ionic strength and target peptide. The final chapter sums up my conclusions, where I propose additional experiments and computational analyses that could be carried out to further our understanding of allostery.
  • Thumbnail Image
    Item
    ANALYZING SEMI-LOCAL LINK COHESION TO DETECT COMMUNITIES AND ANOMALIES IN COMPLEX NETWORKS
    (2021) Schwartz, Catherine; Czaja, Wojciech; Applied Mathematics and Scientific Computation; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
    Link cohesion is a new type of metric used to assess how supported an edge is relativeto other edges, accounting for nearby alternate paths and associated vertex degrees. A deterministic, scalable, and parallelizable link cohesion metric was shown to be useful in supporting edge scoring and simplifying highly connected networks, making key cohesive subgraphs easier to detect. In this dissertation, the link cohesion metric and a modified version of the metric are analyzed to determine their ability to improve the communities detected in different types of networks when used as a pre-weighting step to traditional algorithms like the Louvain method. Additional observations are made on the utility of analyzing the modified metric to gain insights on whether a network has community structure. The two different link cohesion metrics are also used to create vertex-level features that have the potential for being useful in detecting fake accounts in online social networks. These features are used in conjunction with a new interpretable anomaly detection method which performs well with a small amount of training data, yielding the potential for humanin- the-loop interactions that can allow users to tailor the type of anomalies to prioritize.
  • Thumbnail Image
    Item
    UNCOVERING PATTERNS IN COMPLEX DATA WITH RESERVOIR COMPUTING AND NETWORK ANALYTICS: A DYNAMICAL SYSTEMS APPROACH
    (2020) Krishnagopal, Sanjukta; Girvan, Michelle; Physics; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
    In this thesis, we explore methods of uncovering underlying patterns in complex data, and making predictions, through machine learning and network science. With the availability of more data, machine learning for data analysis has advanced rapidly. However, there is a general lack of approaches that might allow us to 'open the black box'. In the machine learning part of this thesis, we primarily use an architecture called Reservoir Computing for time-series prediction and image classification, while exploring how information is encoded in the reservoir dynamics. First, we investigate the ways in which a Reservoir Computer (RC) learns concepts such as 'similar' and 'different', and relationships such as 'blurring', 'rotation' etc. between image pairs, and generalizes these concepts to different classes unseen during training. We observe that the high dimensional reservoir dynamics display different patterns for different relationships. This clustering allows RCs to perform significantly better in generalization with limited training compared with state-of-the-art pair-based convolutional/deep Siamese Neural Networks. Second, we demonstrate the utility of an RC in the separation of superimposed chaotic signals. We assume no knowledge of the dynamical equations that produce the signals, and require only that the training data consist of finite time samples of the component signals. We find that our method significantly outperforms the optimal linear solution to the separation problem, the Wiener filter. To understand how representations of signals are encoded in an RC during learning, we study its dynamical properties when trained to predict chaotic Lorenz signals. We do so by using a novel, mathematical fixed-point-finding technique called directional fibers. We find that, after training, the high dimensional RC dynamics includes fixed points that map to the known Lorenz fixed points, but the RC also has spurious fixed points, which are relevant to how its predictions break down. While machine learning is a useful data processing tool, its success often relies on a useful representation of the system's information. In contrast, systems with a large numbers of interacting components may be better analyzed by modeling them as networks. While numerous advances in network science have helped us analyze such systems, tools that identify properties on networks modeling multi-variate time-evolving data (such as disease data) are limited. We close this gap by introducing a novel data-driven, network-based Trajectory Profile Clustering (TPC) algorithm for 1) identification of disease subtypes and 2) early prediction of subtype/disease progression patterns. TPC identifies subtypes by clustering patients with similar disease trajectory profiles derived from bipartite patient-variable networks. Applying TPC to a Parkinson’s dataset, we identify 3 distinct subtypes. Additionally, we show that TPC predicts disease subtype 4 years in advance with 74% accuracy.
  • Thumbnail Image
    Item
    ADVENTURES ON NETWORKS: DEGREES AND GAMES
    (2015) Pal, Siddharth; Makowski, Armand; La, Richard; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)
    A network consists of a set of nodes and edges with the edges representing pairwise connections between nodes. Examples of real-world networks include the Internet, the World Wide Web, social networks and transportation networks often modeled as random graphs. In the first half of this thesis, we explore the degree distributions of such random graphs. In homogeneous networks or graphs, the behavior of the (generic) degree of a single node is often thought to reflect the degree distribution of the graph defined as the usual fractions of nodes with given degree. To study this preconceived notion, we introduce a general framework to discuss the conditions under which these two degree distributions coincide asymptotically in large random networks. Although Erdos-Renyi graphs along with other well known random graph models satisfy the aforementioned conditions, we show that there might be homogeneous random graphs for which such a conclusion may fail to hold. A counterexample to this common notion is found in the class of random threshold graphs. An implication of this finding is that random threshold graphs cannot be used as a substitute to the Barabasi-Albert model for scale-free network modeling, as proposed in some works. Since the Barabasi-Albert model was proposed, other network growth models were introduced that were shown to generate scale-free networks. We study one such basic network growth model, called the fitness model, which captures the inherent attributes of individual nodes through fitness values (drawn from a fitness distribution) that influence network growth. We characterize the tail of the network-wide degree distribution through the fitness distribution and demonstrate that the fitness model is indeed richer than the Barabasi-Albert model, in that it is capable of producing power-law degree distributions with varying parameters along with other non-Poisson degree distributions. In the second half of the thesis, we look at the interactions between nodes in a game-theoretic setting. As an example, these nodes could represent interacting agents making decisions over time while the edges represent the dependence of their payoffs on the decisions taken by other nodes. We study learning rules that could be adopted by the agents so that the entire system of agents reaches a desired operating point in various scenarios motivated by practical concerns facing engineering systems. For our analysis, we abstract out the network and represent the problem in the strategic-form repeated game setting. We consider two classes of learning rules -- a class of better-reply rules and a new class of rules, which we call, the class of monitoring rules. Motivated by practical concerns, we first consider a scenario in which agents revise their actions asynchronously based on delayed payoff information. We prove that, under the better-reply rules (when certain mild assumptions hold), the action profiles played by the agents converge almost surely to a pure-strategy Nash equilibrium (PSNE) with finite expected convergence time in a large class of games called generalized weakly acyclic games (GWAGs). A similar result is shown to hold for the monitoring rules in GWAGs and also in games satisfying a payoff interdependency structure. Secondly, we investigate a scenario in which the payoff information is unreliable, causing agents to make erroneous decisions occasionally. When the agents follow the better-reply rules and the payoff information becomes more accurate over time, we demonstrate the agents will play a PSNE with probability tending to one in GWAGs. Under a similar setting, when the agents follow the monitoring rule, we show that the action profile weakly converges to certain characterizable PSNE(s). Finally, we study a scenario where an agent might erroneously execute an intended action from time to time. Under such a setting, we show that the monitoring rules ensure that the system reaches PSNE(s) which are resilient to deviations by potentially multiple agents.