Computer Sciences Commons^™

Direct Optimization For Classification With Boosting, Shaodan Zhai

Browse all Theses and Dissertations

Boosting, as one of the state-of-the-art classification approaches, is widely used in the industry for a broad range of problems. The existing boosting methods often formulate classification tasks as a convex optimization problem by using surrogates of performance measures. While the convex surrogates are computationally efficient to globally optimize, they are sensitive to outliers and inconsistent under some conditions. On the other hand, boosting's success can be ascribed to maximizing the margins, but few boosting approaches are designed to directly maximize the margin. In this research, we design novel boosting algorithms that directly optimize non-convex performance measures, including the empirical …

Contrast Pattern Aided Regression And Classification, Vahid Taslimitehrani

Browse all Theses and Dissertations

Regression and classification techniques play an essential role in many data mining tasks and have broad applications. However, most of the state-of-the-art regression and classification techniques are often unable to adequately model the interactions among predictor variables in highly heterogeneous datasets. New techniques that can effectively model such complex and heterogeneous structures are needed to significantly improve prediction accuracy. In this dissertation, we propose a novel type of accurate and interpretable regression and classification models, named as Pattern Aided Regression (PXR) and Pattern Aided Classification (PXC) respectively. Both PXR and PXC rely on identifying regions in the data space where …

Efficient Training Of Small Kernel Convolutional Neural Networks Using Fast Fourier Transform, Tyler Highlander

Browse all Theses and Dissertations

Convolutional neural networks (CNNs) are currently state-of-the-art for various classification tasks, but are computationally expensive. Propagating through the convolutional layers is very slow, as each kernel in each layer must sequentially calculate many inner products for a single forward and backward propagation which equates to O(N^2 n^2) per kernel per layer where the inputs are N x N arrays and the kernels are n x n arrays. Convolution can be efficiently performed as a Hadamard product in the frequency domain. The bottleneck is the transformation which has a cost of O(N^2 log_2 N) using the fast Fourier transform (FFT). However, …

Feature Extraction Using Dimensionality Reduction Techniques: Capturing The Human Perspective, Ashley B. Coleman

Browse all Theses and Dissertations

The purpose of this paper is to determine if any of the four commonly used dimensionality reduction techniques are reliable at extracting the same features that humans perceive as distinguishable features. The four dimensionality reduction techniques that were used in this experiment were Principal Component Analysis (PCA), Multi-Dimensional Scaling (MDS), Isomap and Kernel Principal Component Analysis (KPCA). These four techniques were applied to a dataset of images that consist of five infrared military vehicles. Out of the four techniques three out of the five resulting dimensions of PCA matched a human feature. One out of five dimensions of MDS matched …

A Workload Balanced Mapreduce Framework On Gpu Platforms, Yue Zhang

Browse all Theses and Dissertations

The MapReduce framework is a programming model proposed by Google to process large datasets. It is an efficient framework that can be used in many areas, such as social network, scientific research, electronic business, etc. Hence, more and more MapReduce frameworks are implemented on different platforms, including Phoenix (based on multicore CPU), MapCG (based on GPU), and StreamMR (based on GPU). However, these MapReduce frameworks have limitations, and they cannot handle the collision problem in the map phase, and the unbalanced workload problems in the reduce phase. To improve the performance of the MapReduce framework on GPGPUs, in this thesis, …

Whole-Lake Primary Production Calculator, Colin D. Leong

Browse all Theses and Dissertations

This work describes an implementation of a model for estimation of both benthic and phytoplanktonic primary production in lakes. The web application makes use of interpolation techniques to allow estimates of primary production using values for photosynthesis/irradiance parameters at only 5 depths. These estimates compare favorably in accuracy with estimates using values listed at over one hundred depths. Validation of the implementation was done by comparison with primary production results from the Northern Temperate Lakes Long Term Ecological Research database.

Features For Ranking Tweets Based On Credibility And Newsworthiness, Jacob W. Ross

Browse all Theses and Dissertations

We create a robust and general feature set for learning to rank algorithms that rank tweets based on credibility and newsworthiness. In previous works, it has been demonstrated that when the training and testing data are from two distinct time periods, the ranker performs poorly. We improve upon previous work by creating a feature set that does not over fit a particular year or set of topics. This is critical given how people utilize social media changes as time progresses, and the topics discussed vary. In addition, we are constantly gaining new tweet data. Thus, it is important to be …

Domain-Specific Document Retrieval Framework For Near Real-Time Social Health Data, Swapnil Soni

Browse all Theses and Dissertations

With the advent of web search and microblogging, the percentage of Online Health Information Seekers (OHIS) using these services to share and seek health information in real-time has increased exponentially. Recently, Twitter has emerged as one of the primary mediums for sharing and seeking of the latest information related to a variety of topics, including health information. Although Twitter is an excellent information source, the identification of useful information from the deluge of tweets is one of the major challenges. Twitter search is limited to keyword-based techniques to retrieve information for a given query and sometimes the results do not …

Automatic Emotion Identification From Text, Wenbo Wang

Browse all Theses and Dissertations

People's emotions can be gleaned from their text using machine learning techniques to build models that exploit large self-labeled emotion data from social media. Further, the self-labeled emotion data can be effectively adapted to train emotion classifiers in different target domains where training data are sparse.

Emotions are both prevalent in and essential to most aspects of our lives. They influence our decision-making, affect our social relationships and shape our daily behavior. With the rapid growth of emotion-rich textual content, such as microblog posts, blog posts, and forum discussions, there is a growing need to develop algorithms and techniques for …

Knowledge Enabled Location Prediction Of Twitter Users, Revathy Krishnamurthy

Browse all Theses and Dissertations

As the popularity of online social networking sites such as Twitter and Facebook continues to rise, the volume of textual content generated on the web is increasing rapidly. The mining of user generated content in social media has proven effective in domains ranging from personalization and recommendation systems to crisis management. These applications stand to be further enhanced by incorporating information about the geo-position of social media users in their analysis. Due to privacy concerns, users are largely reluctant to share their location information. As a consequence of this, researchers have focused on automatic inferencing of location information from the …

A Language For Inconsistency-Tolerant Ontology Mapping, Kunal Sengupta

Browse all Theses and Dissertations

Ontology alignment plays a key role in enabling interoperability among various data sources present in the web. The nature of the world is such, that the same concepts differ in meaning, often so slightly, which makes it difficult to relate these concepts. It is the omni-present heterogeneity that is at the core of the web. The research work presented in this dissertation, is driven by the goal of providing a robust ontology alignment language for the semantic web, as we show that description logics based alignment languages are not suitable for aligning ontologies.

The adoption of the semantic web technologies …

Orthogonal Moment-Based Human Shape Query And Action Recognition From 3d Point Cloud Patches, Huaining Cheng

Browse all Theses and Dissertations

With the recent proliferation of 3D sensors such as Light Detection and Ranging (LIDAR), it is essential to develop feature representation methods that can best characterize the point clouds produced by these devices. When these devices are employed in targeting and surveillance of human actions from both ground and aerial platforms, the corresponding point clouds of body shape often comprise low-resolution, disjoint, and irregular patches of points resulted from self-occlusions and viewing angle variations. The prevailing method of depth image analysis has the limitation of relying on 2D features that are not native representation of 3D spatial relationships. On the …

Ontology Pattern-Based Data Integration, Adila Alfa Krisnadhi

Browse all Theses and Dissertations

Data integration is concerned with providing a unified access to data residing at multiple sources. Such a unified access is realized by having a global schema and a set of mappings between the global schema and the local schemas of each data source, which specify how user queries at the global schema can be translated into queries at the local schemas. Data sources are typically developed and maintained independently, and thus, highly heterogeneous. This causes difficulties in integration because of the lack of interoperability in the aspect of architecture, data format, as well as syntax and semantics of the data. …

Learning To Rank Algorithms And Their Application In Machine Translation, Tian Xia

Browse all Theses and Dissertations

In this thesis, we discuss two issues in the learning to rank area, choosing effective objective loss function, constructing effective regresstion trees in the gradient boosting framework, as well as a third issus, applying learning to rank models into statistcal machine translation. First, list-wise based learning to rank methods either directly optimize performance measures or optimize surrogate functions of performance measures that have smaller gaps between optimized losses and performance measures, thus it is generally believed that they should be able to lead to better performance than point-and pair-wise based learning to rank methods. However, in real-world applications, state-of-the-art practical …

Browser Based Visualization For Parameter Spaces Of Big Data Using Client-Server Model, Kurtis M. Glendenning

Browse all Theses and Dissertations

Visualization is an important task in data analytics, as it allows researchers to view abstract patterns within the data instead of reading through extensive raw data. Allowing the ability to interact with the visualizations is an essential aspect since it provides the ability to intuitively explore data to find meaning and patterns more efficiently. Interactivity, however, becomes progressively more difficult as the size of the dataset increases. This project begins by leveraging existing web-based data visualization technologies and extends their functionality through the use of parallel processing. This methodology utilizes state-of-the-art techniques, such as Node.js, to split the visualization rendering …

Owl Query Answering Using Machine Learning, Todd Huster

Browse all Theses and Dissertations

The formal semantics of the Web Ontology Language (OWL) enables automated reasoning over OWL knowledge bases, which in turn can be used for a variety of purposes including knowledge base development, querying and management. Automated reasoning is usually done by means of deductive (proof-theoretic) algorithms which are either provably sound and complete or employ approximate methods to trade some correctness for improved efficiency. As has been argued elsewhere, however, reasoning methods for the Semantic Web do not necessarily have to be based on deductive methods, and approximate reasoning using statistical or machine-learning approaches may bring improved speed while maintaining high …

Temporally Biased Search Result Snippets, J. Abhiram Tatineni

Browse all Theses and Dissertations

The search engine result snippets are an important source of information for the user to obtain quick insights into the corresponding result documents. When the search terms are too general, like a person's name or a company's name, creating an appropriate snippet that effectively summarizes the document's content can be challenging owing to multiple occurrences of the search term in the top ranked documents, without a simple means to select a subset of sentences containing them to form result snippet. In web pages classified as narratives and news articles, multiple references to explicit, implicit and relative temporal expressions can be …

Design Of A Novel Low - Cost, Portable, 3d Ultrasound System With Extended Imaging Capabilities For Point-Of-Care Applications, Michail Tsakalakis

Browse all Theses and Dissertations

Ultrasound Imaging (USI) or Medical Sonography (MS), as it is formally called, has been widely used in biomedical applications over the last decades. USI can provide clinicians with a thorough view of the internal parts of the human body, making use of sound waves of higher frequencies than humans can perceive. USI systems are considered highly portable and of low-cost, compared to other imaging modalities. However, despite those advantages, Ultrasound Systems (US) and especially 3D ones, have not been yet extensively utilized for Point-of-Care (POC) applications, due to numerous restrictions and artifacts that they currently present.

Hardware complexity and real-time …

Mining Behavior Of Citizen Sensor Communities To Improve Cooperation With Organizational Actors, Hemant Purohit

Browse all Theses and Dissertations

Web 2.0 (social media) provides a natural platform for dynamic emergence of citizen (as) sensor communities, where the citizens generate content for sharing information and engaging in discussions. Such a citizen sensor community (CSC) has stated or implied goals that are helpful in the work of formal organizations, such as an emergency management unit, for prioritizing their response needs. This research addresses questions related to design of a cooperative system of organizations and citizens in CSC. Prior research by social scientists in a limited offline and online environment has provided a foundation for research on cooperative behavior challenges, including 'articulation' …

Distributed Local Trust Propagation Model And Its Cloud-Based Implementation, Dharan Kumar Reddy Althuru

Browse all Theses and Dissertations

World Wide Web has grown rapidly in the last two decades with user generated content and interactions. Trust plays an important role in providing personalized content recommendations and in improving our confidence in various online interactions. We review trust propagation models in the context of social networks, semantic web, and recommender systems. With an objective to make trust propagation models more flexible, we propose several extensions to the trust propagation models that can be implemented as configurable parameters in the system. We implement Local Partial Order Trust (LPOT) model that considers trust as well as distrust ratings and perform evaluation …

Combating Integrity Attacks In Industrial Control Systems, Chad Arnold

Browse all Theses and Dissertations

Industrial Control Systems are vulnerable to integrity attacks because of connectivity to the external Internet and trusted internal networking components that can become compromised. Integrity attacks can be modeled, analyzed, and sometimes remedied by exploiting properties of physical devices and reasoning about the trust worthiness of ICS communication components.

Industrial control systems (ICS) monitor and control the processes of public utility that society depends on - the electric power grid, oil and gas pipelines, transportation, and water facilities. Attacks that impact the operations of these critical assets could have devastating consequences. The complexity and desire to interconnect ICS components have …

A Novel Synergistic Model Fusing Electroencephalography And Functional Magnetic Resonance Imaging For Modeling Brain Activities, Konstantinos Michalopoulos

Browse all Theses and Dissertations

Study of the human brain is an important and very active area of research. Unraveling the way the human brain works would allow us to better understand, predict and prevent brain related diseases that affect a significant part of the population. Studying the brain response to certain input stimuli can help us determine the involved brain areas and understand the mechanisms that characterize behavioral and psychological traits.

In this research work two methods used for the monitoring of brain activities, Electroencephalography (EEG) and functional Magnetic Resonance (fMRI) have been studied for their fusion, in an attempt to bridge together the …

Automated Complexity-Sensitive Image Fusion, Brian Patrick Jackson

Browse all Theses and Dissertations

To construct a complete representation of a scene with environmental obstacles such as fog, smoke, darkness, or textural homogeneity, multisensor video streams captured in diferent modalities are considered. A computational method for automatically fusing multimodal image streams into a highly informative and unified stream is proposed. The method consists of the following steps: 1. Image registration is performed to align video frames in the visible band over time, adapting to the nonplanarity of the scene by automatically subdividing the image domain into regions approximating planar patches

2. Wavelet coefficients are computed for each of the input frames in each modality …

The Properties Of Property Alignment On The Semantic Web, Michelle Andreen Cheatham

Browse all Theses and Dissertations

Ontology alignment is an important step in enabling computers to query and reason across the many linked datasets on the semantic web. This is a difficult challenge because the ontologies underlying different linked datasets can vary in terms of subject area coverage, level of abstraction, ontology modeling philosophy, and even language. The alignment approach presented here centers on string similarity metrics. Nearly all ontology alignment systems use a string similarity metric in one form or another, but it seems that the choice of a particular metric is often arbitrary. We begin this dissertation with the most comprehensive survey to date …

An Evolutionary Approximation To Contrastive Divergence In Convolutional Restricted Boltzmann Machines, Ryan R. Mccoppin

Browse all Theses and Dissertations

Deep learning is an emerging area in machine learning that exploits multi-layered neural networks to extract invariant relationships from large data sets. Deep learning uses layers of non-linear transformations to represent data in abstract and discrete forms. Several different architectures have been developed over the past few years specifically to process images including the Convolutional Restricted Boltzmann Machine. The Boltzmann Machine is trained using contrastive divergence, a depth-first gradient based training algorithm. Gradient based training methods have no guarantee of reaching an optimal solution and tend to search a limited region of the solution space. In this thesis, we present …

What Machines Understand About Personality Words After Reading The News, Eric David Moyer

Browse all Theses and Dissertations

Vector-based lexical semantics is a powerful technique that still has many undiscovered applications. In this thesis I apply a vector-space lexical-semantic model newly developed by Mikolov et. al. trained on skip-grams to the lexical hypothesis in personality psychology. The method produces interpretable dimensions that are consistent across several sets of descriptive personality words. The dimensions include ones for conflict and positive and negative evaluation. However they are more descriptive of word usage semantics than of the characteristics of the thing described and thus do not include a recognizable component of the 5 factor model in their first 14 dimensions. They …

Mining Privacy Settings To Find Optimal Privacy-Utility Tradeoffs For Social Network Services, Shumin Guo

Browse all Theses and Dissertations

Privacy has been a big concern for users of social network services (SNS). On recent criticism about privacy protection, most SNS now provide fine privacy controls, allowing users to set visibility levels for almost every profile item. However, this also creates a number of difficulties for users. First, SNS providers often set most items by default to the highest visibility to improve the utility of social network, which may conflict with users' intention. It is often formidable for a user to fine-tune tens of privacy settings towards the user desired settings. Second, tuning privacy settings involves an intricate tradeoff between …

Automatic Identification Of Interestingness In Biomedical Literature, Gaurish Anand

Browse all Theses and Dissertations

This thesis presents research on automatically identifying interestingness in a graph of semantic predications. Interestingness represents a subjective quality of information that represents its value in meeting a user's known or unknown retrieval needs. The perception of information as interesting requires a level of utility for the user as well as a balance between significant novelty and sufficient familiarity. It can also be influenced by additional factors such as unexpectedness or serendipity with recent experiences. The ability to identify interesting information facilitates the development of user-centered retrieval, especially in information semantic summarization and iterative, step-wise searching such as in discovery …

Mining And Understanding Regret Tweets, Lu Zhou

Browse all Theses and Dissertations

Inappropriate tweets may cause severe damages on the authors' reputation or privacy. However, many users do not realize until publishing them after a while. Once published, such tweets have lasting effects that may not be completely eliminated by simple deletion, because other users may have read them or third-party tweet analysis platforms have cached them. In this paper, we study the problem of identifying regret tweets for normal individual users, with the ultimate goal to reduce the occurrences of regret tweets. We develop a machine learning approach to extract a large collection of regret tweets from noisy deleted tweets. We …

Amyna: A Security Generator And Performance Estimator Framework Against Memory-Based Attacks, Anna Trikalinou

Browse all Theses and Dissertations

As we become more and more dependent on computer networks, using them for everything from banking and investing to shopping and communicating, computer security has emerged to be an increasingly important concern, due to undesirable cyber-security attacks. Thus, in response to this issue many efforts have been made towards accurate and robust computer security protection; however, the general problem is very challenging, diverse and ever-changing and remains still open. In this PhD dissertation we offer protection on one of these cyber-security attacks, the memory-based attacks, caused by one of the most critical software errors according to the MITRE ranking. When …