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
52 results
Search Results
Item Gender Effects on Knee Loading and Prediction of Knee Loads Using Instrumented Insoles and Machine Learning(2024) Snyder, Samantha Jane; Miller, Ross H.; Shim, Jae Kun; Kinesiology; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Women are more likely to experience knee osteoarthritis as compared to men, but the underlying mechanisms behind this disparity are unclear. Greater knee loads, knee adduction moment, knee flexion moment, and medial joint contact force, are linked to severity and progression of knee osteoarthritis. However, it is unknown if greater knee loads in healthy, young women during activities of daily living (sit-to-stand, stand-to-sit, walking and running) can partially explain the higher prevalence of knee osteoarthritis rates in women. Although previous research showed no significant differences in peak knee adduction moment and knee flexion moment between men and women, differences in peak medial joint contact force are largely unexplored. Women also tend to take shorter steps and run slower than men. It is unknown if these differences may result in greater cumulative knee loading per unit distance traveled as compared to men. Furthermore, knee loading measurement is typically confined to a gait laboratory, yet the knee is subjected to large cyclical loads throughout daily life. The combination of machine learning techniques and wearable sensors has been shown to improve accessibility of biomechanical measurements without compromising accuracy. Therefore, the goal of this dissertation is to develop a framework for measuring these risk factors using machine learning and novel instrumented insoles, and to investigate differences in peak and cumulative per unit distance traveled knee loads between young, healthy men and women. In study 1 we developed instrumented insoles and examined insole reliability and validity. In study 2, we estimated knee loads for most activities with strong correlation coefficients and low to moderate mean absolute errors. In study 3, we found peak medial joint contact force was not significantly different across activities for men and women. Similarly, in study 4, we found no significant difference between men and women in knee loads per unit distance traveled during walking and running. These findings suggest biomechanical mechanisms alone cannot explain the disproportionate rate of knee osteoarthritis in women. However, in future research, the developed knee loading prediction models can help quantify daily knee loads and aid in reducing knee osteoarthritis risk in both men and women.Item Efficient Models and Learning Strategies for Resource-Constrained Systems(2024) Rabbani, Tahseen Wahed Karim; Huang, Furong; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)The last decade has seen sharp improvements in the performance of machine learning (ML) models but at the cost of vastly increased complexity and size of their underlying architectures. Advances in high-performance computing have enabled researchers to train and deploy models composed of hundreds of billions of parameters. However, harvesting the full utility of large models on smaller clients, such as Internet of Things (IoT) devices, without resorting to external hosting will require a significant reduction of parameters and faster, cheaper inference. In addition to augmenting IoT, efficient models and learning paradigms can reduce energy consumption, encourage technological equity, and are well-suited for deployment in real-world applications that require fast response in low-resource settings. To address these challenges, we introduce multiple, novel strategies for (1) reducing the scale of deep neural networks and (2) faster learning. For the size problem (1), we leverage tools such as tensorization, randomized projections, and locality-sensitive hashing to train on reduced representations of large models without sacrificing performance. For learning efficiency (2), we develop algorithms for cheaper forward passes, accelerated PCA, and asynchronous gradient descent. Several of these methods are tailored for federated learning (FL), a private, distributed learning paradigm where data is decentralized among resource-constrained edge clients. We are exclusively concerned with improving efficiency during training -- our techniques do not process pre-trained models or require a device to train over an architecture in its full entirety.Item DEVELOPMENT AND APPLICATION OF PROPINQUITY MODELING FRAMEWORK FOR IDENTIFICATION AND ANALYSIS OF EXTREME EVENT PATTERNS(2024) kholodovsky, vitaly; Liang, Xin-Zhong; Atmospheric and Oceanic Sciences; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Extreme weather and climate events such as floods, droughts, and heat waves can cause extensive societal damage. While various statistical and climate models have been developed for the purpose of simulating extremes, a consistent definition of extreme events is still lacking. Furthermore, to better assess the performance of the climate models, a variety of spatial forecast verification measures have been developed. However, in most cases, the spatial verification measures that are widely used to compare mean states do not have sufficient theoretical justification to benchmark extreme events. In order to alleviate inconsistencies when defining extreme events within different scientific communities, we propose a new generalized Spatio-Temporal Threshold Clustering method for the identification of extreme event episodes, which uses machine learning techniques to couple existing pattern recognition indices with high or low threshold choices. The method consists of five main steps: construction of essential field quantities, dimension reduction, spatial domain mapping, time series clustering, and threshold selection. We develop and apply this method using a gridded daily precipitation dataset derived from rain gauge stations over the contiguous United States. We observe changes in the distribution of conditional frequency of extreme precipitation from large-scale, well-connected spatial patterns to smaller-scale, more isolated rainfall clusters, possibly leading to more localized droughts and heatwaves, especially during the summer months. Additionally, we compare empirical and statistical probabilities and intensities obtained through the Conventional Location Specific methods, which are deficient in geometric interconnectivity between individual spatial pixels and independent in time, with a new Propinquity modeling framework. We integrate the Spatio-Temporal Threshold Clustering algorithm and the conditional semi-parametric Heffernan and Tawn (2004) model into the Propinquity modeling framework to separate classes of models that can calculate process level dependence of large-scale extreme processes, primarily through the overall extreme spatial field. Our findings reveal significant differences between Propinquity and Conventional Location Specific methods, in both empirical and statistical approaches in shape and trend direction. We also find that the process of aggregating model results without considering interconnectivity between individual grid cells for trend construction can lead to significant variations in the overall trend pattern and direction compared with models that do account for interconnectivity. Based on these results, we recommend avoiding such practices and instead adopting the Propinquity modeling framework or other spatial EVA models that take into account the interconnectivity between individual grid cells. Our aim for the final application is to establish a connection between extreme essential field quantity intensity fields and large-scale circulation patterns. However, the Conventional Location Specific Threshold methods are not appropriate for this purpose as they are memoryless in time and not able to identify individual extreme episodes. To overcome this, we developed the Feature Finding Decomposition algorithm and used it in combination with the Propinquity modeling framework. The algorithm consists of the following three steps: feature finding, image decomposition, and large-scale circulation patterns connection. Our findings suggest that the Western Pacific Index, particularly its 5th percentile and 5th mode of decomposition, is the most significant teleconnection pattern that explains the variation in the trend pattern of the largest feature intensity.Item QUANTUM SIMULATION OF BOSONIC SYSTEM AND APPLICATION OF MACHINE LEARNING(2023) Kuo, En-Jui; Hafezi, Mohammad; Physics; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)First, we introduce the notion of "generalized bosons," whose exchange statistics resemble those of bosons, but the local bosonic commutator $[a_i,a_i^{\dagger}]=1$ is replaced by an arbitrary single-mode operator that is diagonal in the generalized Fock basis. Examples of generalized bosons include boson pairs and spins. We consider the analogue of the boson sampling task for these particles and observe that its output probabilities are still given by permanents, so the results regarding the difficulty of sampling carry over directly. Finally, we propose implementations of generalized boson sampling in circuit-QED and ion-trap platforms. In the rest of the thesis, we move on to different topics. Firstly, we incorporate machine learning techniques in quantum information. We use machine learning to classify rational two-dimensional conformal field theories (CFTs). We first use the energy spectra of these minimal models to train a supervised learning algorithm. In contrast to conventional methods that are typically qualitative and involve system size scaling, our method quantifies the similarity of the spectrum of a system at a fixed size to candidate CFTs. Such an approach allows us to correctly predict the nature and value of critical points of several strongly correlated spin models using only their energy spectra. Our results are also relevant for the ground-state entanglement Hamiltonian of certain topological phases of matter described by CFTs. Remarkably, we achieve high prediction accuracy by only using the lowest few Rényi entropies as the input. Finally, using autoencoders, an unsupervised learning algorithm, we find a hidden variable that has a direct correlation with the central charge and discuss prospects for using machine learning to investigate other conformal field theories, including higher-dimensional ones. Next, we demonstrate how machine learning techniques, especially unsupervised learning algorithms, can be used to study Symmetry-Protected Topological (SPT) phases of matter. SPT phases are short-range entangled phases of matter with a non-local order parameter that are preserved under a local symmetry group. Here, we use an unsupervised learning algorithm, namely diffusion maps, to differentiate between symmetry-broken phases and topologically ordered phases and between non-trivial topological phases in different classes. Specifically, we show that phase transitions associated with these phases can be detected in various bosonic and fermionic models in one dimension, including the interacting SSH model, the AKLT model and its variants, and weakly interacting fermionic models. Our approach provides a cost-effective computational method for detecting topological phase transitions associated with SPT systems, which can also be applied to experimental data obtained from quantum simulators.Item APPLICATIONS OF ARTIFICIAL NEURAL NETWORKS IN LEARNING QUANTUM SYSTEMS(2023) Pan, Ruizhi; Clark, Charles; Physics; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Quantum machine learning is an emerging field that combines techniques in the disciplines of machine learning (ML) and quantum physics. Research in this field takes three broad forms: applications of classical ML techniques to quantum physical systems, quantum computing and algorithms for classical ML problems, and new ideas inspired by the intersection of the two disciplines. We mainly focus on the power of artificial neural networks (NNs) in quantum-state representation and phase classification in this work. In the first part of the dissertation, we study NN quantum states which are used as wave-function ans{\" a}tze in the context of quantum many-body physics. While these states have achieved success in simulating low-lying eigenstates and short-time unitary dynamics of quantum systems and efficiently representing particular states such as those with a stabilizer nature, more rigorous quantitative analysis about their expressibility and complexity is warranted. Here, our analysis of the restricted Boltzmann machine (RBM) state representation of one-dimensional (1D) quantum spin systems provides new insight into their computational complexity. We define a class of long-range-fast-decay (LRFD) RBM states with quantifiable upper bounds on truncation errors and provide numerical evidence for a large class of 1D quantum systems that may be approximated by LRFD RBMs of at most polynomial complexities. These results lead us to conjecture that the ground states of a wide range of quantum systems may be exactly represented by LRFD RBMs or a variant of them, even in cases where other state representations become less efficient. At last, we provide the relations between multiple typical state manifolds. Our work proposes a paradigm for doing complexity analysis for generic long-range RBMs which naturally yields a further classification of this manifold. This paradigm and our characterization of their nonlocal structures may pave the way for understanding the natural measure of complexity for quantum many-body states described by RBMs and are generalizable for higher-dimensional systems and deep neural-network quantum states. In the second part, we use RBMs to investigate, in dimensions $D=1$ and $2$, the many-body excitations of long-range power-law interacting quantum spin models. We develop an energy-shift method to calculate the excited states of such spin models and obtain a high-precision momentum-resolved low-energy spectrum. This enables us to identify the critical exponent where the maximal quasiparticle group velocity transits from finite to divergent in the thermodynamic limit numerically. In $D=1$, the results agree with an analysis using the field theory and semiclassical spin-wave theory. Furthermore, we generalize the RBM method for learning excited states in nonzero-momentum sectors from 1D to 2D systems. At last, we analyze and provide all possible values ($3/2$, $2$ and $3$) of the critical exponent for 1D generic quadratic bosonic and fermionic Hamiltonians with long-range hoppings and pairings which serves for understanding the speed of information propagation in quantum systems. In the third part, we study deep NNs as phase classifiers. We analyze the phase diagram of a 2D topologically nontrivial fermionic model Hamiltonian with pairing terms at first and then demonstrate that deep NNs can learn the band-gap closing conditions only based on wave-function samples of several typical energy eigenstates, thus being able to identify the phase transition point without knowledge of Hamiltonians.Item Efficient Machine Learning Techniques for Neural Decoding Systems(2022) wu, xiaomin; Bhattacharyya, Shuvra S.; Chen, Rong; Electrical Engineering; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)In this thesis, we explore efficient machine learning techniques for calcium imaging based neural decoding in two directions: first, techniques for pruning neural network models to reduce computational complexity and memory cost while retaining high accuracy; second, new techniques for converting graph-based input into low-dimensional vector form, which can be processed more efficiently by conventional neural network models. Neural decoding is an important step in connecting brain activity to behavior --- e.g., to predict movement based on acquired neural signals. Important application areas for neural decoding include brain-machine interfaces and neuromodulation. For application areas such as these, real-time processing of neural signals is important as well as high quality information extraction from the signals. Calcium imaging is a modality that is of increasing interest for studying brain activity. Miniature calcium imaging is a neuroimaging modality that can observe cells in behaving animals with high spatial and temporalresolution, and with the capability to provide chronic imaging. Compared to alternative modalities, calcium imaging has potential to enable improved neural decoding accuracy. However, processing calcium images in real-time is a challenging task as it involves multiple time-consuming stages: neuron detection, motion correction, and signal extraction. Traditional neural decoding methods, such as those based on Wiener and Kalman filters, are fast; however, they are outperformed in terms of accuracy by recently-developed deep neural network (DNN) models. While DNNs provide improved accuracy, they involve high computational complexity, which exacerbates the challenge of real-time processing. Addressing the challenges of high-accuracy, real-time, DNN-based neural decoding is the central objective of this research. As a first step in addressing these challenges, we have developed the NeuroGRS system. NeuroGRS is designed to explore design spaces for compact DNN models and optimize the computational complexity of the models subject to accuracy constraints. GRS, which stands for Greedy inter-layer order with Random Selection of intra-layer units, is an algorithm that we have developed for deriving compact DNN structures. We have demonstrated the effectiveness of GRS to transform DNN models into more compact forms that significantly reduce processing and storage complexity while retaining high accuracy. While NeuroGRS provides useful new capabilities for deriving compact DNN models subject to accuracy constraints, the approach has a significant limitation in the context of neural decoding. This limitation is its lack of scalability to large DNNs. Large DNNs arise naturally in neural decoding applications when the brain model under investigation involves a large number of neurons. As the size of the input DNN increases, NeuroGRS becomes prohibitively expensive in terms of computationaltime. To address this limitation, we have performed a detailed experimental analysis of how pruned solutions evolve as GRS operates, and we have used insights from this analysis to develop a new DNN pruning algorithm called Jump GRS (JGRS). JGRS maintains similar levels of model quality --- in terms of predictive accuracy --- as GRS while operating much more efficiently and being able to handle much larger DNNs under reasonable amounts of time and reasonable computational resources. Jump GRS incorporates a mechanism that bypasses (``jumps over'') validation and retraining during carefully-selected iterations of the pruning process. We demonstrate the advantages and improved scalability of JGRS compared to GRS through extensive experiments in the context of DNNs for neural decoding. We have also developed methods for raising the level of abstraction in the signal representation used for calcium imaging analysis. As a central part of this work, we invented the WGEVIA (Weighted Graph Embedding with Vertex Identity Awareness) algorithm, which enables DNN-based processing of neuron activity that is represented in the form of microcircuits. In contrast to traditional representations of neural signals, which involve spiking signals, a microcircuit representation is a graphical representation. Each vertex in a microcircuit corresponds to a neuron, and each edge carries a weight that captures information about firing relationships between the neurons associated with the vertices that are incident to the edge. Our experiments demonstrate that WGEVIA is effective at extracting information from microcircuits. Moreover,raising the level of abstraction to microcircuit analysis has the potential to enable more powerful signal extraction under limited processing time and resources.Item USING SOCIAL MEDIA AS A DATA SOURCE IN PUBLIC HEALTH RESEARCH(2022) Sigalo, Nekabari; Frias-Martinez, Vanessa; Library & Information Services; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Researchers have increasingly looked to social media data as a means of measuring population health and well-being in a less intrusive and more scalable manner compared to traditional public health data sources. In this dissertation, I outline three studies that leverage social media as a data source, to answer research questions related to public health and compare traditional public health data sources to social media data sources. In Study #1, I conduct a study with the aim of developing, from geotagged Twitter data, a predictive model for the identification of food deserts in the United States, using the linguistic constructs found in food-related tweets. The results from this study suggest the food-ingestion language found in tweets, such as census-tract level measures of food sentiment and healthiness, are associated with census tract-level food desert status. Additionally, the results suggest that including food ingestion language derived from tweets in classification models that predict food desert status improves model performance when compared to baseline models that only include socio-economic characteristics. In Study #2, I evaluate whether attitudes towards COVID-19 vaccines collected from the Household Pulse Survey can be predicted using attitudes extracted from Twitter. The results reveal that attitudes toward COVID-19 vaccines found in tweets explain 61-72% of the variability in the percentage of HPS respondents that were vaccine hesitant or compliant. The results also reveal significant statistical relationships between perceptions expressed on Twitter and in the survey. In Study #3, I conduct a study to examine whether supplementing COVID-19 vaccine uptake forecast models with the attitudes found in tweets improves over baseline models that only use historical vaccination data. The results of this study reveal that supplementing baseline forecast models with both historical vaccination data and COVID-19 vaccine attitudes found in tweets reduce RMSE by as much as 9%. The studies outlined in this dissertation suggest there is a valuable signal for public health research in Twitter data.Item ROBUSTNESS AND UNDERSTANDABILITY OF DEEP MODELS(2022) Ghiasi, Mohammad Amin; Goldstein, Thomas; Computer Science; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Deep learning has made a considerable leap in the past few decades, from promising models for solving various problems to becoming state-of-the-art. However, unlike classical machine learning models, it is sometimes difficult to explain why and how deep learning models make decisions. It is also interesting that their performance can drop with small amounts of noise. In short, deep learning models are well-performing, easily corrupted, hard-to-understand models that beat human beings in many tasks. Consequently, improving these deep models requires a deep understanding. While deep learning models usually generalize well on unseen data, adding negligible amounts of noise to their input can flip their decision. This interesting phenomenon is known as "adversarial attacks." In this thesis, we study several defense methods against such adversarial attacks. More specifically, we focus on defense methods that, unlike traditional methods, use less computation or fewer training examples. We also show that despite the improvements in adversarial defenses, even provable certified defenses can be broken. Moreover, we revisit regularization to improve adversarial robustness. Over the past years, many techniques have been developed for understanding and explaining how deep neural networks make a decision. This thesis introduces a new method for studying the building blocks of neural networks' decisions. First, we introduce the Plug-In Inversion, a new method for inverting and visualizing deep neural network architectures, including Vision Transformers. Then we study the features a ViT learns to make a decision. We compare these features when the network trains on labeled data versus when it uses a language model's supervision for training, such as in CLIP. Last, we introduce feature sonification, which borrows feature visualization techniques to study models trained for speech recognition (non-vision) tasks.Item Tracking the dynamics of the opioid crisis in the United States over space and time(2022) Xia, Zhiyue; Stewart, Kathleen; Geography; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Millions of adolescents and adults in the United States suffer from drug problems such as substance use disorder, referring to clinical impairments including mental illnesses and disabilities caused by drugs. The Substance Abuse and Mental Health Services Administration reported the estimated number of illicit drug users increased to 59.3 million in 2020, or 21.4% of the U.S. population, which made drug misuse one of the most concerning public health issues. Opioids are a category of drugs that can be highly addictive, including heroin and synthetic drugs such as fentanyl. Centers for Disease Control and Prevention (CDC) indicated that about 74.8% of drug overdose deaths involved opioids in 2020. The opioid crisis has hit American cities hard, spreading across the U.S. beginning with the west coast, and then expanding to heavily impact the central, mid-Atlantic, and east coast of the U.S. as well as states in the southeast. In this dissertation, I work on three studies to track the dynamics of the opioid crisis in the U.S. over space and time from a geographic perspective using spatiotemporal data science methods including clustering analysis, time-series models and machine learning approaches. The first study focused on the geospatial patterns of illicit drug-related activities (e.g., possession, delivery, and manufacture of opioids) in a typical U.S. city (Chicago as a case study area). By analyzing more than 52,000 reported drug activities, I built a data-driven machine learning model for predicting opioid hot zones and identifying correlated built environment and sociodemographic factors that drove the opioid crisis in an urban setting. The second study of my dissertation is to analyze the opioid crisis in the context of the global pandemic of SARS-CoV-2 (COVID-19). In 2020, COVID-19 outbroke and affected hundreds of millions of people across the globe. The COVID-19 pandemic is also impacting the community of opioid misusers in the U.S. The major research objective of Study 2 is to understand how the opioid crisis is impacted by the COVID-19 pandemic and to find neighborhood characteristics and economic factors that have driven the variations before and during the pandemic. Study 3 focuses on analyzing the crisis risen by synthetic opioids (including fentanyl) that are more potent and dangerous than other drugs. This study analyzed the geographic patterns of synthetic opioids spreading across the U.S. between 2013 and 2020, a period when synthetic opioids rose to be a major risk factor for public health. The significance of this dissertation is that the three studies investigate the opioid crisis in the U.S. in a comprehensive manner and these studies can facilitate public health stakeholders with effective decision making on healthcare planning relating to drug problems. Tracking the dynamics of the opioid crisis by drug type, including modeling and predicting the geographic patterns of opioid misuse involving particular opioids (e.g, heroin and synthetic opioids), can provide an important basis for applying further treatment services and mitigation efforts, and also be useful for assessing current services and efforts.Item Causal Survival Analysis – Machine Learning Assisted Models: Structural Nested Accelerated Failure Time Model and Threshold Regression(2022) Chen, Yiming; Lee, Mei-Ling ML; Mathematical Statistics; Digital Repository at the University of Maryland; University of Maryland (College Park, Md.)Time-varying confounding for intervention complicates causal survival analysis when the data are collected in a longitudinal manner. Traditional survival models that only adjust for time-dependent covariates provide a biased causal conclusion for the intervention effect. Some techniques have been developed to address this challenge. Nevertheless, these existing methods may still lack power, and suffer from computational burden given high dimensional data with a temporally connected nature. The first part of this dissertation focuses on one of the methods that deal with time-varying confounding, the Structural Nested Model and associated G-estimation. Two Neural Networks (GE-SCORE and GE-MIMIC) were proposed to estimate the Structural Nested Accelerated Failure Time Model. The proposed algorithms can provide less biased and individualized intervention causal effect estimation. The second part explored the causal interpretations and applications of the First-Hitting-Time based Threshold Regression Model using a Wiener process. Moreover, a Neural Network expansion of this specific type of Threshold Regression (TRNN) was explored for the first time.