WorldCat Identities

! University of Georgia.! Department of Computer Science

Overview
Works: 49 works in 49 publications in 1 language and 48 library holdings
Genres: Academic theses 
Publication Timeline
.
Most widely held works by ! University of Georgia.! Department of Computer Science
Learning driver preferences for freeway merging using multitask irl by Sanath Govinda Bhat( )

1 edition published in 2017 in English and held by 1 WorldCat member library worldwide

Most automobile manufacturers today have invested heavily in the research and design of implementing autonomy in their cars. One important and challenging problem faced by a self-driven car on highways is merging into the highway from an acceleration ramp. Successful merging needs consideration of the behaviors of cars driving in the outermost highway lane which is adjacent to the merging lane, especially, the behaviors of those cars that would potentially become the leading or following car after a successful merge. We attempt to predict the motivation for the behaviors of those cars driving on the outermost highway lanes near the merging area hypothesizing that they perform a series of tasks each of which is driven by different motivations while passing through each section of the merging area. We use a Hierarchical Bayesian model to model the preferences in each task and the priors over those preferences
A distributed cloud-based platform for FMRI big data analytics by Milad Sadeghi Makkie( )

1 edition published in 2018 in English and held by 1 WorldCat member library worldwide

The sheer complexity of the brain has forced the neuroscience community and specifically the neuroimaging experts to transit from the smaller brain datasets to much larger hard-to-handle ones. The primary goal of flagship projects such as the BRAIN Initiative and Human Brain Project is to gain a better understanding of the human brain and to treat the neurological and psychiatric disorders through the cutting-edge technologies in the biomedical imaging field. In the context of fMRI, the primary challenge is obtaining meaningful results from the intrinsic complex structure of large fMRI data and lack of clear insight into the underlying neural activities. However, archiving, analyzing, and sharing the fast-growing neuroimaging datasets posed significant challenges. New computational methods and technologies have emerged in the domain of Big Data but have not been fully adapted for use in neuroimaging. In this dissertation, I introduce my efforts toward creating a comprehensive platform to store, to manage and to process such datasets. I further present my GPU-based deep learning solution for distributed data processing that employs TensorFlow, Apache Spark, and Hadoop using cloud computing services. Finally, I demonstrate the significant performance gains of our platform enabling data-driven extraction of hierarchical information from massive fMRI data using a distributed deep convolutional autoencoder model
Vision and language : an application-oriented perspective by Karan Sharma( )

1 edition published in 2018 in English and held by 1 WorldCat member library worldwide

Vision and language are two of the most critical human faculties. If we are to develop more useful Artificial Intelligence (AI) systems, these modalities will need to work in tandem. Although we are still far from the ultimate goal of synergetic integration of vision and language, several practical applications lying at the intersection of computer vision (CV) and natural language processing (NLP) have experienced a huge upsurge in recent times. This upsurge in the integration of vision and language has been accelerated by recent advances in deep learning and ready availability of both, benchmark and real-world datasets. In this dissertation, we address a few interesting and important applications, such as automated image captioning and classification of objects and actions in images, that lie at the intersection of CV and NLP and have a significant potential impact in important problem domains such as information retrieval and product marketing. First, we propose an approach to speed up image caption retrieval guided by the top object detected in an image. Second, we propose an approach to classify an action in an image without executing explicit action classifiers on the image. In this approach, we first detect objects in an image and then, with the aid of top objects and associated word embeddings obtained via training on a natural language corpus, we infer the the most probable action in the image. Next, we propose a model to guess objects in an image in situations where the datasets for training classifiers for such objects are unavailable. Finally, we conduct a similarity study on consumer products using both visual and textual features. We believe that these studies and the proposed models will provide practitioners with insights that they could apply in designing AI systems for specific applications
Enabling efficient detection and forensic investigation of malicious software downloads by Krishna Phani Kumar Vadrevu( )

1 edition published in 2017 in English and held by 1 WorldCat member library worldwide

Malware infections have been on a steady rise recently causing increasing amounts of financial damage and data loss. The advent of techniques like code polymorphism has resulted in a huge proliferation of new malware. Traditional malware detection and analysis methods have lagged behind in grappling with this problem. In this research work, we developed novel malware detection and analysis systems to help alleviate this problem in a comprehensive manner. Firstly, we developed a system called AMICO that can detect malicious software down- loads in live web traffic. AMICO uses supervised learning techniques to learn a prove- nance classifier that takes into consideration the network user behavior and can differentiate between malware and benign software downloads. Pilot tests showed that AMICO can detect up to 90% of the malware downloads at about 0.1% false positive rate. Furthermore, to make the analysis of all downloaded malware more scalable, we developed a system called MAXS. MAXS is a novel probabilistic framework for scaling execution of malware in analysis envi- ronments. A prototype implementation and evaluation with large real-world datasets showed that MAXS can reduce up to 50% of malware execution time with less than 0.3% information loss. Finally, to aid in the forensic investigation all these malware downloads, we developed ChromePic, a web browser with an embedded forensic engine. ChromePic is a light-weight, portable, efficient and always-on tool that records fine-grained browser logs, including screen- shots and state of the DOM during user interactions. Experimental evaluation on multiple platforms showed that ChromePic is useful in deconstructing various real world web attacks. Also, it has an overhead of less than 150 ms thus making the tool practically imperceptible to the end user
Sail: a system for adaptive interest-based learning in stem education by Karen Elizabeth Aguar( )

1 edition published in 2017 in English and held by 1 WorldCat member library worldwide

The aim of this research is to alleviate many challenges faced in STEM education through the creation of a scalable, adaptive learning framework that supports interest-based learning (IBL) in multiple domains. Adaptive Learning is the idea that software and material should "adapt" to individual student's needs, typically based on previous knowledge, pace, or learning style. This research takes a less explored approach by adapting content and practice problems based on a student's interests. Interest-based learning (IBL) has been shown to improve intrinsic motivation, leading to better learning and achievements, but no solution currently exists to facilitate and promote IBL across multiple domains. This work presents the design and pilot of SAIL, a System for Adaptive Interest-based Learning, to easily facilitate IBL in an adaptive and scalable platform. SAIL is not limited by domain, but was designed with STEM subjects in mind due to their high applicability in other fields. With SAIL, one student in an introductory programming course could practice loops through sports-themed examples while another could learn through music or science. SAIL was designed to help alleviate many of the concerns in STEM education by providing a competent and compelling curriculum delivering individualized instruction to help increase motivation, performance and fill the gaps in STEM education. SAIL showcases the interconnectivity of STEM subjects with other fields, combatting misperceptions and increasing motivation to help attract and retain a larger and more diverse population of students. With SAIL, students become active participants in their learning experience as they utilize an interactive map to traverse their unique path through interest-based course material. A large pilot study (N=307) in the context of introductory programming (Java) was conducted comparing a class using SAIL to three other classes with varying control conditions. This study resulted in new quantitative and qualitative knowledge about how SAIL can impact introductory Computer Science (CS) as well as assessing viability for other STEM fields, including K-12 STEM education. Via SAIL, we raise the standard of education, increase enjoyment, remedy gender disparities, and aid in encouraging more students to continue their CS education
Query processing in graph databases by Supriya Ramireddy( )

1 edition published in 2017 in English and held by 1 WorldCat member library worldwide

Graph data are extensively associated with state-of-the-art applications in a variety of domains which include Linked Data and Social Media. This drives the need to have graph databases that can effectively store and manage graph data. Relational query processing has become efficient due to many decades of research in the field of data management and processing, among which translating SQL into relational algebra operations plays a key role in query processing. Based on relational algebra, many graph algebras have been defined that can be used for query processing and optimization in graph databases. We propose a graph algebra which operates on graph databases, for processing queries. We have implemented a graph algebra as a part of ScalaTion and compared it with Neo4j and MySQL with respect to query processing times. Various queries are tested on datasets with a few vertices to a large number of vertices. Graph databases perform well when the database gets larger compared to relational databases. Increase in the number of joins in queries, decreases the performance of relational databases, whereas equivalent queries in graph databases comparatively exhibit good performance. Among graph databases compared in the study, ScalaTion shows better performance
Supporting open science in big data frameworks and data science education by Michael Edward Cotterell( )

1 edition published in 2017 in English and held by 1 WorldCat member library worldwide

As the prevalence of data grows throughout the Big Data era, so does a need to provide and improve tools for the education and application of data-driven ana- lytics and scientific investigation. The main contributions of this research can be summarized as follows: i) We provide an overview of the open source ScalaTion project, a big data framework that supports big data analytics, simulation mod- eling, and functional data analysis. ii) We outline some of the Functional Data support in ScalaTion, including a performance comparison for the evaluation of B-spline basis functions that shows that our method is faster than some other popular libraries. iii) To demonstrate how to provide lightweight big data frame- work integration in open notebooks, we present the open source ScalaTion Kernel project, a custom Jupyter kernel that enables ScalaTion support in Jupyter note- books. iv) To demonstrate research using ScalaTion, we outline and evaluate a tight clustering algorithm, written using ScalaTion, for the functional data anal- ysis of time course omics data. v) To promote reproducibility in open science, we present the Applied Open Data Science (AODS) project, a collection of customized web applications for the hosting and sharing of open notebooks with ScalaTion support. This project also includes shareable, executable, and modifiable exam- ple notebooks that utilize ScalaTion to demonstrate various data science topics as well as detailed documentation on how to easily reproduce the environment in which the notebooks are hosted. Specifically, we propose and demonstrate, via readily accessible examples, methods to facilitate openness and reproducibility (both of results and infrastructure) in data science investigations using a big data framework
Experimental study of shapelet based personalized human activity classification by Nitin Saroha( )

1 edition published in 2018 in English and held by 1 WorldCat member library worldwide

Human activity recognition (HAR) has many applications in both research and industry domains. It has a variety of applications in healthcare, cognitive assistance, and tracking, because the sensors are becoming both better and smaller to be used in a wearable device. Many HAR systems have been developed to detect human activities after the emergence of wearable motion sensors. But most use complex feature extraction methods to classify the activities to attain high accuracy. In this work, we use a waveform pattern matching approach that allows us to extract a small representative waveform for each activity called "Activity Shapelet" ("A-Shapelet", for short). The focus of this work is to use a shapelet to accurately classify activities. The advantages of this method is to use raw data to classify activities. This eliminates the need for defining and distinguishing among different features of each activity for classifying them accurately. We provide important concepts related to shapelets and the process involved in the classification of activities. We give detailed analysis of accuracy results based on different parameters
Probabilistic topic modeling based framework for exploration of rdf data by Seyedamin Pouriyeh( )

1 edition published in 2018 in English and held by 1 WorldCat member library worldwide

During the past few decades, the amount of Web content has grown exponentially. Recently, a vast amount of datasets in a variety of domains has been published as part of the Linked Open Data (LOD) project. As a result, exploring and exploiting these massive heterogeneous datasets, which are typically represented using the Resource Description Framework (RDF), has gained a considerable attention. With this work, we aim to address exploring and exploiting of such datasets within two broad categories: RDF dataset summarization and RDF dataset profiling. With respect to RDF dataset summarization, we focus on entity summarization, which aims to produce an abridged, but sufficient descriptions of all entities in the dataset. In other words, entity summarization is a way to absorb and distill descriptive knowledge from RDF datasets. We propose a probabilistic topic model using Latent Dirichlet Allocation (LDA) for the entity summarization task called ES-LDA and its extension, ES-LDA$ext, which combines prior knowledge with statistical learning techniques within a single framework, in order to create more reliable and representative summaries of entities. We demonstrate the effectiveness of our approach by conducting extensive experiments and show that our models outperform state-of-the-art techniques and enhance the quality of the entity summaries. RDF dataset profiling is a task that involves generating a proper profile for RDF datasets on the Web so that they can be discovered more easily. Basically, RDF dataset profiles are expected to facilitate data discovery, consumption, and integration with statistics and useful metadata about the content of the RDF datasets. We propose topic-wise RDF dataset profiling, called R-LDA, using LDA technique. In our model, we identify a number of topics that can represent an RDF dataset and assign a set of Wikipedia categories to the obtained topics that are semantically relevant, understandable, and cover the discovered topics well. The union of the assigned categories serves as a profile of the dataset, in a sense that it provides an overall characterization of the dataset's content
Improving the dual cardinality simulation algorithms by Luis Anggelo Bernaola Ibarra( )

1 edition published in 2018 in English and held by 1 WorldCat member library worldwide

Graph pattern matching is typically defined in terms of subgraph isomorphism, which makes it an NP-complete/NP-hard problem. Isomorphism algorithms requires bijective functions which can be too restrictive to identify patterns in real-world applications. Moreover, real-world graphs may contain some noise and the problem of finding the exact match can be very expensive. In order to avoid the combinatorial worst-case time complexity of subgraph isomorphism, we extend prior work on dual cardinality simulation. According to our experiments, this type of graph simulation offers high precision with good performance in large graphs. Precision is acquired because dual cardinality simulation checks the constraint in which the number of matching children or parents with the same label in the data graph should not be less than their correspondents in the query graphs. For improving the performance, we have introduced the concept of count sets which are computed before dual cardinality simulation is executed. Experiments are done on large graphs using synthetic and real-world graphs
Brain controlled robot navigation based on low cost EEG by Yang Shi( )

1 edition published in 2017 in English and held by 1 WorldCat member library worldwide

This thesis focuses on employing low cost EEG signals to control robots for navigation tasks. A data driven signal processing and machine learning framework is proposed and applied. Power Spectral Density (PSD) and Spectral Analysis are used for feature extraction, and I examined the result of Principle Component Analysis (PCA), and chose non-linear classifiers for machine learning. The algorithm for classification is Quadratic Discriminant Analysis (QDA), and achieved around 88% to 91% accuracy for five-fold cross validation. When testing with a new dataset, the accuracy is around 82%, but will be low in contaminated datasets and at varying electrode locations. I also experimented the real-time system, and most instructions are correctly classified. This thesis provides a novel system for EEG data processing, especially for situations of low cost, low channel amount equipment
Secured mobile medical image sharing by Jinze Li( )

1 edition published in 2017 in English and held by 1 WorldCat member library worldwide

Secured Mobile Medical Image Sharing provides a new solution to share medical images on mobile device. It successfully constructs a bridge between doctors or other hospital staff and patients. More importantly, encrypting DICOM images before it is shared from inner network to outer network can effectively protect patient privacy. Nowadays, most medical information is shared by physical or digital copies. However, both of them have some disadvantages, such as high cost, transfer security and etc. Although it is impossible to completely solve all those problems through utilizing this new solution, it is necessary to reduce negative impact of those issues. In this thesis, I propose an application on Android device to securely store and share medical images by utilizing cloud storage and asymmetric encryption. A testing was carried out on Android devices
Model-based IRL with continuous action spaces by Anuja Pradeep Nagare( )

1 edition published in 2018 in English and held by 1 WorldCat member library worldwide

Inverse reinforcement learning (IRL) seeks to learn the preferences of an expert agent performing a task from the expert's demonstrations. More specifically, it seeks to find the reward function of the expert modelled as a Markov decision process from observations of its state-action trajectories. IRL's ability to use an expert agent's demonstrations of real-world activities, such as driving, locomotion tasks, and other robotic tasks to build intelligent agents makes IRL significant. This research provides a novel method for preference learning by developing a model-based IRL algorithm for continuous action spaces. It generalizes a previous Bayesian approach to IRL to include continuous action spaces and uses the trust region policy optimization in the method. Action space densities are generated for each state using a random walk, and an online transition model is used. Our method learns the reward function of an expert agent with a continuous action space and uses this learned function to complete the underlying MDP and predict an optimal policy. Experimental results over a benchmark problem domain called Object-World and toward modelling driver behavior on congested freeways offer evidence about the benefits of this approach
Accurate prediction of human mirna targets via graph modeling and machine learning approaches by Mohammad Mohebbi( )

1 edition published in 2017 in English and held by 1 WorldCat member library worldwide

MiRNAs are small endogenous non-coding RNA molecules that have a critical function in suppressing genes and they also correlate with many diseases and cancers. Due to the importance of their effects in several cell activities, discovering their mechanisms is an important task. Because the functionality of miRNAs tightly connected to the way they recognize their targets miRNA target prediction has received a lot of attentions in research. Despite that, most of current methods suffer from high false positive rates and they are not able to provide much insight to the actual process of miRNA targeting. In this dissertation, we present two novel approaches aimed at addressing existing issues in miRNA target prediction; one approach to improve false positive rate and the other to substantiate multiple hypotheses pertaining to biological mechanism of miRNA targeting and to provide insight into the actual mechanism. To address the first issue, we present Correlation Graph model that captures nucleotide correlations between miRNA sequence and the target. This model makes it possible to characterize nucleotide correlations other than Watson-Crick base pairings between two parts of the duplex. We designed an SVM based algorithm and tested our model on human data and it achieved a sensitivity of 86% with a false positive rate below 13% which is a significant performance improvement in comparison to the state-of-the-art methods miRanda and RNAhybrid. The second part of this dissertation addresses the issue of understanding the mechanism of miRNA targeting. It contains a multi-hypothesis learner algorithm that utilizes features collected from literature pertaining to the mechanisms of targeting. These features enable the algorithm to partition data in a way very relevant to the biological features. The algorithm uses these partitions to learn multiple hypotheses. Our evaluations on human and mouse datasets show our method has comparable performance to that of high performance classifiers such as RandomForest. Moreover, feature selection on the resulting partitions confirms that the partitioning mechanism is compatible with biological mechanisms. These partitions could be used for further in vivo experiments to verify the currently proposed targeting approaches and to discover the new mechanisms
Frameworks and algorithms for individual planning under cooperation by Muthukumaran Chandrasekaran( )

1 edition published in 2017 in English and held by 1 WorldCat member library worldwide

Interactive Dynamic Influence Diagrams (I-DIDs) and Interactive Partially Observable Markov Decision Processes (I-POMDPs) are well-established finitely-nested frameworks that operationalize the planning and decision-making of a self-interested agent in a multiagent setting under uncertainty. Furthermore, I-DIDs (and I-POMDPs) take the perspective of a single agent in a multiagent setting and assume no communication or pre-coordination between agents. Therefore, intuitively, they are naturally suited for ad hoc or impromptu teamwork. However, we show that teamwork is implausible due to the way such frameworks operationalize bounded rationality of the agents. Before that, we first seek to scale I-DIDs in the number of agents by addressing the curse of dimensionality due to exponential growth in the number of models ascribed to the others by the subject agent over time using the well-known concept of stochastic bisimulation. Next, we investigate the implausibility of teamwork in such frameworks and present a principled way to induce it by augmenting I-DIDs with level-0 models enhanced with superior reasoning capabilities using reinforcement learning. We further investigate teamwork in open settings -- where one or more agents may leave or re-enter the system at will without announcing their arrival or departure to the others. As such, individual planning under the constraints of uncertainty, and lack of pre-coordination or communication between agents is complex. This complexity is exacerbated by agent openness. We present ways for individual agents to plan and act in open-agent teams within the context of a variant of the I-POMDP framework. Finally, we expose the void in the literature for theoretical frameworks that may be suitable for analyzing pragmatic interactions spanning several time steps between typed agents in multiagent teams. We fill that void by formally establishing a novel game-theoretic framework, called the Bayesian Markov Game, where bayesian agents with explicitly-defined finite-level types engage in a markov game where each agent has private but incomplete information regarding others' types. We characterize an equilibrium in this game and establish the conditions for its existence. In addition to laying strong theoretical foundations, we also empirically demonstrate the effectiveness of all our approaches and algorithms on multiple benchmark cooperative domains
A study of the correspondence problem and relevant applications in computer vision and graphics by Somenath Das( )

1 edition published in 2018 in English and held by 1 WorldCat member library worldwide

Correspondence determination between different objects plays a pivotal role in a wide range of applications in computer vision and computer graphics. In this dissertation, we address some key problems in computer vision and computer graphics that are dependent on accurate correspondence determination between the underlying objects under consideration. Following a general introduction to the correspondence problem in Chapter 1, in Chapter 2, we introduce a pairwise geodesic distance-based global shape representation for 3D shapes and exploit the spectrum of this representation to address correspondence determination between 3D shapes and, self-symmetry detection and detection of stable regions within 3D shapes. A surface differential-oriented global shape representation is introduced in Chapter 3 that is shown to encode the local surface geometry. We successfully exploit the spectrum of this representation for symmetry detection within a 3D shape and correspondence determination between 3D shapes. Furthermore, a novel criterion is introduced to measure the compatibility of the representation spectrum in the context of an important application such as deformation transfer. All the shapes under consideration are isometric transformation pairs (i.e., related via an isometric transformation). In Chapter 4, we present a comparative study of the performance of the shape representations introduced in Chapters 2 and 3 in the presence of noise. In addition, we introduce in Chapter 4 a biharmonic density-based surface point feature that is computed by exploiting the eigenspectra of the shape representations that are quintessential for establishing correspondence between shapes. Furthermore, we successfully apply the shape representation to address deformation transfer from a given source shape to a target shape. In Chapter 5, we address non-rigid structure from motion, a very important problem in computer vision, to extract 3D information from a 2D image sequence. To address this problem we impose a constraint on the distribution of the 2D correspondences between consecutive frames of the temporal image sequence. Finally, we conclude in Chapter 6 by giving an outline of some possible direction towards future extensions of the works presented. All the problems and applications considered in this dissertation either directly address or indirectly depend upon accurate correspondence determination between the different objects under consideration. In computer graphics these objects are the 3D shapes, whereas in computer vision the objects are the regions within the images. The results of the proposed framework in each chapter are compared to those from other relevant state-of-the-art schemes. It is shown that the proposed schemes perform competitively when compared with their state-of-the-art counterparts
Towards ciliary motion subtyping: representing patients as mixture of motion patterns by Alekhya Chennupati( )

1 edition published in 2017 in English and held by 1 WorldCat member library worldwide

Cilia are microscopic hair like projections that lie on almost every cell of the body. Motile cilia have a rhythmic beating motion to clear mucus and irritants. If the mucociliary defense mechanism does not work properly, it leads to a wide spectrum of diseases called ciliopathies. Identifying ciliopathies early and implementing proactive therapies is clinically compelling to minimize procedural invasiveness. However, previous work in this area was limited to separating normal from abnormal ciliary motion, and ignored the existence of broader spectrum of ciliary beat patterns that may have clinical implications with different disorders. Hence, defining a universal, quantitative "language" that describes phenotypes of ciliary motion is of particular clinical and translational interest. The analysis presented here groups patients with similar ciliary motion patterns, establishing a platform that can unravel ciliary motion subtypes in patients
HOON - a formalism supporting adaptive workflows by Yanbo Han( Book )

1 edition published in 1998 in English and held by 1 WorldCat member library worldwide

Analyzing android ad-libraries by Pranav Kalyan Panage( )

1 edition published in 2017 in English and held by 1 WorldCat member library worldwide

The business model for Mobile app developers today is in-app advertisements. These advertisements are implemented by including third-party Ad-libraries in Apk's. We analyzed applications to get more detailed statistics about the current trends in ad-library usage. Secondly, We generate forensic log information for ad-libraries behavior at the framework level. To achieve this, We hook all of the Android Framework API's that are exposed to applications. Along with this, we have altered the underlying Android operating system to get more precise information about activity component. These logs showed us what information is collected by ad-libraries from our devices. This framework API logging module is also aimed at improving our previous work, DroidForensics
An open science approach to exploring time-accuracy trade-offs in recommender systems by Khalid Jahangeer( )

1 edition published in 2017 in English and held by 1 WorldCat member library worldwide

Recommender Systems have become an integral part of our consumer dominated world. With the evolution of Big Data and the exponential expansion of consumerism it is imperative that we design efficient recommender systems. Collaborative Filtering techniques have been popularly adopted by researchers for developing robust recommender systems. In this paper we discuss some of the techniques that fall under the Collaborative Filtering umbrella and have been implemented within the ScalaTion big data analytics framework. Apart from discussing the implementation we analyze the execution time and accuracy of these techniques. This analysis has been performed to explore trade-offs between time and accuracy that occur while performing predictions using these techniques
 
moreShow More Titles
fewerShow Fewer Titles
Audience Level
0
Audience Level
1
  Kids General Special  
Audience level: 0.70 (from 0.69 for Secured mo ... to 0.94 for HOON - a f ...)

Languages
English (20)