Matching Items (95)
Filtering by
- Language: English
Description
Image denoising, a fundamental task in computer vision, poses significant challenges due to its inherently inverse and ill-posed nature. Despite advancements in traditional methods and supervised learning approaches, particularly in medical imaging such as Medical Resonance Imaging (MRI) scans, the reliance on paired datasets and known noise distributions remains a practical hurdle. Recent progress in noise statistical independence theory and diffusion models has revitalized research interest, offering promising avenues for unsupervised denoising. However, existing methods often yield overly smoothed results or introduce hallucinated structures, limiting their clinical applicability. This thesis tackles the core challenge of progressing towards unsupervised denoising of MRI scans. It aims to retain intricate details without smoothing or introducing artificial structures, thus ensuring the production of high-quality MRI images. The thesis makes a three-fold contribution: Firstly, it presents a detailed analysis of traditional techniques, early machine learning algorithms for denoising, and new statistical-based models, with an extensive evaluation study on self-supervised denoising methods highlighting their limitations. Secondly, it conducts an evaluation study on an emerging class of diffusion-based denoising methods, accompanied by additional empirical findings and discussions on their effectiveness and limitations, proposing solutions to enhance their utility. Lastly, it introduces a novel approach, Unsupervised Multi-stage Ensemble Deep Learning with diffusion models for denoising MRI scans (MEDL). Leveraging diffusion models, this approach operates independently of signal or noise priors and incorporates weighted rescaling of multi-stage reconstructions to balance over-smoothing and hallucination tendencies. Evaluation using benchmark datasets demonstrates an average gain of 1dB and 2% in PSNR and SSIM metrics, respectively, over existing approaches.
ContributorsVora, Sahil (Author) / Li, Baoxin (Thesis advisor) / Wang, Yalin (Committee member) / Zhou, Yuxiang (Committee member) / Arizona State University (Publisher)
Created2024
Description
Unsupervised learning of time series data, also known as temporal clustering, is a challenging problem in machine learning. This thesis presents a novel algorithm, Deep Temporal Clustering (DTC), to naturally integrate dimensionality reduction and temporal clustering into a single end-to-end learning framework, fully unsupervised. The algorithm utilizes an autoencoder for temporal dimensionality reduction and a novel temporal clustering layer for cluster assignment. Then it jointly optimizes the clustering objective and the dimensionality reduction objective. Based on requirement and application, the temporal clustering layer can be customized with any temporal similarity metric. Several similarity metrics and state-of-the-art algorithms are considered and compared. To gain insight into temporal features that the network has learned for its clustering, a visualization method is applied that generates a region of interest heatmap for the time series. The viability of the algorithm is demonstrated using time series data from diverse domains, ranging from earthquakes to spacecraft sensor data. In each case, the proposed algorithm outperforms traditional methods. The superior performance is attributed to the fully integrated temporal dimensionality reduction and clustering criterion.
ContributorsMadiraju, NaveenSai (Author) / Liang, Jianming (Thesis advisor) / Wang, Yalin (Thesis advisor) / He, Jingrui (Committee member) / Arizona State University (Publisher)
Created2018
Description
A piezoelectric transducer, comprised of electroded and active pad PZT layer atop a backing PZT layer and protected with an acoustic matching layer, and operating under a pulse-echo technique for longitudinal ultrasonic imaging, acts as both source and detector.
Ultrasonic transducer stacks (modules), which had failed or passed during pulse-echo sensitivity testing, were received from Consortium X. With limited background information on these stacks, the central theme was to determine the origin(s) of failure via the use of thermal and physicochemical characterization techniques.
The optical and scanning electron microscopy revealed that contact electrode layers are discontinuous in all samples, while delaminations between electrodes and pad layer were observed in failed samples. The X-ray diffraction data on the pad PZT revealed an overall c/a ratio of 1.022 ratio and morphotropic boundary composition, with significant variations of the Zr to Ti ratio within a sample and between samples. Electron probe microanalysis confirmed that the overall Zr to Ti ratio of the pad PZT was 52/48, and higher amounts of excess PbO in failed samples, whereas, inductively coupled plasma mass spectrometry revealed the presence of Mn, Al, and Sb (dopants) and presence of Cu (sintering aid) in in this hard (pad) PZT. Additionally, three exothermic peaks during thermal analysis was indicative of incomplete calcination of pad PZT. Moreover, transmission electron microscopy and scanning transmission electron microscopy revealed the presence of parylene at the Ag-pad PZT interface and within the pores of pad PZT (in failed samples subjected to electric fields). This further dilutes the electrical, mechanical, and electromechanical properties of the pad PZT, which in turn detrimentally influences the pulse echo sensitivity.
Ultrasonic transducer stacks (modules), which had failed or passed during pulse-echo sensitivity testing, were received from Consortium X. With limited background information on these stacks, the central theme was to determine the origin(s) of failure via the use of thermal and physicochemical characterization techniques.
The optical and scanning electron microscopy revealed that contact electrode layers are discontinuous in all samples, while delaminations between electrodes and pad layer were observed in failed samples. The X-ray diffraction data on the pad PZT revealed an overall c/a ratio of 1.022 ratio and morphotropic boundary composition, with significant variations of the Zr to Ti ratio within a sample and between samples. Electron probe microanalysis confirmed that the overall Zr to Ti ratio of the pad PZT was 52/48, and higher amounts of excess PbO in failed samples, whereas, inductively coupled plasma mass spectrometry revealed the presence of Mn, Al, and Sb (dopants) and presence of Cu (sintering aid) in in this hard (pad) PZT. Additionally, three exothermic peaks during thermal analysis was indicative of incomplete calcination of pad PZT. Moreover, transmission electron microscopy and scanning transmission electron microscopy revealed the presence of parylene at the Ag-pad PZT interface and within the pores of pad PZT (in failed samples subjected to electric fields). This further dilutes the electrical, mechanical, and electromechanical properties of the pad PZT, which in turn detrimentally influences the pulse echo sensitivity.
ContributorsPeri, Prudhvi Ram (Author) / Dey, Sandwip (Thesis advisor) / Smith, David (Committee member) / Alford, Terry (Committee member) / Arizona State University (Publisher)
Created2018
Description
Computer assisted language learning (CALL) has become increasingly common as a means of helping learners develop essential skills in a second or foreign language. However, while many CALL programs claim to be based on principles of second language acquisition (SLA) theory and research, evaluation of design and learning outcomes at the level of individual CALL exercises is lacking in the existing literature. The following proposed study will explore the design of computer-based vocabulary matching exercises using both written text and images and the effects of various design manipulations on learning outcomes. The study will use eye-tracking to investigate what users attend to on screen as they work through a series of exercises with different configurations of written words and images. It will ask whether manipulation of text and image features and combinations can have an effect on learners’ attention to the various elements, and if so, whether differences in levels of attention results in higher or lower scores for measures of learning. Specifically, eye-tracking data will be compared to post-test scores for recall and recognition of target vocabulary items to look for a correlation between levels of attention to written forms in-task and post-test gains in scores for vocabulary learning.
ContributorsPatchin, Colleen (Author) / Smith, David (Thesis advisor) / Ross, Andrew (Committee member) / James, Mark (Committee member) / Arizona State University (Publisher)
Created2019
Description
This dissertation describes fundamental studies of hollow carbon nanostructures, which may be used as electrodes for practical energy storage applications such as batteries or supercapacitors. Electron microscopy is heavily utilized for the nanoscale characterization. To control the morphology of hollow carbon nanostructures, ZnO nanowires serve as sacrificial templates. The first part of this dissertation focuses on the optimization of synthesis parameters and the scale-up production of ZnO nanowires by vapor transport method. Uniform ZnO nanowires with 40 nm width can be produced by using 1100 °C reaction temperature and 20 sccm oxygen flow rate, which are the two most important parameters.
The use of ethanol as carbon source with or without water steam provides uniform carbonaceous deposition on ZnO nanowire templates. The amount of as-deposited carbonaceous material can be controlled by reaction temperature and reaction time. Due to the catalytic property of ZnO surface, the thicknesses of carbonaceous layers are typically in nanometers. Different methods to remove the ZnO templates are explored, of which hydrogen reduction at temperatures higher than 700 °C is most efficient. The ZnO templates can also be removed under ethanol environment, but the temperatures need to be higher than 850 °C for practical use.
Characterizations of hollow carbon nanofibers show that the hollow carbon nanostructures have a high specific surface area (>1100 m2/g) with the presence of mesopores (~3.5 nm). The initial data on energy storage as electrodes of electrochemical double layer capacitors show that high specific capacitance (> 220 F/g) can be obtained, which is related to the high surface area and unique porous hollow structure with a thin wall.
The use of ethanol as carbon source with or without water steam provides uniform carbonaceous deposition on ZnO nanowire templates. The amount of as-deposited carbonaceous material can be controlled by reaction temperature and reaction time. Due to the catalytic property of ZnO surface, the thicknesses of carbonaceous layers are typically in nanometers. Different methods to remove the ZnO templates are explored, of which hydrogen reduction at temperatures higher than 700 °C is most efficient. The ZnO templates can also be removed under ethanol environment, but the temperatures need to be higher than 850 °C for practical use.
Characterizations of hollow carbon nanofibers show that the hollow carbon nanostructures have a high specific surface area (>1100 m2/g) with the presence of mesopores (~3.5 nm). The initial data on energy storage as electrodes of electrochemical double layer capacitors show that high specific capacitance (> 220 F/g) can be obtained, which is related to the high surface area and unique porous hollow structure with a thin wall.
ContributorsSong, Yian (Author) / Liu, Jingyue (Committee member) / Smith, David (Committee member) / McCartney, Martha (Committee member) / Chen, Tingyong (Committee member) / Arizona State University (Publisher)
Created2016
Description
This dissertation presents research findings regarding the exploitation of localized surface plasmon (LSP) of epitaxial Ag islands as a means to enhance the photoluminescence (PL) of Germanium (Ge) quantum dots (QDs). The first step of this project was to investigate the growth of Ag islands on Si(100). Two distinct families of Ag islands have been observed. “Big islands” are clearly faceted and have basal dimensions in the few hundred nm to μm range with a variety of basal shapes. “Small islands” are not clearly faceted and have basal diameters in the 10s of nm range. Big islands form via a nucleation and growth mechanism, and small islands form via precipitation of Ag contained in a planar layer between the big islands that is thicker than the Stranski-Krastanov layer existing at room-temperature.
The pseudodielectric functions of epitaxial Ag islands on Si(100) substrates were investigated with spectroscopic ellipsometry. Comparing the experimental pseudodielectric functions obtained for Si with and without Ag islands clearly identifies a plasmon mode with its dipole moment perpendicular to the surface. This observation is confirmed using a simulation based on the thin island film (TIF) theory. Another mode parallel to the surface may be identified by comparing the experimental pseudodielectric functions with the simulated ones from TIF theory. Additional results suggest that the LSP energy of Ag islands can be tuned from the ultra-violet to the infrared range by an amorphous Si (α-Si) cap layer.
Heterostructures were grown that incorporated Ge QDs, an epitaxial Si cap layer and Ag islands grown atop the Si cap layer. Optimum growth conditions for distinct Ge dot ensembles and Si cap layers were obtained. The density of Ag islands grown on the Si cap layer depends on its thickness. Factors contributing to this effect may include the average strain and Ge concentration on the surface of the Si cap layer.
The effects of the Ag LSP on the PL of Ge coherent domes were investigated for both α-Si capped and bare Ag islands. For samples with low-doped substrates, the LSPs reduce the Ge dot-related PL when the Si cap layer is below some critical thickness and have no effect on the PL when the Si cap layer is above the critical thickness. For samples grown on highly-doped wafers, the LSP of bare Ag islands enhanced the PL of Ge QDs by ~ 40%.
The pseudodielectric functions of epitaxial Ag islands on Si(100) substrates were investigated with spectroscopic ellipsometry. Comparing the experimental pseudodielectric functions obtained for Si with and without Ag islands clearly identifies a plasmon mode with its dipole moment perpendicular to the surface. This observation is confirmed using a simulation based on the thin island film (TIF) theory. Another mode parallel to the surface may be identified by comparing the experimental pseudodielectric functions with the simulated ones from TIF theory. Additional results suggest that the LSP energy of Ag islands can be tuned from the ultra-violet to the infrared range by an amorphous Si (α-Si) cap layer.
Heterostructures were grown that incorporated Ge QDs, an epitaxial Si cap layer and Ag islands grown atop the Si cap layer. Optimum growth conditions for distinct Ge dot ensembles and Si cap layers were obtained. The density of Ag islands grown on the Si cap layer depends on its thickness. Factors contributing to this effect may include the average strain and Ge concentration on the surface of the Si cap layer.
The effects of the Ag LSP on the PL of Ge coherent domes were investigated for both α-Si capped and bare Ag islands. For samples with low-doped substrates, the LSPs reduce the Ge dot-related PL when the Si cap layer is below some critical thickness and have no effect on the PL when the Si cap layer is above the critical thickness. For samples grown on highly-doped wafers, the LSP of bare Ag islands enhanced the PL of Ge QDs by ~ 40%.
ContributorsKong, Dexin (Author) / Drucker, Jeffery (Thesis advisor) / Chen, Tingyong (Committee member) / Ros, Robert (Committee member) / Smith, David (Committee member) / Arizona State University (Publisher)
Created2015
Description
Understanding the complexity of temporal and spatial characteristics of gene expression over brain development is one of the crucial research topics in neuroscience. An accurate description of the locations and expression status of relative genes requires extensive experiment resources. The Allen Developing Mouse Brain Atlas provides a large number of in situ hybridization (ISH) images of gene expression over seven different mouse brain developmental stages. Studying mouse brain models helps us understand the gene expressions in human brains. This atlas collects about thousands of genes and now they are manually annotated by biologists. Due to the high labor cost of manual annotation, investigating an efficient approach to perform automated gene expression annotation on mouse brain images becomes necessary. In this thesis, a novel efficient approach based on machine learning framework is proposed. Features are extracted from raw brain images, and both binary classification and multi-class classification models are built with some supervised learning methods. To generate features, one of the most adopted methods in current research effort is to apply the bag-of-words (BoW) algorithm. However, both the efficiency and the accuracy of BoW are not outstanding when dealing with large-scale data. Thus, an augmented sparse coding method, which is called Stochastic Coordinate Coding, is adopted to generate high-level features in this thesis. In addition, a new multi-label classification model is proposed in this thesis. Label hierarchy is built based on the given brain ontology structure. Experiments have been conducted on the atlas and the results show that this approach is efficient and classifies the images with a relatively higher accuracy.
ContributorsZhao, Xinlin (Author) / Ye, Jieping (Thesis advisor) / Wang, Yalin (Thesis advisor) / Li, Baoxin (Committee member) / Arizona State University (Publisher)
Created2016
Description
Photocatalytic water splitting is a promising technique to produce H2 fuels from water using sustainable solar energy. To better design photocatalysts, the understanding of charge transfer at surfaces/interfaces and the corresponding structure change during the reaction is very important. Local structural and chemical information on nanoparticle surfaces or interfaces can be achieved through characterizations on transmission electron microscopy (TEM). Emphasis should be put on materials structure changes during the reactions in their “working conditions”. Environmental TEM with in situ light illumination system allows the photocatalysts to be studied under light irradiation when exposed to H2O vapor. A set of ex situ and in situ TEM characterizations are carried out on typical types of TiO2 based photocatalysts. The observed structure changes during the reaction are correlated with the H2 production rate for structure-property relationships.
A surface disordering was observed in situ when well-defined anatase TiO2 rhombohedral nanoparticles were exposed to 1 Torr H2O vapor and 10suns light inside the environmental TEM. The disordering is believed to be related to high density of hydroxyl groups formed on surface oxygen vacancies during water splitting reactions.
Pt co-catalyst on TiO2 is able to split pure water producing H2 and O2. The H2 production rate drops during the reaction. Particle size growth during reaction was discovered with Z-contrast images. The particle size growth is believed to be a photo-electro-chemical Ostwald ripening.
Characterizations were also carried out on a more complicated photocatalyst system: Ni/NiO core/shell co-catalyst on TiO2. A decrease of the H2 production rate resulting from photo-corrosion was observed. The Ni is believed to be oxidized to Ni2+ by OH• radicals which are intermediate products of H2O oxidation. The mechanism that the OH• radicals leak into the cores through cracks on NiO shells is more supported by experiments.
Overall this research has done a comprehensive ex situ and in situ TEM characterizations following some typical TiO2 based photocatalysts during reactions. This research has shown the technique availability to study photocatalyst inside TEM in photocatalytic conditions. It also demonstrates the importance to follow structure changes of materials during reactions in understanding deactivation mechanisms.
A surface disordering was observed in situ when well-defined anatase TiO2 rhombohedral nanoparticles were exposed to 1 Torr H2O vapor and 10suns light inside the environmental TEM. The disordering is believed to be related to high density of hydroxyl groups formed on surface oxygen vacancies during water splitting reactions.
Pt co-catalyst on TiO2 is able to split pure water producing H2 and O2. The H2 production rate drops during the reaction. Particle size growth during reaction was discovered with Z-contrast images. The particle size growth is believed to be a photo-electro-chemical Ostwald ripening.
Characterizations were also carried out on a more complicated photocatalyst system: Ni/NiO core/shell co-catalyst on TiO2. A decrease of the H2 production rate resulting from photo-corrosion was observed. The Ni is believed to be oxidized to Ni2+ by OH• radicals which are intermediate products of H2O oxidation. The mechanism that the OH• radicals leak into the cores through cracks on NiO shells is more supported by experiments.
Overall this research has done a comprehensive ex situ and in situ TEM characterizations following some typical TiO2 based photocatalysts during reactions. This research has shown the technique availability to study photocatalyst inside TEM in photocatalytic conditions. It also demonstrates the importance to follow structure changes of materials during reactions in understanding deactivation mechanisms.
ContributorsZhang, Liuxian (Author) / Crozier, Peter (Thesis advisor) / Smith, David (Committee member) / Chan, Candace (Committee member) / Liu, Jingyue (Committee member) / Arizona State University (Publisher)
Created2015
Description
The rapid growth of social media in recent years provides a large amount of user-generated visual objects, e.g., images and videos. Advanced semantic understanding approaches on such visual objects are desired to better serve applications such as human-machine interaction, image retrieval, etc. Semantic visual attributes have been proposed and utilized in multiple visual computing tasks to bridge the so-called "semantic gap" between extractable low-level feature representations and high-level semantic understanding of the visual objects.
Despite years of research, there are still some unsolved problems on semantic attribute learning. First, real-world applications usually involve hundreds of attributes which requires great effort to acquire sufficient amount of labeled data for model learning. Second, existing attribute learning work for visual objects focuses primarily on images, with semantic analysis on videos left largely unexplored.
In this dissertation I conduct innovative research and propose novel approaches to tackling the aforementioned problems. In particular, I propose robust and accurate learning frameworks on both attribute ranking and prediction by exploring the correlation among multiple attributes and utilizing various types of label information. Furthermore, I propose a video-based skill coaching framework by extending attribute learning to the video domain for robust motion skill analysis. Experiments on various types of applications and datasets and comparisons with multiple state-of-the-art baseline approaches confirm that my proposed approaches can achieve significant performance improvements for the general attribute learning problem.
Despite years of research, there are still some unsolved problems on semantic attribute learning. First, real-world applications usually involve hundreds of attributes which requires great effort to acquire sufficient amount of labeled data for model learning. Second, existing attribute learning work for visual objects focuses primarily on images, with semantic analysis on videos left largely unexplored.
In this dissertation I conduct innovative research and propose novel approaches to tackling the aforementioned problems. In particular, I propose robust and accurate learning frameworks on both attribute ranking and prediction by exploring the correlation among multiple attributes and utilizing various types of label information. Furthermore, I propose a video-based skill coaching framework by extending attribute learning to the video domain for robust motion skill analysis. Experiments on various types of applications and datasets and comparisons with multiple state-of-the-art baseline approaches confirm that my proposed approaches can achieve significant performance improvements for the general attribute learning problem.
ContributorsChen, Lin (Author) / Li, Baoxin (Thesis advisor) / Turaga, Pavan (Committee member) / Wang, Yalin (Committee member) / Liu, Huan (Committee member) / Arizona State University (Publisher)
Created2016
Description
In many fields one needs to build predictive models for a set of related machine learning tasks, such as information retrieval, computer vision and biomedical informatics. Traditionally these tasks are treated independently and the inference is done separately for each task, which ignores important connections among the tasks. Multi-task learning aims at simultaneously building models for all tasks in order to improve the generalization performance, leveraging inherent relatedness of these tasks. In this thesis, I firstly propose a clustered multi-task learning (CMTL) formulation, which simultaneously learns task models and performs task clustering. I provide theoretical analysis to establish the equivalence between the CMTL formulation and the alternating structure optimization, which learns a shared low-dimensional hypothesis space for different tasks. Then I present two real-world biomedical informatics applications which can benefit from multi-task learning. In the first application, I study the disease progression problem and present multi-task learning formulations for disease progression. In the formulations, the prediction at each point is a regression task and multiple tasks at different time points are learned simultaneously, leveraging the temporal smoothness among the tasks. The proposed formulations have been tested extensively on predicting the progression of the Alzheimer's disease, and experimental results demonstrate the effectiveness of the proposed models. In the second application, I present a novel data-driven framework for densifying the electronic medical records (EMR) to overcome the sparsity problem in predictive modeling using EMR. The densification of each patient is a learning task, and the proposed algorithm simultaneously densify all patients. As such, the densification of one patient leverages useful information from other patients.
ContributorsZhou, Jiayu (Author) / Ye, Jieping (Thesis advisor) / Mittelmann, Hans (Committee member) / Li, Baoxin (Committee member) / Wang, Yalin (Committee member) / Arizona State University (Publisher)
Created2014