Physical Sciences and Mathematics Commons^™

Neuron Clustering For Mitigating Catastrophic Forgetting In Supervised And Reinforcement Learning, Benjamin Frederick Goodrich

Doctoral Dissertations

Neural networks have had many great successes in recent years, particularly with the advent of deep learning and many novel training techniques. One issue that has affected neural networks and prevented them from performing well in more realistic online environments is that of catastrophic forgetting. Catastrophic forgetting affects supervised learning systems when input samples are temporally correlated or are non-stationary. However, most real-world problems are non-stationary in nature, resulting in prolonged periods of time separating inputs drawn from different regions of the input space.

Reinforcement learning represents a worst-case scenario when it comes to precipitating catastrophic forgetting in neural networks. …

Algorithm-Based Fault Tolerance For Two-Sided Dense Matrix Factorizations, Yulu Jia

Doctoral Dissertations

The mean time between failure (MTBF) of large supercomputers is decreasing, and future exascale computers are expected to have a MTBF of around 30 minutes. Therefore, it is urgent to prepare important algorithms for future machines with such a short MTBF. Eigenvalue problems (EVP) and singular value problems (SVP) are common in engineering and scientific research. Solving EVP and SVP numerically involves two-sided matrix factorizations: the Hessenberg reduction, the tridiagonal reduction, and the bidiagonal reduction. These three factorizations are computation intensive, and have long running times. They are prone to suffer from computer failures.

We designed algorithm-based fault tolerant (ABFT) …

Batched Linear Algebra Problems On Gpu Accelerators, Tingxing Dong

Doctoral Dissertations

The emergence of multicore and heterogeneous architectures requires many linear algebra algorithms to be redesigned to take advantage of the accelerators, such as GPUs. A particularly challenging class of problems, arising in numerous applications, involves the use of linear algebra operations on many small-sized matrices. The size of these matrices is usually the same, up to a few hundred. The number of them can be thousands, even millions.

Compared to large matrix problems with more data parallel computation that are well suited on GPUs, the challenges of small matrix problems lie in the low computing intensity, the large sequential operation …

Automated Style Feedback For Advanced Beginner Java Programmers, Hannah Blau

Doctoral Dissertations

FrenchPress is an Eclipse plug-in that partially automates the task of giving students feedback on their Java programs. It is designed not for novices but for students taking their second or third Java course: students who know enough Java to write a working program but lack the judgment to recognize bad code when they see it. FrenchPress does not diagnose compile-time or run-time errors, or logical errors that produce incorrect output. It targets silent flaws, flaws the student is unable to identify for himself because nothing in the programming environment alerts him. FrenchPress diagnoses flaws characteristic of programmers who have …

Content Placement As A Key To A Content-Dominated, Highly Mobile Internet, Abhigyan Sharma

Doctoral Dissertations

Most of the Internet traffic is content, and most of the Internet connected hosts are mobile. Our work focuses on the design of infrastructure services needed to support such a content-dominated, highly mobile Internet. In the design of these services, three sets of decisions arise frequently: (1) Content placment for selecting the locations where a content is placed, (2) request redirection for selecting the location where a particular request is served from and (3) network routing for selecting the physical path between clients and the services they are accessing. Our central thesis is that content placement is a powerful factor, …

Informed Search For Learning Causal Structure, Brian J. Taylor

Doctoral Dissertations

Over the past twenty-five years, a large number of algorithms have been developed to learn the structure of causal graphical models. Many of these algorithms learn causal structures by analyzing the implications of observed conditional independence among variables that describe characteristics of the domain being analyzed. They do so by applying inference rules, data analysis operations such as the conditional independence tests, each of which can eliminate large parts of the space of possible causal structures. Results show that the sequence of inference rules used by PC, a widely applied algorithm for constraint-based learning of causal models, is effective but …

Exploiting Social Media Sources For Search, Fusion And Evaluation, Chia-Jung Lee

Doctoral Dissertations

The web contains heterogeneous information that is generated with different characteristics and is presented via different media. Social media, as one of the largest content carriers, has generated information from millions of users worldwide, creating material rapidly in all types of forms such as comments, images, tags, videos and ratings, etc. In social applications, the formation of online communities contributes to conversations of substantially broader aspects, as well as unfiltered opinions about subjects that are rarely covered in public media. Information accrued on social platforms, therefore, presents a unique opportunity to augment web sources such as Wikipedia or news pages, …

Safe Reinforcement Learning, Philip S. Thomas

Doctoral Dissertations

This dissertation proposes and presents solutions to two new problems that fall within the broad scope of reinforcement learning (RL) research. The first problem, high confidence off-policy evaluation (HCOPE), requires an algorithm to use historical data from one or more behavior policies to compute a high confidence lower bound on the performance of an evaluation policy. This allows us to, for the first time, provide the user of any RL algorithm with confidence that a newly proposed policy (which has never actually been used) will perform well. The second problem is to construct what we call a safe reinforcement learning …

Forensic And Management Challenges In Wireless And Mobile Network Environment, Sookhyun Yang

Doctoral Dissertations

The Internet recently passed an historic inflection point, with the number of broadband wireless/mobile devices surpassing the number of wired PCs and servers connected to the Internet. Smartphones, laptops, tablets, machine-to-machine (M2M) devices, and other portable devices have penetrated our daily lives. According to Cisco, by 2018, wired devices will account for only 39% of IP traffic, with the remaining traffic produced by wireless/mobile devices. This proliferation of wireless/mobile devices is profoundly changing many of the characteristics of network applications, protocols, and operation, and posing fundamental challenges to the Internet architecture. In light of this new trend, this thesis focuses …

Design And Implementation Of An Economy Plane For The Internet, Xinming Chen

Doctoral Dissertations

The Internet has been very successful in supporting many network applications. As the diversity of uses for the Internet has increased, many protocols and services have been developed by the industry and the research community. However, many of them failed to get deployed in the Internet. One challenge of deploying these novel ideas in operational network is that the network providers need to be involved in the process. Many novel network protocols and services, like multicast and end-to-end QoS, need the support from network providers. However, since network providers are typically driven by business reasons, if they can not get …

Energy-Efficient Content Delivery Networks, Vimal Mathew

Doctoral Dissertations

Internet-scale distributed systems such as content delivery networks (CDNs) operate hundreds of thousands of servers deployed in thousands of data center locations around the globe. Since the energy costs of operating such a large IT infrastructure are a significant fraction of the total operating costs, we argue for redesigning them to incorporate energy optimization as a first-order principle. We focus on CDNs and demonstrate techniques to save energy while meeting client-perceived service level agreements (SLAs) and minimizing impact on hardware reliability. Servers deployed at individual data centers can be switched off at low load to save energy. We show that …

Skeleton Structures And Origami Design, John C. Bowers

Doctoral Dissertations

In this dissertation we study problems related to polygonal skeleton structures that have applications to computational origami. The two main structures studied are the straight skeleton of a simple polygon (and its generalizations to planar straight line graphs) and the universal molecule of a Lang polygon. This work builds on results completed jointly with my advisor Ileana Streinu. Skeleton structures are used in many computational geometry algorithms. Examples include the medial axis, which has applications including shape analysis, optical character recognition, and surface reconstruction; and the Voronoi diagram, which has a wide array of applications including geographic information systems …

General Program Synthesis From Examples Using Genetic Programming With Parent Selection Based On Random Lexicographic Orderings Of Test Cases, Thomas Helmuth

Doctoral Dissertations

Software developers routinely create tests before writing code, to ensure that their programs fulfill their requirements. Instead of having human programmers write the code to meet these tests, automatic program synthesis systems can create programs to meet specifications without human intervention, only requiring examples of desired behavior. In the long-term, we envision using genetic programming to synthesize large pieces of software. This dissertation takes steps toward this goal by investigating the ability of genetic programming to solve introductory computer science programming problems. We present a suite of 29 benchmark problems intended to test general program synthesis systems, which we systematically …

Exploiting Concepts In Videos For Video Event Detection, Ethem Can

Doctoral Dissertations

Video event detection is the task of searching videos for events of interest to a user where an event is a complex activity which is localized in time and space. The video event detection problem has gained more importance as the amount of online video is increasing by more than 300 hours every minute on Youtube alone. In this thesis, we tackle three major video event detection problems: video event detection with exemplars (VED-ex), where a large number of example videos are associated with queries; video event detection with few exemplars (VED-ex_few), in which only a small number of example …

Network Characteristics And Dynamics: Reciprocity, Competition And Information Dissemination, Bo Jiang

Doctoral Dissertations

Networks are commonly used to study complex systems. This often requires a good understanding of the structural characteristics and evolution dynamics of networks, and also their impacts on a variety of dynamic processes taking place on top of them. In this thesis, we study various aspects of networks characteristics and dynamics, with a focus on reciprocity, competition and information dissemination. We first formulate the maximum reciprocity problem and study its use in the interpretation of reciprocity in real networks. We propose to interpret reciprocity based on its comparison with the maximum possible reciprocity for a network exhibiting the same degrees. …

Threat Analysis, Countermeaures And Design Strategies For Secure Computation In Nanometer Cmos Regime, Raghavan Kumar

Doctoral Dissertations

Advancements in CMOS technologies have led to an era of Internet Of Things (IOT), where the devices have the ability to communicate with each other apart from their computational power. As more and more sensitive data is processed by embedded devices, the trend towards lightweight and efficient cryptographic primitives has gained significant momentum. Achieving a perfect security in silicon is extremely difficult, as the traditional cryptographic implementations are vulnerable to various active and passive attacks. There is also a threat in the form of "hardware Trojans" inserted into the supply chain by the untrusted third-party manufacturers for economic incentives. Apart …

Variation In Human-Intensive Systems: A Conceptual Framework For Characterizing, Modeling, And Analyzing Families Of Systems, Borislava I. Simidchieva

Doctoral Dissertations

A system model---namely a formal definition of the coordination of people, hardware devices, and software components performing activities, using resources and artifacts, and producing various outputs---can aid understanding of the real-world system it models. Complex real-world systems, however, exhibit considerable amounts of variation that can be difficult or impossible to represent within a single model. This dissertation evaluates the hypothesis that the careful characterization and representation of system variation can aid in the generation and analysis of concrete system instances related to one another in specified ways and manifesting different kinds of variation. When a set of closely related systems …

A Platform For Scalable Low-Latency Analytics Using Mapreduce, Boduo Li

Doctoral Dissertations

Today, the ability to process "big data" has become crucial to the information needs of many enterprise businesses, scientific applications, and governments. Recently, there have been increasing needs of processing data that is not only "big" but also "fast". Here "fast data" refers to high-speed real-time and near real-time data streams, such as Twitter feeds, search query streams, click streams, impressions, and system logs. To handle both historical data and real-time data, many companies have to maintain multiple systems. However, recent real-world case studies show that maintaining multiple systems cause not only code duplication, but also intensive manual work to …

Robust Mobile Data Transport: Modeling, Measurements, And Implementation, Yung-Chih Chen

Doctoral Dissertations

Advances in wireless technologies and the pervasive influence of multi-homed devices have significantly changed the way people use the Internet. These changes of user behavior and the evolution of multi-homing technologies have brought a huge impact to today's network study and provided new opportunities to improve mobile data transport. In this thesis, we investigate challenges related to human mobility, with emphases on network performance at both system level and user level. More specifically, we seek to answer the following two questions: 1) How to model user mobility in the networks and use the model for network provisioning? 2) Is it …

Versatile Three-Phase Power Electronics Converter Based Real-Time Load Emulators, Jing Wang

Doctoral Dissertations

This dissertation includes the methodology, implementation, validation, as well as real-time modeling of a load emulator for a reconfigurable grid emulation platform of hardware test-bed (HTB). This test-bed was proposed by Center of Ultra-wide-area Resilient Electric Energy Transmission Network (CURENT) at the University of Tennessee, at Knoxville in 2011, to address the transmission level system challenges posed by contemporary fast changing energy technologies.

Detailed HTB introduction, including design concept, fundamental units and hardware construction, is elaborated. In the development, current controlled three-phase power electronics converter based emulator unit is adopted to create desired power system loading conditions.

In the application, …

Nucleosynthesis In Self-Consistent Core-Collapse Supernova Models Using Multidimensional Chimera Simulations, James Austin Harris

Doctoral Dissertations

Observations of nuclear abundances in core-collapse supernova (CCSN) ejecta, highlighted by γ-ray [gamma-ray] observations of the ⁴⁴Ti [titanium-44] spatial distribution in the nearby supernova remnants Cassiopeia A and SN 1987A, allow nucleosynthesis calculations to place powerful constraints on conditions deep in the interiors of supernovae and their progenitor stars. This ability to probe where direct observations cannot makes such calculations an invaluable tool for understanding the CCSN mechanism. Unfortunately, despite knowing for two decades that supernovae are intrinsically multi-dimensional events, discussions of CCSN nucleosynthesis have been predominantly based on spherically symmetric (1D) models, which employ a contrived energy source …

Intrusion Detection System Of Industrial Control Networks Using Network Telemetry, Stanislav Ponomarev

Doctoral Dissertations

Industrial Control Systems (ICSs) are designed, implemented, and deployed in most major spheres of production, business, and entertainment. ICSs are commonly split into two subsystems - Programmable Logic Controllers (PLCs) and Supervisory Control And Data Acquisition (SCADA) systems - to achieve high safety, allow engineers to observe states of an ICS, and perform various configuration updates. Before wide adoption of the Internet, ICSs used "air-gap" security measures, where the ICS network was isolated from other networks, including the Internet, by a physical disconnect [1]. This level of security allowed ICS protocol designers to concentrate on the availability and safety of …

Neuroscience-Inspired Dynamic Architectures, Catherine Dorothy Schuman

Doctoral Dissertations

Biological brains are some of the most powerful computational devices on Earth. Computer scientists have long drawn inspiration from neuroscience to produce computational tools. This work introduces neuroscience-inspired dynamic architectures (NIDA), spiking neural networks embedded in a geometric space that exhibit dynamic behavior. A neuromorphic hardware implementation based on NIDA networks, Dynamic Adaptive Neural Network Array (DANNA), is discussed. Neuromorphic implementations are one alternative/complement to traditional von Neumann computation. A method for designing/training NIDA networks, based on evolutionary optimization, is introduced. We demonstrate the utility of NIDA networks on classification tasks, a control task, and an anomaly detection task. There …

Computational Framework For Small Animal Spect Imaging: Simulation And Reconstruction, Sang Hyeb Lee

Doctoral Dissertations

Small animal Single Photon Emission Computed Tomography (SPECT) has been an invaluable asset in biomedical science since this non-invasive imaging technique allows the longitudinal studies of animal models of human diseases. However, the image degradation caused by non-stationary collimator-detector response and single photon emitting nature of SPECT makes it difficult to provide a quantitative measure of 3D radio-pharmaceutical distribution inside the patient. Moreover, this problem exacerbates when an intra-peritoneal X-ray contrast agent is injected into a mouse for low-energy radiotracers.

In this dissertation, we design and develop a complete computational framework for the entire SPECT scan procedure from the radio-pharmaceutical …

Sensitivity Of Mixed Models To Computational Algorithms Of Time Series Data, Gunaime Nevine

Doctoral Dissertations

Statistical analysis is influenced by implementation of the algorithms used to execute the computations associated with various statistical techniques. Over many years; very important criteria for model comparison has been studied and examined, and two algorithms on a single dataset have been performed numerous times. The goal of this research is not comparing two or more models on one dataset, but comparing models with numerical algorithms that have been used to solve them on the same dataset.

In this research, different models have been broadly applied in modeling and their contrasting which are affected by the numerical algorithms in different …

Epistemological Databases For Probabilistic Knowledge Base Construction, Michael Louis Wick

Doctoral Dissertations

Knowledge bases (KB) facilitate real world decision making by providing access to structured relational information that enables pattern discovery and semantic queries. Although there is a large amount of data available for populating a KB; the data must first be gathered and assembled. Traditionally, this integration is performed automatically by storing the output of an information extraction pipeline directly into a database as if this prediction were the ``truth.'' However, the resulting KB is often not reliable because (a) errors accumulate in the integration pipeline, and (b) they persist in the KB even after new information arrives that could rectify …

Long Range Motion Estimation And Applications, Laura Sevilla-Lara

Doctoral Dissertations

Finding correspondences between images underlies many computer vision problems, such as op- tical flow, tracking, stereovision and alignment. Finding these correspondences involves formulating a matching function and optimizing it. This optimization process is often gradient descent, which avoids exhaustive search, but relies on the assumption of being in the basin of attraction of the right local minimum. This is often the case when the displacement is small, and current methods obtain very accurate results for small motions. However, when the motion is large and the matching function is abrupt this assumption is less likely to be true. One traditional way …

Learning With Joint Inference And Latent Linguistic Structure In Graphical Models, Jason Narad

Doctoral Dissertations

Constructing end-to-end NLP systems requires the processing of many types of linguistic information prior to solving the desired end task. A common approach to this problem is to construct a pipeline, one component for each task, with each system's output becoming input for the next. This approach poses two problems. First, errors propagate, and, much like the childhood game of "telephone", combining systems in this manner can lead to unintelligible outcomes. Second, each component task requires annotated training data to act as supervision for training the model. These annotations are often expensive and time-consuming to produce, may differ from each …

Effectiveness Of Cloud Services For Scientific And Vod Applications, Dilip Kumar Krishnappa

Doctoral Dissertations

Cloud platforms have emerged as the primary data warehouse for a variety of applications, such as DropBox, iCloud, Google Music, etc. These applications allow users to store data in the cloud and access it from anywhere in the world. Commercial clouds are also well suited for providing high-end servers for rent to execute applications that require computation resources sporadically. Cloud users only pay for the time they actually use the hardware and the amount of data that is transmitted to and from the cloud, which has the potential to be more cost effective than purchasing, hosting, and maintaining dedicated hardware. …

Model-Based Guidance For Human-Intensive Processes, Stefan Christov

Doctoral Dissertations

Human-intensive processes (HIPs), such as medical processes involving coordination among doctors, nurses, and other medical staff, often play a critical role in society. Despite considerable work and progress in error reduction, human errors are still a major concern for many HIPs. To address this problem of human errors in HIPs, this thesis investigates two approaches for online process guidance, i.e., for guiding process performers while a process is being executed. Both approaches rely on monitoring a process execution and base the guidance they provide on a detailed formal process model that captures the recommended ways to perform the corresponding HIP. …