Matching Items (120)
Filtering by
- Genre: Academic theses
- Creators: Yang, Yezhou
Description
Thiol functionalization is one potentially useful way to tailor physical and chemical properties of graphene oxides (GOs) and reduced graphene oxides (RGOs). Despite the ubiquitous presence of thiol functional groups in diverse chemical systems, efficient thiol functionalization has been challenging for GOs and RGOs, or for carbonaceous materials in general. In this work, thionation of GOs has been achieved in high yield through two new methods that also allow concomitant chemical reduction/thermal reduction of GOs; a solid-gas metathetical reaction method with boron sulfides (BxSy) gases and a solvothermal reaction method employing phosphorus decasulfide (P4S10). The thionation products, called "mercapto reduced graphene oxides (m-RGOs)", were characterized by employing X-ray photoelectron spectroscopy, powder X-ray diffraction, UV-Vis spectroscopy, FT-IR spectroscopy, Raman spectroscopy, electron probe analysis, scanning electron microscopy, (scanning) transmission electron microscopy, nano secondary ion mass spectrometry, Ellman assay and atomic force microscopy. The excellent dispersibility of m-RGOs in various solvents including alcohols has allowed fabrication of thin films of m-RGOs. Deposition of m-RGOs on gold substrates was achieved through solution deposition and the m-RGOs were homogeneously distributed on gold surface shown by atomic force microscopy. Langmuir-Blodgett (LB) films of m-RGOs were obtained by transferring their Langmuir films, formed by simple drop casting of m-RGOs dispersion on water surface, onto various substrates including gold, glass and indium tin oxide. The m-RGO LB films showed low sheet resistances down to about 500 kΩ/sq at 92% optical transparency. The successful results make m-RGOs promising for applications in transparent conductive coatings, biosensing, etc.
ContributorsJeon, Kiwan (Author) / Seo, Dong-Kyun (Thesis advisor) / Jones, Anne K (Committee member) / Yarger, Jeffery (Committee member) / Arizona State University (Publisher)
Created2013
Description
Geopolymers, a class of X-ray amorphous, ceramic-like aluminosilicate materials are produced at ambient temperatures through a process called geopolymerization. Due to both low energy requirement during synthesis and interesting mechanical and chemical properties, geopolymers are grabbing enormous attention. Although geopolymers have a broad range of applications including thermal/acoustic insulation and waste immobilization, they are always prepared in monolithic form. The primary aim of this study is to produce new nanostructured materials from the geopolymerization process, including porous monoliths and powders.
In view of the current interest in porous geopolymers for non-traditional applications, it is becoming increasingly important to develop synthetic techniques to introduce interconnected pores into the geopolymers. This study presents a simple synthetic route to produce hierarchically porous geopolymers via a reactive emulsion templating process utilizing triglyceride oil. In this new method, highly alkaline geopolymer resin is mixed with canola oil to form a homogeneous viscous emulsion which, when cured at 60 °C, gives a hard monolithic material. During the process, the oil in the alkaline emulsion undergoes a saponification reaction to decompose into water-soluble soap and glycerol molecules which are extracted to yield porous geopolymers. Nitrogen sorption studies indicates the presence of mesopores, whereas the SEM studies reveals that the mesoporous geopolymer matrix is dotted with spherical macropores. The method exhibits flexibility in that the pore structure of the final porous geopolymers products can be adjusted by varying the precursor composition.
In a second method, the geopolymerization process is modified to produce highly dispersible geopolymer particles, by activating metakaolin with sodium silicate solutions containing excess alkali, and curing for short duration under moist conditions. The produced geopolymer particles exhibit morphology similar to carbon blacks and structured silicas, while also being stable over a wide pH range.
Finally, highly crystalline hierarchical faujasite zeolites are prepared by yet another modification of the geopolymerization process. In this technique, the second method is combined with a saponification reaction of triglyceride oil. The resulting hierarchical zeolites exhibit superior CO2-sorption properties compared to equivalent commercially available and currently reported materials. Additionally, the simplicity of all three of these techniques means they are readily scalable.
In view of the current interest in porous geopolymers for non-traditional applications, it is becoming increasingly important to develop synthetic techniques to introduce interconnected pores into the geopolymers. This study presents a simple synthetic route to produce hierarchically porous geopolymers via a reactive emulsion templating process utilizing triglyceride oil. In this new method, highly alkaline geopolymer resin is mixed with canola oil to form a homogeneous viscous emulsion which, when cured at 60 °C, gives a hard monolithic material. During the process, the oil in the alkaline emulsion undergoes a saponification reaction to decompose into water-soluble soap and glycerol molecules which are extracted to yield porous geopolymers. Nitrogen sorption studies indicates the presence of mesopores, whereas the SEM studies reveals that the mesoporous geopolymer matrix is dotted with spherical macropores. The method exhibits flexibility in that the pore structure of the final porous geopolymers products can be adjusted by varying the precursor composition.
In a second method, the geopolymerization process is modified to produce highly dispersible geopolymer particles, by activating metakaolin with sodium silicate solutions containing excess alkali, and curing for short duration under moist conditions. The produced geopolymer particles exhibit morphology similar to carbon blacks and structured silicas, while also being stable over a wide pH range.
Finally, highly crystalline hierarchical faujasite zeolites are prepared by yet another modification of the geopolymerization process. In this technique, the second method is combined with a saponification reaction of triglyceride oil. The resulting hierarchical zeolites exhibit superior CO2-sorption properties compared to equivalent commercially available and currently reported materials. Additionally, the simplicity of all three of these techniques means they are readily scalable.
ContributorsMedpelli, Dinesh (Author) / Seo, Dong-Kyun (Thesis advisor) / Herckes, Pierre (Committee member) / Petuskey, William (Committee member) / Arizona State University (Publisher)
Created2015
Description
This thesis focused on physicochemical and electrochemical projects directed towards two electrolyte types: 1) class of ionic liquids serving as electrolytes in the catholyte for alkali-metal ion conduction in batteries and 2) gel membrane for proton conduction in fuel cells; where overall aims were encouraged by the U.S. Department of Energy.
Large-scale, sodium-ion batteries are seen as global solutions to providing undisrupted electricity from sustainable, but power-fluctuating, energy production in the near future. Foreseen ideal advantages are lower cost without sacrifice of desired high-energy densities relative to present lithium-ion and lead-acid battery systems. Na/NiCl2 (ZEBRA) and Na/S battery chemistries, suffer from high operation temperature (>300ºC) and safety concerns following major fires consequent of fuel mixing after cell-separator rupturing. Initial interest was utilizing low-melting organic ionic liquid, [EMI+][AlCl4-], with well-known molten salt, NaAlCl4, to create a low-to-moderate operating temperature version of ZEBRA batteries; which have been subject of prior sodium battery research spanning decades. Isothermal conductivities of these electrolytes revealed a fundamental kinetic problem arisen from "alkali cation-trapping effect" yet relived by heat-ramping >140ºC.
Battery testing based on [EMI+][FeCl4-] with NaAlCl4 functioned exceptional (range 150-180ºC) at an impressive energy efficiency >96%. Newly prepared inorganic ionic liquid, [PBr4+][Al2Br7-]:NaAl2Br7, melted at 94ºC. NaAl2Br7 exhibited super-ionic conductivity 10-1.75 Scm-1 at 62ºC ensued by solid-state rotator phase transition. Also improved thermal stability when tested to 265ºC and less expensive chemical synthesis. [PBr4+][Al2Br7-] demonstrated remarkable, ionic decoupling in the liquid-state due to incomplete bromide-ion transfer depicted in NMR measurements.
Fuel cells are electrochemical devices generating electrical energy reacting hydrogen/oxygen gases producing water vapor. Principle advantage is high-energy efficiency of up to 70% in contrast to an internal combustion engine <40%. Nafion-based fuel cells are prone to carbon monoxide catalytic poisoning and polymer membrane degradation unless heavily hydrated under cell-pressurization. This novel "SiPOH" solid-electrolytic gel (originally liquid-state) operated in the fuel cell at 121oC yielding current and power densities high as 731mAcm-2 and 345mWcm-2, respectively. Enhanced proton conduction significantly increased H2 fuel efficiency to 89.7% utilizing only 3.1mlmin-1 under dry, unpressurized testing conditions. All these energy devices aforementioned evidently have future promise; therefore in early developmental stages.
Large-scale, sodium-ion batteries are seen as global solutions to providing undisrupted electricity from sustainable, but power-fluctuating, energy production in the near future. Foreseen ideal advantages are lower cost without sacrifice of desired high-energy densities relative to present lithium-ion and lead-acid battery systems. Na/NiCl2 (ZEBRA) and Na/S battery chemistries, suffer from high operation temperature (>300ºC) and safety concerns following major fires consequent of fuel mixing after cell-separator rupturing. Initial interest was utilizing low-melting organic ionic liquid, [EMI+][AlCl4-], with well-known molten salt, NaAlCl4, to create a low-to-moderate operating temperature version of ZEBRA batteries; which have been subject of prior sodium battery research spanning decades. Isothermal conductivities of these electrolytes revealed a fundamental kinetic problem arisen from "alkali cation-trapping effect" yet relived by heat-ramping >140ºC.
Battery testing based on [EMI+][FeCl4-] with NaAlCl4 functioned exceptional (range 150-180ºC) at an impressive energy efficiency >96%. Newly prepared inorganic ionic liquid, [PBr4+][Al2Br7-]:NaAl2Br7, melted at 94ºC. NaAl2Br7 exhibited super-ionic conductivity 10-1.75 Scm-1 at 62ºC ensued by solid-state rotator phase transition. Also improved thermal stability when tested to 265ºC and less expensive chemical synthesis. [PBr4+][Al2Br7-] demonstrated remarkable, ionic decoupling in the liquid-state due to incomplete bromide-ion transfer depicted in NMR measurements.
Fuel cells are electrochemical devices generating electrical energy reacting hydrogen/oxygen gases producing water vapor. Principle advantage is high-energy efficiency of up to 70% in contrast to an internal combustion engine <40%. Nafion-based fuel cells are prone to carbon monoxide catalytic poisoning and polymer membrane degradation unless heavily hydrated under cell-pressurization. This novel "SiPOH" solid-electrolytic gel (originally liquid-state) operated in the fuel cell at 121oC yielding current and power densities high as 731mAcm-2 and 345mWcm-2, respectively. Enhanced proton conduction significantly increased H2 fuel efficiency to 89.7% utilizing only 3.1mlmin-1 under dry, unpressurized testing conditions. All these energy devices aforementioned evidently have future promise; therefore in early developmental stages.
ContributorsTucker, Telpriore G (Author) / Angell, Charles A. (Committee member) / Moore, Ana (Committee member) / Seo, Dong-Kyun (Committee member) / Arizona State University (Publisher)
Created2014
Description
Increasing concentrations of carbon dioxide in the atmosphere will inevitably lead to long-term changes in climate that can have serious consequences. Controlling anthropogenic emission of carbon dioxide into the atmosphere, however, represents a significant technological challenge. Various chemical approaches have been suggested, perhaps the most promising of these is based on electrochemical trapping of carbon dioxide using pyridine and derivatives. Optimization of this process requires a detailed understanding of the mechanisms of the reactions of reduced pyridines with carbon dioxide, which are not currently well known. This thesis describes a detailed mechanistic study of the nucleophilic and Bronsted basic properties of the radical anion of bipyridine as a model pyridine derivative, formed by one-electron reduction, with particular emphasis on the reactions with carbon dioxide. A time-resolved spectroscopic method was used to characterize the key intermediates and determine the kinetics of the reactions of the radical anion and its protonated radical form. Using a pulsed nanosecond laser, the bipyridine radical anion could be generated in-situ in less than 100 ns, which allows fast reactions to be monitored in real time. The bipyridine radical anion was found to be a very powerful one-electron donor, Bronsted base and nucleophile. It reacts by addition to the C=O bonds of ketones with a bimolecular rate constant around 1* 107 M-1 s-1. These are among the fastest nucleophilic additions that have been reported in literature. Temperature dependence studies demonstrate very low activation energies and large Arrhenius pre-exponential parameters, consistent with very high reactivity. The kinetics of E2 elimination, where the radical anion acts as a base, and SN2 substitution, where the radical anion acts as a nucleophile, are also characterized by large bimolecular rate constants in the range ca. 106 - 107 M-1 s-1. The pKa of the bipyridine radical anion was measured using a kinetic method and analysis of the data using a Marcus theory model for proton transfer. The bipyridine radical anion is found to have a pKa of 40±5 in DMSO. The reorganization energy for the proton transfer reaction was found to be 70±5 kJ/mol. The bipyridine radical anion was found to react very rapidly with carbon dioxide, with a bimolecular rate constant of 1* 108 M-1 s-1 and a small activation energy, whereas the protonated radical reacted with carbon dioxide with a rate constant that was too small to measure. The kinetic and thermodynamic data obtained in this work can be used to understand the mechanisms of the reactions of pyridines with carbon dioxide under reducing conditions.
ContributorsRanjan, Rajeev (Author) / Gould, Ian R (Thesis advisor) / Buttry, Daniel A (Thesis advisor) / Yarger, Jeff (Committee member) / Seo, Dong-Kyun (Committee member) / Arizona State University (Publisher)
Created2015
Description
Deoxyribonucleic acid (DNA) has been treated as excellent building material for nanoscale construction because of its unique structural features. Its ability to self-assemble into predictable and addressable nanostructures distinguishes it from other materials. A large variety of DNA nanostructures have been constructed, providing scaffolds with nanometer precision to organize functional molecules. This dissertation focuses on developing biologically replicating DNA nanostructures to explore their biocompatibility for potential functions in cells, as well as studying the molecular behaviors of DNA origami tiles in higher-order self-assembly for constructing DNA nanostructures with large size and complexity. Presented here are a series of studies towards this goal. First, a single-stranded DNA tetrahedron was constructed and replicated in vivo with high efficiency and fidelity. This study indicated the compatibility between DNA nanostructures and biological systems, and suggested a feasible low-coast method to scale up the preparation of synthetic DNA. Next, the higher-order self-assembly of DNA origami tiles was systematically studied. It was demonstrated that the dimensional aspect ratio of origami tiles as well as the intertile connection design were essential in determining the assembled superstructures. Finally, the effects of DNA hairpin loops on the conformations of origami tiles as well as the higher-order assembled structures were demonstrated. The results would benefit the design and construction of large complex nanostructures.
ContributorsLi, Zhe (Author) / Yan, Hao (Thesis advisor) / Liu, Yan (Thesis advisor) / Seo, Dong-Kyun (Committee member) / Wachter, Rebekka (Committee member) / Arizona State University (Publisher)
Created2012
Description
Nanoporous electrically conducting materials can be prepared with high specific pore volumes and surface areas which make them well-suited for a wide variety of technologies including separation, catalysis and owing to their conductivity, energy related applications like solar cells, batteries and capacitors. General synthetic methods for nanoporous conducting materials that exhibit fine property control as well as facility and efficiency in their implementation continue to be highly sought after. Here, general methods for the synthesis of nanoporous conducting materials and their characterization are presented. Antimony-doped tin oxide (ATO), a transparent conducting oxide (TCO), and nanoporous conducting carbon can be prepared through the step-wise synthesis of interpenetrating inorganic/organic networks using well-established sol-gel methodology. The one-pot method produces an inorganic gel first that encompasses a solution of organic precursors. The surface of the inorganic gel subsequently catalyzes the formation of an organic gel network that interpenetrates throughout the inorganic gel network. These mutually supporting gel networks strengthen one another and allow for the use of evaporative drying methods and heat treatments that would usually destroy the porosity of an unsupported gel network. The composite gel is then selectively treated to either remove the organic network to provide a porous inorganic network, as is the case for antimony-doped tin oxide, or the inorganic network can be removed to generate a porous carbon material. The method exhibits flexibility in that the pore structure of the final porous material can be modified through the variation of the synthetic conditions. Additionally, porous carbons of hierarchical pore size distributions can be prepared by using wet alumina gel as a template dispersion medium and as a template itself. Alumina gels exhibit thixotropy, which is the ability of a solid to be sheared to a liquid state and upon removal of the shear force, return to a solid gel state. Alumina gels were prepared and blended with monomer solutions and sacrificial template particles to produce wet gel composites. These composites could then be treated to remove the alumina and template particles to generate hierarchically porous carbon.
ContributorsVolosin, Alex (Author) / Seo, Dong-Kyun (Thesis advisor) / Buttry, Daniel (Committee member) / Gust, John D (Committee member) / Arizona State University (Publisher)
Created2012
Description
In a collaborative environment where multiple robots and human beings are expected
to collaborate to perform a task, it becomes essential for a robot to be aware of multiple
agents working in its work environment. A robot must also learn to adapt to
different agents in the workspace and conduct its interaction based on the presence
of these agents. A theoretical framework was introduced which performs interaction
learning from demonstrations in a two-agent work environment, and it is called
Interaction Primitives.
This document is an in-depth description of the new state of the art Python
Framework for Interaction Primitives between two agents in a single as well as multiple
task work environment and extension of the original framework in a work environment
with multiple agents doing a single task. The original theory of Interaction
Primitives has been extended to create a framework which will capture correlation
between more than two agents while performing a single task. The new state of the
art Python framework is an intuitive, generic, easy to install and easy to use python
library which can be applied to use the Interaction Primitives framework in a work
environment. This library was tested in simulated environments and controlled laboratory
environment. The results and benchmarks of this library are available in the
related sections of this document.
to collaborate to perform a task, it becomes essential for a robot to be aware of multiple
agents working in its work environment. A robot must also learn to adapt to
different agents in the workspace and conduct its interaction based on the presence
of these agents. A theoretical framework was introduced which performs interaction
learning from demonstrations in a two-agent work environment, and it is called
Interaction Primitives.
This document is an in-depth description of the new state of the art Python
Framework for Interaction Primitives between two agents in a single as well as multiple
task work environment and extension of the original framework in a work environment
with multiple agents doing a single task. The original theory of Interaction
Primitives has been extended to create a framework which will capture correlation
between more than two agents while performing a single task. The new state of the
art Python framework is an intuitive, generic, easy to install and easy to use python
library which can be applied to use the Interaction Primitives framework in a work
environment. This library was tested in simulated environments and controlled laboratory
environment. The results and benchmarks of this library are available in the
related sections of this document.
ContributorsKumar, Ashish, M.S (Author) / Amor, Hani Ben (Thesis advisor) / Zhang, Yu (Committee member) / Yang, Yezhou (Committee member) / Arizona State University (Publisher)
Created2017
Description
Computer Vision as a eld has gone through signicant changes in the last decade.
The eld has seen tremendous success in designing learning systems with hand-crafted
features and in using representation learning to extract better features. In this dissertation
some novel approaches to representation learning and task learning are studied.
Multiple-instance learning which is generalization of supervised learning, is one
example of task learning that is discussed. In particular, a novel non-parametric k-
NN-based multiple-instance learning is proposed, which is shown to outperform other
existing approaches. This solution is applied to a diabetic retinopathy pathology
detection problem eectively.
In cases of representation learning, generality of neural features are investigated
rst. This investigation leads to some critical understanding and results in feature
generality among datasets. The possibility of learning from a mentor network instead
of from labels is then investigated. Distillation of dark knowledge is used to eciently
mentor a small network from a pre-trained large mentor network. These studies help
in understanding representation learning with smaller and compressed networks.
The eld has seen tremendous success in designing learning systems with hand-crafted
features and in using representation learning to extract better features. In this dissertation
some novel approaches to representation learning and task learning are studied.
Multiple-instance learning which is generalization of supervised learning, is one
example of task learning that is discussed. In particular, a novel non-parametric k-
NN-based multiple-instance learning is proposed, which is shown to outperform other
existing approaches. This solution is applied to a diabetic retinopathy pathology
detection problem eectively.
In cases of representation learning, generality of neural features are investigated
rst. This investigation leads to some critical understanding and results in feature
generality among datasets. The possibility of learning from a mentor network instead
of from labels is then investigated. Distillation of dark knowledge is used to eciently
mentor a small network from a pre-trained large mentor network. These studies help
in understanding representation learning with smaller and compressed networks.
ContributorsVenkatesan, Ragav (Author) / Li, Baoxin (Thesis advisor) / Turaga, Pavan (Committee member) / Yang, Yezhou (Committee member) / Davulcu, Hasan (Committee member) / Arizona State University (Publisher)
Created2017
Description
With the rise of the Big Data Era, an exponential amount of network data is being generated at an unprecedented rate across a wide-range of high impact micro and macro areas of research---from protein interaction to social networks. The critical challenge is translating this large scale network data into actionable information.
A key task in the data translation is the analysis of network connectivity via marked nodes---the primary focus of our research. We have developed a framework for analyzing network connectivity via marked nodes in large scale graphs, utilizing novel algorithms in three interrelated areas: (1) analysis of a single seed node via it’s ego-centric network (AttriPart algorithm); (2) pathway identification between two seed nodes (K-Simple Shortest Paths Multithreaded and Search Reduced (KSSPR) algorithm); and (3) tree detection, defining the interaction between three or more seed nodes (Shortest Path MST algorithm).
In an effort to address both fundamental and applied research issues, we have developed the LocalForcasting algorithm to explore how network connectivity analysis can be applied to local community evolution and recommender systems. The goal is to apply the LocalForecasting algorithm to various domains---e.g., friend suggestions in social networks or future collaboration in co-authorship networks. This algorithm utilizes link prediction in combination with the AttriPart algorithm to predict future connections in local graph partitions.
Results show that our proposed AttriPart algorithm finds up to 1.6x denser local partitions, while running approximately 43x faster than traditional local partitioning techniques (PageRank-Nibble). In addition, our LocalForecasting algorithm demonstrates a significant improvement in the number of nodes and edges correctly predicted over baseline methods. Furthermore, results for the KSSPR algorithm demonstrate a speed-up of up to 2.5x the standard k-simple shortest paths algorithm.
A key task in the data translation is the analysis of network connectivity via marked nodes---the primary focus of our research. We have developed a framework for analyzing network connectivity via marked nodes in large scale graphs, utilizing novel algorithms in three interrelated areas: (1) analysis of a single seed node via it’s ego-centric network (AttriPart algorithm); (2) pathway identification between two seed nodes (K-Simple Shortest Paths Multithreaded and Search Reduced (KSSPR) algorithm); and (3) tree detection, defining the interaction between three or more seed nodes (Shortest Path MST algorithm).
In an effort to address both fundamental and applied research issues, we have developed the LocalForcasting algorithm to explore how network connectivity analysis can be applied to local community evolution and recommender systems. The goal is to apply the LocalForecasting algorithm to various domains---e.g., friend suggestions in social networks or future collaboration in co-authorship networks. This algorithm utilizes link prediction in combination with the AttriPart algorithm to predict future connections in local graph partitions.
Results show that our proposed AttriPart algorithm finds up to 1.6x denser local partitions, while running approximately 43x faster than traditional local partitioning techniques (PageRank-Nibble). In addition, our LocalForecasting algorithm demonstrates a significant improvement in the number of nodes and edges correctly predicted over baseline methods. Furthermore, results for the KSSPR algorithm demonstrate a speed-up of up to 2.5x the standard k-simple shortest paths algorithm.
ContributorsFreitas, Scott (Author) / Tong, Hanghang (Thesis advisor) / Maciejewski, Ross (Committee member) / Yang, Yezhou (Committee member) / Arizona State University (Publisher)
Created2018
Description
The performance of most of the visual computing tasks depends on the quality of the features extracted from the raw data. Insightful feature representation increases the performance of many learning algorithms by exposing the underlying explanatory factors of the output for the unobserved input. A good representation should also handle anomalies in the data such as missing samples and noisy input caused by the undesired, external factors of variation. It should also reduce the data redundancy. Over the years, many feature extraction processes have been invented to produce good representations of raw images and videos.
The feature extraction processes can be categorized into three groups. The first group contains processes that are hand-crafted for a specific task. Hand-engineering features requires the knowledge of domain experts and manual labor. However, the feature extraction process is interpretable and explainable. Next group contains the latent-feature extraction processes. While the original feature lies in a high-dimensional space, the relevant factors for a task often lie on a lower dimensional manifold. The latent-feature extraction employs hidden variables to expose the underlying data properties that cannot be directly measured from the input. Latent features seek a specific structure such as sparsity or low-rank into the derived representation through sophisticated optimization techniques. The last category is that of deep features. These are obtained by passing raw input data with minimal pre-processing through a deep network. Its parameters are computed by iteratively minimizing a task-based loss.
In this dissertation, I present four pieces of work where I create and learn suitable data representations. The first task employs hand-crafted features to perform clinically-relevant retrieval of diabetic retinopathy images. The second task uses latent features to perform content-adaptive image enhancement. The third task ranks a pair of images based on their aestheticism. The goal of the last task is to capture localized image artifacts in small datasets with patch-level labels. For both these tasks, I propose novel deep architectures and show significant improvement over the previous state-of-art approaches. A suitable combination of feature representations augmented with an appropriate learning approach can increase performance for most visual computing tasks.
The feature extraction processes can be categorized into three groups. The first group contains processes that are hand-crafted for a specific task. Hand-engineering features requires the knowledge of domain experts and manual labor. However, the feature extraction process is interpretable and explainable. Next group contains the latent-feature extraction processes. While the original feature lies in a high-dimensional space, the relevant factors for a task often lie on a lower dimensional manifold. The latent-feature extraction employs hidden variables to expose the underlying data properties that cannot be directly measured from the input. Latent features seek a specific structure such as sparsity or low-rank into the derived representation through sophisticated optimization techniques. The last category is that of deep features. These are obtained by passing raw input data with minimal pre-processing through a deep network. Its parameters are computed by iteratively minimizing a task-based loss.
In this dissertation, I present four pieces of work where I create and learn suitable data representations. The first task employs hand-crafted features to perform clinically-relevant retrieval of diabetic retinopathy images. The second task uses latent features to perform content-adaptive image enhancement. The third task ranks a pair of images based on their aestheticism. The goal of the last task is to capture localized image artifacts in small datasets with patch-level labels. For both these tasks, I propose novel deep architectures and show significant improvement over the previous state-of-art approaches. A suitable combination of feature representations augmented with an appropriate learning approach can increase performance for most visual computing tasks.
ContributorsChandakkar, Parag Shridhar (Author) / Li, Baoxin (Thesis advisor) / Yang, Yezhou (Committee member) / Turaga, Pavan (Committee member) / Davulcu, Hasan (Committee member) / Arizona State University (Publisher)
Created2017