UMD Theses and Dissertations
Permanent URI for this collectionhttp://hdl.handle.net/1903/3
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 given thesis/dissertation in DRUM.
More information is available at Theses and Dissertations at University of Maryland Libraries.
Browse
2 results
Search Results
Item A Probabilistic Approach to Modeling Socio-Behavioral Interactions(2016) Ramesh, Arti; Getoor, Lise; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)In our ever-increasingly connected world, it is essential to build computational models that represent, reason, and model the underlying characteristics of real-world networks. Data generated from these networks are often heterogeneous, interlinked, and exhibit rich multi-relational graph structures having unobserved latent characteristics. My work focuses on building computational models for representing and reasoning about rich, heterogeneous, interlinked graph data. In my research, I model socio-behavioral interactions and predict user behavior patterns in two important online interaction platforms: online courses and online professional networks. Structured data from these interaction platforms contain rich behavioral and interaction data, and provide an opportunity to design machine learning methods for understanding and interpreting user behavior. The data also contains unstructured data, such as natural language text from forum posts and other online discussions. My research aims at constructing a family of probabilistic models for modeling social interactions involving both structured and unstructured data. In the early part of this thesis, I present a family of probabilistic models for online courses for: 1) modeling student engagement, 2) predicting student completion and dropouts, 3) modeling student sentiment toward various course aspects (e.g., content vs. logistics), 4) detecting coarse and fine-grained course aspects (e.g., grading, video, content), and 5) modeling evolution of topics in repeated offerings of online courses. These methods have the potential to improve student experience and focus limited instructor resources in ways that will have the most impact. In the latter part of this thesis, I present methods to model multi-relational influence in online professional networks. I test the effectiveness of this model via experimentation on the professional network, LinkedIn. My models can potentially be adapted to address a wide range of problems in real-world networks including predicting user interests, user retention, personalization, and making recommendations.Item On the Stability of Structured Prediction(2015) London, Benjamin Alexei; Getoor, Lise; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Many important applications of artificial intelligence---such as image segmentation, part-of-speech tagging and network classification---are framed as multiple, interdependent prediction tasks. These structured prediction problems are typically modeled using some form of joint inference over the outputs, to exploit the relational dependencies. Joint reasoning can significantly improve predictive accuracy, but it introduces a complication in the analysis of structured models: the stability of inference. In optimizations involving multiple interdependent variables, such as joint inference, a small change to the input or parameters could induce drastic changes in the solution. In this dissertation, I investigate the impact of stability in structured prediction. I explore two topics, connected by the stability of inference. First, I provide generalization bounds for learning from a limited number of examples with large internal structure. The effective learning rate can be significantly sharper than rates given in related work. Under certain conditions on the data distribution and stability of the predictor, the bounds decrease with both the number of examples and the size of each example, meaning one could potentially learn from a single giant example. Secondly, I investigate the benefits of learning with strongly convex variational inference. Using the duality between strong convexity and stability, I demonstrate, both theoretically and empirically, that learning with a strongly convex free energy can result in significantly more accurate marginal probabilities. One consequence of this work is a new technique that ``strongly convexifies" many free energies used in practice. These two seemingly unrelated threads are tied by the idea that stable inference leads to lower error, particularly in the limited example setting, thereby demonstrating that inference stability is of critical importance to the study and practice of structured prediction.