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
4 results
Search Results
Item INFERENCE-BASED MODELING, MONITORING, AND CONTROL ALGORITHMS FOR AUTONOMOUS MEDICAL CARE(2022) Tivay, Ali; Hahn, Jin-Oh; Mechanical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Autonomous medical care systems are relatively recent developments in biomedical research that aim to leverage the vigilance, precision, and processing power of computers to assist (or replace) humans in providing medical care to patients. Indeed, past research has demonstrated initial promise for autonomous medical care in applications related to anesthesia, hemodynamic management, and diabetes management, to name a few. However, many of these technologies yet do not exhibit the maturity necessary for widespread real-world adoption and regulatory approval. This can be attributed, in part, to several outstanding challenges associated with the design and development of algorithms that interact with physiological processes. Ideally, an autonomous medical care system should be equipped to exhibit (i) transparent behavior, where the system’s perceptions, reasoning, and decisions are human-interpretable; (ii) context-aware behavior, where the system is capable of remaining mindful of contextual and peripheral information in addition to its primary goal; (iii) coordinated behavior, where the system can coordinate multiple actions in synergistic ways to best achieve multiple objectives; (iv) adaptable behavior, where the system is equipped to identify and adapt to variabilities that exist within and across different patients; and (v) uncertainty-aware behavior, where the system can handle imperfect measurements, quantify the uncertainties that arise as a result, and incorporate them into its decisions. As these desires and challenges are specific to autonomous medical care applications and not fully explored in past research in this area, this dissertation presents a sequence of methodologies to model, monitor, and control a physiological process with special emphasis on addressing these challenges. For this purpose, first, a collective variational inference (C-VI) method is presented that facilitates the creation of personalized and generative physiological models from low-information and heterogeneous datasets. The generative physiological model is of special importance for the purposes of this work, as it encodes physiological knowledge by reproducing the patterned randomness that is observed in physiological datasets. Second, a population-informed particle filtering (PIPF) method is presented that fuses the information encoded in the generative model with real-time clinical data to form perceptions of a patient’s states, characteristics, and events. Third, a population-informed variational control (PIVC) method is presented that leverages the generative model, the perceptions of the PIPF algorithm, and user-defined definitions of actions and rewards in order to search for optimal courses of treatment for a patient. These methods together form a physiological decision-support and closed-loop control (PCLC) framework that is intended to facilitate the desirable behaviors sought in the motivations of this work. The performance, merits, and limitations of this framework are analyzed and discussed based on clinically-important case studies on fluid resuscitation for hemodynamic management.Item Expressive Knowledge Resources in Probabilistic Models(2014) Hu, Yuening; Boyd-Graber, Jordan; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Understanding large collections of unstructured documents remains a persistent problem. Users need to understand the themes of a corpus and to explore documents of interest. Topic models are a useful and ubiquitous tool to discover the main themes (namely topics) of the corpus. Topic models have been successfully applied in natural language processing, computer vision, information retrieval, cognitive science, etc. However, the discovered topics are not always meaningful: some topics confuse two or more themes into one topic; two different topics can be near duplicates; and some topics make no sense at all. Adding knowledge resources into topic models can improve the topics. However, how to encode knowledge into topic models and where to find these knowledge resources remain two scientific challenges. To address these problems, this thesis presents tree-based topic models to encode prior knowledge, a mechanism incorporating knowledge from untrained users, a polylingual tree-based topic model based on existing dictionaries as knowledge resources, an exploration of regularizing spectral methods to encode prior knowledge into topic models, and a model for automatically building hierarchies of prior knowledge for topic models. To encode knowledge resources into topic models, we first present tree-based topic models, where correlations between word types are modeled as a prior tree and applied to topic models. We also develop more efficient inference algorithms for tree- based topic models. Experiments on multiple corpora show that efficiency is greatly improved on different number of topics, number of correlations and vocabulary size. Because users decide whether the topics are useful or not, users' feedback is necessary for effective topic modeling. We thus propose a mechanism for giving normal users a voice to topic models by encoding users' feedback as correlations between word types into tree-based topic models. This framework, interactive topic modeling (ITM), allows untrained users to encode their feedback easily and iteratively into the topic models. We validate the framework both with simulated and real users and discuss strategies for improving the user experience to adapt models to what users need. Existing knowledge resources such as dictionaries can also improve the model. We propose polylingual tree-based topic models based on bilingual dictionaries and apply this model to domain adaptation for statistical Machine Translation. We derive three different inference schemes and evaluate the efficacy of our model on a Chinese to English translation system, and obtain up to 1.2 BLEU improvement over the machine translation baseline. This thesis further explores an alternative way--regularizing spectral methods for topic models--to encode prior knowledge into topic models. Spectral methods offer scalable alternatives to Markov chain Monte Carlo and expectation maximization. However, these new methods lack the priors that are associated with probabilistic models. We examine Arora et al.'s anchor algorithm for topic models and encode prior knowledge by regularizing the anchor algorithm to improve the interpretability and generalizability of topic models. Because existing knowledge resources are limited and because obtaining the knowledge from users is expensive and time-consuming, automatic techniques should also be considered to extract knowledge from the corpus. This thesis further presents a Bayesian hierarchical clustering technique with the Beta coalescent, which provides a possible way to build up the prior tree automatically. Because of its computational complexity, we develop new sampling schemes using sequential Monte carlo and Dirichlet process mixture models, which render the inference practical and efficient. This thesis explores sources of prior knowledge, presents different ways to encode these expressive knowledge resources into probabilistic topic models, and also applies these models in translation domain adaptation. We also discuss further extensions in a bigger picture of interactive machine learning techniques and domain adaptation for downstream tasks.Item Scalable Statistical Modeling and Query Processing over Large Scale Uncertain Databases(2011) Kanagal Shamanna, Bhargav; Deshpande, Amol; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)The past decade has witnessed a large number of novel applications that generate imprecise, uncertain and incomplete data. Examples include monitoring infrastructures such as RFIDs, sensor networks and web-based applications such as information extraction, data integration, social networking and so on. In my dissertation, I addressed several challenges in managing such data and developed algorithms for efficiently executing queries over large volumes of such data. Specifically, I focused on the following challenges. First, for meaningful analysis of such data, we need the ability to remove noise and infer useful information from uncertain data. To address this challenge, I first developed a declarative system for applying dynamic probabilistic models to databases and data streams. The output of such probabilistic modeling is probabilistic data, i.e., data annotated with probabilities of correctness/existence. Often, the data also exhibits strong correlations. Although there is prior work in managing and querying such probabilistic data using probabilistic databases, those approaches largely assume independence and cannot handle probabilistic data with rich correlation structures. Hence, I built a probabilistic database system that can manage large-scale correlations and developed algorithms for efficient query evaluation. Our system allows users to provide uncertain data as input and to specify arbitrary correlations among the entries in the database. In the back end, we represent correlations as a forest of junction trees, an alternative representation for probabilistic graphical models (PGM). We execute queries over the probabilistic database by transforming them into message passing algorithms (inference) over the junction tree. However, traditional algorithms over junction trees typically require accessing the entire tree, even for small queries. Hence, I developed an index data structure over the junction tree called INDSEP that allows us to circumvent this process and thereby scalably evaluate inference queries, aggregation queries and SQL queries over the probabilistic database. Finally, query evaluation in probabilistic databases typically returns output tuples along with their probability values. However, the existing query evaluation model provides very little intuition to the users: for instance, a user might want to know Why is this tuple in my result? or Why does this output tuple have such high probability? or Which are the most influential input tuples for my query ?'' Hence, I designed a query evaluation model, and a suite of algorithms, that provide users with explanations for query results, and enable users to perform sensitivity analysis to better understand the query results.Item Representing and Querying Uncertain Data(2009) Sen, Prithviraj; Getoor, Lise C; Deshpande, Amol V; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)There has been a longstanding interest in building systems that can handle uncertain data. Traditional database systems inherently assume exact data and harbor fundamental limitations when it comes to handling uncertain data. In this dissertation, we present a probabilistic database model that can compactly represent uncertainty models in full generality. Our representation is associated with precise and intuitive semantics and we show that the answer to every user-submitted query can be obtained by performing probabilistic inference. To query large-scale probabilistic databases, we propose a number of techniques that help scale probabilistic inference. Foremost among these techniques is a novel lifted inference algorithm that determines and exploits symmetries in the uncertainty model to speed up query evaluation. For cases when the uncertainty model stored in the database does not contain symmetries, we propose a number of techniques that perform approximate lifted inference. Our techniques for approximate lifted inference have the added advantage of allowing the user to control the degree of approximation through a handful of tunable parameters. Besides scaling probabilistic inference, we also develop techniques that alter the structure of inference required to evaluate a query. More specifically, we show that for a restricted model of our probabilistic database, if each result tuple can be represented by a boolean formula with special characteristics, i.e., it is a read-once function, then the complexity of inference can be drastically reduced. We conclude the dissertation with a listing of directions for future work.