Search

Biological processes are often far too complex to predict. For those phenomenon that still evade understanding, it is helpful to visualize a black box—a system with clear inputs and outputs, but an unknowable, labyrinthine set of interior mechanisms to produce this transformation. To determine what happened, controlled experiments are often run on a culture of cells or tissues using a multi-well plate format. This structure is relatively simple and can be applied in most laboratory environments. This paper discusses newly built technologies that take this definition of an experiment and adapt it to detect genotypes as well as phenotypes; promisingly, these technologies are highly multiplexible by using a modified version of Fluoresecnt in situ Sequencing (FISSEQ).

Interactive learning tools allow students to explore STEM concepts deeply, improving educational outcomes. Interactive tools that use cloud computing resources can (1) explore computationally intensive concepts and (2) seamlessly link concepts to graphical representations. When these goals conflict, good design, and implementation principles are needed to (1) preserve interactive response times and (2) uphold curriculum goals. For this paper, we present an interactive cloud-based learning tool that allows students to interactively explore algebraic formulas, their corresponding graphs (including axes) and generating data. Our tool prioritizes the following design principles: (1) co-design curriculum and interactive systems support, (2) efficiently collect and share student activities between students, teachers and researchers, and (3) integrate with existing classroom management platforms. Our implementation employs a careful division of work between client-side browsers and cloud servers to provide response times that are within human perception limits. We achieve these design goals by (1) client-side programming for interactive components, (2) sending student activity data to servers at different rates for different sharing types, and (3) align student's data with Moodle's (a popular open source classroom management system) schema. Twenty Math teachers were trained to teach Algebra to students using our interactive service. Over eighty percent of teachers trained with our system adopted it. We provide students response times (1) with interactive graphing less than 10 milliseconds (2) tail latency of 99% requests below 0.1 seconds. More broadly, our system has received positive feedback as we have scaled it to over 5 schools, 15 classrooms, and 700 students.

Students who can link algebraic functions to their corresponding graphs perform well in STEM courses. Increasingly, early algebra curricula teaches these concepts in tandem. However, it is challenging to assess whether students are linking the concepts. Tests, video analyses, interviews and other traditional methods that aim to quantify how students link the concepts taught in school require precious classroom and teacher time. In this paper, we use web logs collected by a smart classroom web server to infer learning. Compared to traditional methods, web logs are widely available, voluminous and amenable to data science. However, web logs are constrained by factors such as data collection practices by system managers and the design of the web interface. Our approach partitions the web interface into components related to data and graph concepts. We collect click and mouse movement data as users interact with these components. We used statistical and data mining techniques like k means clustering, principal component analysis and frequent sequence patterns to model their learning behavior. We compared our models with traditional methods to assess learning behavior for a workshop presented in Summer 2016. Students in the workshop were middle-school math teachers planning to use a new early algebra curriculum in their own classrooms. First, we used our models to assess participation levels, a prerequisite indicator for learning. Our models aligned with ground-truth traditional methods for 17 of 18 students. The results from machine learning models which do not consider the computational thinking aspect of the web components aligned with ground-truth methods in only 10 of 18 students. Unlike traditional methods, our models were computed in minutes and not days. The results of the models w.r.t the two types of components of the web portal have been used to infer possible data or graph oriented cognitive bias in the students.

Computational creativity researchers interested in applying machine learning to computer composition often use the music of J.S. Bach to train their systems. Working with Bach, though, requires grappling with the conventions of tonal music, which can be difficult for computer systems to learn. In this paper, we propose and implement an alternate approach to composition and harmonization of chorales based on pitch-relative note encodings to avoid tonality altogether. We then evaluate our approach using a survey and expert analysis, and find that pitch-relative encodings do not significantly affect human-comparability, likability or creativity. However, an extension of this model that better addresses the criteria survey participants used to evaluate the music, such as instrument timbre and harmonic dissonance, still shows promise.

This paper uses player tracking data from a subset of games from the 2015-16 NBA season to estimate the relationship between energy expenditure and player performance. Using locational tracking data, several measures of energy expenditure are calculated, along with one baseline measure. To account for each individual player's skill level, a ridge regression is performed and the residuals are used as the dependent variable in a second regression. Binning is used to estimate the relationship between measures of energy expenditure and player performance, and the coefficients from binning are subsequently used to estimate parametric transformations of energy expenditure that are included in the final model. Finally, generalized additive models (GAMs) are used to estimate the relationship between energy expenditure and performance for individual players. I find three notable results. First, energy expenditure is a significant predictor of a player's deviation from their average skill level. Second, player tracking data can be used to generate measures of energy expenditure that allow for significantly more accurate predictions of player performance than measures not derived from player tracking data. Third, there is strong evidence that players require a period of time when they first enter the game to warm up, during which they perform at a lower level than normal.

This paper examines prediction of financial distress using machine learning techniques. It uses accounting and equity price variables found to be associated with financial distress in Campbell et al. (2008) as input variables to predict output variables of corporate bankruptcy and issuers’ credit rating. By training logistic regression based on the training method in machine learning, I find that the model offers a high ratio of true positive rate to false positive rate in identifying bankrupt firms, which implies its effectiveness to serve as a screener against distressed se- curities. For credit rating prediction, I find statistical evidences that the trained machine learning models provide improved accuracy over the base rate model of weighted random predictions. Lastly, while these models provide relatively low accuracy for predicting the exactly correct rating, its accuracy is much higher for predicting the correct direction of rating changes, which offers a potential for im- proved trading strategy.

High-resolution digital images are known as efficient carriers of information. Billions of them are distorted while being captured, stored, and shared. Thus, Image Quality Assessment (IQA) techniques are utilized to measure the quality of the source image in a way to match with subjective quality measured by human evaluation. Due to the lack of both reference image information and distortion type of the test image, Non-Distortion-Specific No-Reference (NDS NR) IQA focuses on improving assessment performance more intelligently. The idea of automatically obtaining features and weighting them without using reference images leads us to propose an innovative Deep Convolutional-Neural-Network (Deep CNN) architecture consisting of 17 layers to measure the quality of the NDS NR images. Our proposed Deep CNN employs both low-level and high-level features and provides more accurate performance, as well as noticeable improvement in computational time. The proposed network attains the best performance on Spearman Rank Order Correlation Coefficient (SRCC) evaluations among all popular NR-IQA methods listed in the paper. Furthermore, it achieves the highest Linear Correlation Coefficient (LCC) among all well-known No-Reference (NR) and Full-Reference (FR) IQA methods.

In an earlier paper, a method called One-class Classification using Length statistics of (jumping) Emerging Patterns (OCLEP) was introduced for masquerader detection.Jumping emerging patterns (JEPs) for a test instance are minimal patterns that match the test instance but they do not match any normal instances.OCLEP was based on the observation that one needs long JEPs to differentiate an instance of one class from instances of the same class, but needs short JEPs to differentiate an instance of one class from instances of a different class. In this thesis, we present OCLEP+, One-class Classification using Length statistics of Emerging Patterns Plus by adding several new ideas to OCELP. OCLEP+ retains the one-class training feature of OCELP, hence it only requires the normal class data for training. Moreover, OCELP+ has the advantage of being not model or signature based, making it hard to evade. OCLEP+ uses only length statistics of JEPs, making it a robust method. Experiments show that OCELP+ is more accurate than OCLEP and one-class SVM, on the NSL-KDD datasets.

Despite its relative youth, computer science has become a well-established discipline, granting over 2% of the bachelors degrees in the United States (U.S. Department of Education, 2010). For this reason, it is important that we understand the nature of computer science and the likely direction for the development of inquiry in computer science in the future. This paper examines several perspectives on the nature of the methods of computer science inquiry. These are empiricist methods, rationalist methods, and an engineering stance. It argues that empiricist and rationalist stances play identifiable roles in the scientific nature of computer science reasoning but that the engineering stance does not. Following the trend in the maturation of other sciences, this paper recommends an overhaul in computer science curricula.

What does it mean to have random numbers? Without understanding where a group of numbers came from, it is impossible to know if they were randomly generated. However, common sense claims that if the process to generate these numbers is truly understood, then the numbers could not be random. Methods that are able to let their internal workings be known without sacrificing random results are what this paper sets out to describe. Beginning with a study of what it really means for something to be random, this paper dives into the topic of random number generators and summarizes the key areas. It covers the two main groups of generators, true-random and pseudo-random, and gives practical examples of both. To make the information more applicable, real life examples of currently used and currently available generators are provided as well. Knowing the how and why of a number sequence without knowing the values that will come is possible, and this thesis explains how it is accomplished.

In the field of unmanned aerial vehicles (UAVs), several control processes must be ac- tive to maintain safe, autonomous flight. When flying multiple UAVs simultaneously, these aircraft must be capable of performing mission tasks while maintaining a safe distance from each other and obstacles in the air. Despite numerous proposed collision avoidance algo- rithms, there is little research comparing these algorithms in a single environment. This paper outlines a system built on the Robot Operating System (ROS) environment that allows for control of autonomous aircraft from a base station. This base station allows a re- searcher to test different collision avoidance algorithms in both the real world and simulated environments. Data is then gathered from three prominent collision avoidance algorithms based on safety and efficiency metrics. These simulations use different configurations based on airspace size and number of UAVs present at the start of the test. The three algorithms tested in this paper are based on mixed integer linear programming (MILP), the A* al- gorithm, and artificial potential fields. The results show that MILP excelled with a small number of aircraft on the field, but has computation issues with a large number of aircraft. The A* algorithm struggled with small field sizes but performed very well with a larger airspace. Artificial potential fields maintained strong performance across all categories be- cause of the algorithm’s handling of many special cases. While no algorithms were perfect, these algorithms demonstrated the ability to handle up to eight aircraft on a 500 meter square field and sixteen aircraft safely on a 1000 meter square field.

Network simulations are heavily used in the networking community to evaluate the performance of computer networks and their protocols. Simulations are often chosen over alternatives such as live experiments due to limited resources in terms of scalability, as well as reproduceability of the experiments. Many of the routing protocols designed for Mobile Ad Hoc Networks (MANETs) are evaluated solely on their performance calculated by these simulations, but the simulation environments the routing protocols are exposed to are often limited in scope. Only certain aspects of the routing protocols are tested, so the protocols are only understood in terms of the fabricated scenarios that they are subjected to.We first investigate the current best practices in simulation-based multi-hop wireless ad-hoc network (MANET) protocol evaluation to examine how wide-spread this problem is in the networking community. We extend a prior characterization of the settings and parameters used in MANET simulations by studying the papers published in one of the premier mobile networking conferences between 2006 and 2010. We find that there are still several configuration pitfalls which many papers fall victim to, which in turn damages the integrity of the results as well as any research aimed at reproducing and extending these results. We then describe the simulation “design space” of MANET routing in terms of its basic dimensions and corresponding parameters. We then discuss the benchmark infrastructure that was created to provide an easy to use solution for testing these protocols in a wide range of scenarios. The following chapter looks extensively at the realistic scenarios provided with the benchmark that act as sample scenarios to promote modeling simulations after real-world situations, and to show the flexibility in adding new scenarios. We also propose four “auxiliary” metrics to increase simulation integrity. Next, we show results generated by the benchmarking tool and provide our concluding thoughts.

The work presented in this thesis was performed under the EU's Framework 7 (FP7) project, 'REAL3D'. The aim of this project is to develop methods based on digital holography for real time capture and display of 3D objects. This thesis forms a small subset of all the work done in this project. Much of the research work was aimed towards fullling our part of the requirements of the REAL3D project. The central theme of the research presented in this thesis is that of improving the performance of the digital holographic imaging system for its use in 3D display. This encompasses research into speed up of reconstruction algorithms, understanding the in uence of noise and developing techniques to increase resolution and angular perspective range in reconstructions. The main original contributions of this research work presented in this thesis are: A computer-interfaced automatic digital holographic imaging system employing `phase shifting' has been built. This system is capable of recording high-quality digital holograms of a real world 3D object. The object can be rotated on a rotational stage and a full 360 range of perspectives can be recorded. Speckle reduction using moving diusers can be performed to improve the image quality of the reconstructed images. A LabView based user friendly interface has been developed. Novel methods based on space-time tradeo and xed point arithmetic have been developed and implemented for speed- ing up the reconstruction algorithm used in digital holography. This has resulted in the publication of one peer-reviewed journal pub- lication and one conference proceeding [1, 2]. The in uence of additive noise, particularly quantization noise in digital holography has been studied in detail. A model has been developed to understand the in uence of noise on the re- constructed image quality. Based on this model, a method has been developed to suppress quantization noise in a memory ecient man- ner. This work led to the publication of two peer-reviewed journal publications [3, 4]. A novel method of removing the twin image has been devel- oped. Methods to increase the perspectives in holography based on synthetic aperture have been implemented. Apart from these primary contributions, the author of this thesis has also contributed in the form of assisting in experiments, creating gures for various papers, writing computer programs and discussions during group meetings. In total, 6 peer-reviewed journal papers (3 being primary author) have been published and 6 conference proceedings (3 being primary author) have been published. Additionally, 2 talks have been given at international conferences.

In the field of augmented furniture, many pieces of furniture use technology to comment on human interaction. One piece, Nikolovska’ Conversation Table, uses microphone and LEDs to comment on dominance in conversation. This paper develops this notion further by incorporating psychological research to create a furniture set that can read subtleties in dyadic interaction. This Coordination Table is a table and two chairs that use sensors to read the posture of each person. A connected laptop analyzes this data in realtime to calculate how similar the users¿¿¿¿ posture is. According to psychological research, people with similar posture, especially over time, tend to have higher rapport. The Coordination Table uses this finding to estimate the rapport currently experienced by the dyad, based on its data.This paper outlines the software and technology used to construct a prototype Coordination Table. The table was also informally tested to prove that it could detect differences in posture due to higher or lower rapport. Although it has not been evaluated to work exactly as expected, it was shown to detect differences between users engaging in different activities. The author also discusses potential applications for the Coordination Table technology, including its usefulness for psychological research, its potential utility in performance art, and its potential feasibility as a commercial product.

This paper will discuss the development of a music classification system as a component in a music recommendation system. The front-end portion of the system is an Android Media Player application named SmartPlayer. The player is an intelligent mobile recommendation system as well as a media player that is capable to playing high quality videos. In this paper the specifics of the underlying system and the front-end components will be discussed in detail. Other methods and future aspirations will also be discussed. The system performs automatic genre classification with one feature-vector per audio file. The backbone of the system is utilizing the existing SmartPlayer SQL database containing some 100,000 YouTube music videos in mp4 format. The system utilizes multiple components of the Marsyas (Music Analysis, Retrieval and Synthesis for Audio Signals) open source project, as well as STANN (Simple Thread-safe Approximate Nearest Neighbors). A different method that will be introduced is the K-nearest neighbor method to cluster and compare existing videos in the database. The dynamic time warping method will be used to compare different time series data derived from MFCCs. This method is generally used in comparing two time series data plots but does so with respect to relative aspects of the data.

Passwords are still one of the most common means of securing computer systems. Most organizations rely on password authentication systems, and therefore, it is very important for them to enforce their users to have strong passwords. They usually try to enforce security by mandating users to follow password creation policies. They force users to follow some rules such as a minimum length, or using symbols and numbers. However, these policies are not consistent with each other; for example, the length of a good password is different in each policy. They usually ignore the importance of usability of the password for the users. The more complex they are the more they frustrate users and they end up with some coping strategies such as adding "123" at the end of their passwords or repeating a word to make their passwords longer, which reduces the security of the password, and more importantly there is no scientific basis for these password creation policies to make sure that passwords that are created based on these rules are resistance against real attacks. In fact, there are studies that show that even the NIST proposal for a password creation policy that results in strong passwords is not valid. This paper describes different password creation policies and password checkers that try to help users create strong passwords and addresses their issues. Metrics for password strength are explored in this paper and new approaches to calculate these metrics for password distributions are introduced. Furthermore, a new technique to estimate password strength based on its likelihood of being cracked by an attacker is described. In addition, a tool called PAM has been developed and explained in details in this paper to help users have strong passwords using these metrics. PAM is a password analyzer and modifier, which rejects weak passwords and suggests a new stronger password with slight changes to the original one to ensure the usability of the password for each individual.

Clustering is one of the most basic mental activities used by humans to handle the huge amount of information they receive every day. As such, clustering has been extensively studied in different disciplines including: statistics, pattern recognition, machine learning and data mining. Nevertheless, the body of knowledge concerning clustering has focused on objects represented as feature vectors stored in a single dataset. Clustering in this setting aims at grouping objects of a single type in a single table into clusters using the feature vectors. On the other hand, modern real-world applications are composed of multiple, large interrelated datasets comprising distinct attribute sets and containing objects from many domains; typically such data is stored in an information network. The types of patterns and knowledge desired in these applications goes far beyond grouping similar homogeneous objects, but rather involves unveiling dependency structures in the data in addition to pinpointing hidden associations across objects in multiple datasets and domains. For example consider an information network that contains the domains of authors, papers and conferences. Two authors a1 and a2 may work in the same research field but never publish in the same conference. Hence clustering only the domains of authors and conference would fail to place a1 and a2 in the same cluster; however considering the entire information network would reveal a hidden link via the papers domain, placing a1 and a2 in the same cluster. This form of relational clustering is essential for knowledge discovery in several applications such as: bioinformatics, social networking, and recommender systems.Information-network clustering advances knowledge discovery in two manners. First, hidden associations amongst objects from differing domains are unveiled, leading to a better understanding of the hidden structure of the entire network. Second, local clusters of the objects within a domain are sharpened and put into greater context, leading to more accurate local clustering. In order to extract knowledge of this form, information network clustering algorithms must consider 1) overlapping clusters and 2) a clustering structure that relates the clusters. In this dissertation we develop a framework and several algorithms for information network clustering by leveraging the above two fundamental aspects that facilitate knowledge discovery.Current state-of-the-art information network clustering algorithms have had relative success in addressing the computational challenge of high dimensional data; however, the majority of these approaches have not addressed the fundamental aspects of overlapping clusters and clustering structure. In this dissertation, we address the information-network clustering problem from a fresh perspective and introduce a novel framework based on Formal Concept Analysis (FCA). Based on mathematicalorder theory, in particular, the theory of complete lattices, FCA provides a rich theoretical basis for investigating and structuring overlapping relational clustering in a single dataset. Shortcomings of previous methods were overcome by extending FCA to information networks, yielding effective and efficient information network clustering algorithms. Several empirical evaluations performed on a large variety of real-world information networks reveal that the FCA-based algorithms work more effectively and efficiently than the current state of the art.

This study was a combination and continuation of previous studies done by BGSU graduate students and faculty. It explored the use of interactive 3D stereo graphics as a component in the delivery of instructions for a construction task and compared this to a non-stereo version as well as a paper-based approach. The principal goal was to determine the relative effectiveness of the three approaches as measured by time to completion and accuracy. A secondary goal was to determine the role of inherent object complexity on performance as measured by time to completion. In terms of accuracy, the results show no signicant differences between any of the presentations. However, they also show that subjects in the stereo condition were the slowest, with paper being slightly faster and non-stereo being the best overall. Further, an analysis of time (in seconds) per rotation reveals that the subjects in the stereo condition rotated the models less frequently than subjects in the non-stereo condition. In terms of inherent object complexity, larger amounts of space between parts of an object increases construction time.

Culturally Situated Design Tools (CSDTs) are web-based software applications that allow students to create simulations of cultural arts: Native American beadwork, African American cornrow hairstyles, urban graffiti, and so forth; using these underlying mathematical principles. CSDTs are the rationale of creating a set of culturally designed games utilizes gaming as a teaching tool to attract and instruct students with familiar methods and environments. This paper will review the development and evaluation of CSDTs, and discuss how various activities attempt to navigate through the potential dangers and rewards of this potent hybrid of information technology (CSDTs), traditional culture and individual creativity. Increasing accessibility to Computer Science and Technology is essential for a discipline that relies on creativity and diverse perspectives. Creating equity in Computer Science learning opportunities is also a social justice issue. With the educational research community having begun to explore the causes behind the underrepresentation of females and students of color in computing courses, outreach efforts have commenced to overcome these enrollment discrepancies. The purpose of this paper and project is to address these issues. The project we focused on out of the CSDT Series is the Break Dancer tool, located on the CSDT website. The Break Dancer tool is a software-simulated game that teaches 3-dimensional space or solid geometry. Real-world objects exist in 3-dimensions. In this project, the Break Dancer tool integrates the youth subculture, promotes physical activity, and makes it into an educational math game.

Model Driven Software Development is an attempt to build software systems by combining the two disciplines of software programming and modelling. But due to the gap in technology, there are a number of problems in using these models to generate the final source code implementation. This has meant that Model Driven Software Development has become a future goal rather than a current reality. The Java Model Parser and Printer project tries to bridge this gap by turning Java into a model. This means that Java can now be used and implemented like any of the other models within the Model Driven Software Development process. This paper will take a look at the effectiveness of the Java Model Parser and Printer project by performing a number of System tests on it. This will help indicate now well the project can transform Java input files to java models. Coverage readings will be taken on these tests to look at how much of the project‟s grammar and abstract syntax tree are being used during the process of transforming Java source code into Java models. A look will then be taken at the coverage results to see if a correlation exists between the grammar and the abstract syntax tree results. Finally conclusions and future work will be discussed to close the paper.

At its heart, a password cracking attack is just a guessing attack. An attacker makes guesses about a user's password until they guess correctly or they give up. While the defender may limit the number of guesses an attacker is allowed, a password's strength often depends on how hard it is for an attacker to model and reproduce the way a user created their password. If humans were effective at practicing unique habits, or generating and remembering random values, cracking passwords would be a near impossible task. In reality, that isn't true. A vast majority of people still follow common patterns, from capitalizing the first letter of their password to putting numbers at the end. What is changing though are the protective techniques being employed that are independent of user behavior. Practices such as salting password hashes negate the ability to pre-compute attacks. Likewise, password hashes are becoming more computationally complex, raising the costs for each guess an attacker makes. While before an attacker could rely on simple brute force methods and ad-hoc models, there is a growing demand for more effective ways to predict what a user's password will be. The need for this is especially strong in the law enforcement community, where tough encryption is encountered regularly. It is also important for the defender to be able to accurately model the security that user generated passwords provide. This paper details several new ways that probability information can be applied to maximize the success of password cracking attacks. From evaluating the effectiveness of known probabilistic techniques such as Markov models, to designing novel techniques such as using probabilistic context free grammars to create password guesses, there are many different ways probability information can be incorporated into modeling user behavior. Furthermore, the techniques described in this paper have been developed using real life passwords and have been tested in actual controlled password cracking attacks. This focus on training and testing against large sets of real life passwords is fairly unique, and only possible due to the increasing availability of disclosed password lists. In addition to allowing the development of more effective attacks, knowledge of how people select passwords can then be applied to evaluating the effectiveness of password creation policies. For example, how much stronger is an eight character password compared to a seven character password? In short, a better understanding of how users create passwords can benefit both the attacker and the defender.

A biofilm is a colony of micro-organisms attached to a surface that is held together by acomplex matrix of proteins, polysaccharides, and nucleic acid. There has been much interestin biofilms because they display a stronger resistance to antimicrobials than their planktoniccounterparts. In this paper, a model will be set up for the growth of a biofilm throughnutrient, and the decay of the biofilm through various applications of antimicrobial. Thesoluble components of this model are nutrient and antimicrobial, and will be representedin reaction-diffusion equations. The particulate components will be represented in partialdifferential equations, as will the matrix that holds the biofilm together, known as the extra-cellular polymeric substance. This system will be reduced to ordinary differential equationsthrough asymptotic limits and then be solved numerically, using the MATLAB scheme. An-timicrobial will be delivered by two different methods. The first method will be to apply itas a liquid through an inhalation device, a nebulizer, which delivers the antimicrobial to thesurface of the biofilm. The second method of application will be in the form of nanosphereswhich will be imbedded within the biofilm and as they degrade, will release antimicrobialwithin the biofilm. We will compare results of each treatment separately, then with theresults of a combined delivery of both. This paper will show that the combined delivery ofantimicrobial by nebulizer and nanospheres is the optimal method of treatment.

This paper examines the concept of unauthorized access crimes statutes in the fifty states and the federal system in order to determine the concept of what exactly the crime of unauthorized access constitutes. This paper will also determine whether these same crimes could be prosecuted under preexisting statutes that were already in effect in the United States criminal codes prior to the invention of the computer. Through a review of previous literature on the subject of computer crime laws, and more specifically unauthorized access laws, additional insight as to the purpose and intent of these laws will also be drawn. Case law on unauthorized access and any related criminal activity will also be examined. This information will help to determine whether the intentions of the legislatures have been to criminalize what is commonly referred to as hacking.

Traditionally, a chain of evidence or chain of custody refers to the chronological documentation, or paper trail, showing the seizure, custody, control, transfer, analysis, and disposition of evidence, physical or electronic. Whether in the criminal justice system, military applications, or natural disasters, ensuring the accuracy and integrity of such chains is of paramount importance. Intentional or unintentional alteration, tampering, or fabrication of digital evidence can lead to undesirable effects. We find despite the consequences at stake, historically, no unique protocol or standardized procedure exists for establishing such chains. Current practices rely on traditional paper trails and handwritten signatures as the foundation of chains of evidence.;Copying, fabricating or deleting electronic data is easier than ever and establishing equivalent digital chains of evidence has become both necessary and desirable. We propose to consider a chain of digital evidence as a multi-component validation problem. It ensures the security of access control, confidentiality, integrity, and non-repudiation of origin. Our framework, includes techniques from cryptography, keystroke analysis, digital watermarking, and hardware source identification. The work offers contributions to many of the fields used in the formation of the framework. Related to biometric watermarking, we provide a means for watermarking iris images without significantly impacting biometric performance. Specific to hardware fingerprinting, we establish the ability to verify the source of an image captured by biometric sensing devices such as fingerprint sensors and iris cameras. Related to keystroke dynamics, we establish that user stimulus familiarity is a driver of classification performance. Finally, example applications of the framework are demonstrated with data collected in crime scene investigations, people screening activities at port of entries, naval maritime interdiction operations, and mass fatality incident disaster responses.

This research is about information technology (IT) and merger and acquisition(M&A). In recent years, M&A has been a common tool and a fast way for companies to grow and do business. Thus, M&A activity still continues to flourish despite the current economic downturn. M&A is a business transaction and IT has become an increasingly important part of the business. Studies show that while many M&A have failed for different reasons, the major reason is that the Chief Executive Officer (CEO), the Chief Information Officer (CIO), and the key handler of the M&A project did not bring the IT department on board early in the project, due to lack of understanding of the critical role that IT plays in M&A deals. In order to successfully execute M&A, it is essential to have a well-built IT department as a model for M&A with well thought out governance and policies in order to set a good foundation for a successful M&A project. This paper offers some viewpoints, approaches, and methods to help ensure the success of the M&A project; this paper also illustrates some real world successful examples which show the M&A project's ability to perform the synergy capture which drives most M&A.

In this paper, we explore the concept of epistemic visualization in interrogative domains. Epistemic visualization is the process and result of developing visual models that capture the structure, content, justification and acquisition of knowledge obtained by a software agent in a knowledge-based system. The knowledge is the foundation in which the agent can respond to queries against a corpus containing questions and answers. The visualizations are therefore used to examine the quality of the software agent's knowledge. The visual models will include justification and commitment artifacts as well as knowledge acquisition flow. The visualization will demarcate the a priori and posteriori knowledge. The knowledge of the software agent is stored in epistemic structures which are knowledge representation schemes that supports the basic concepts of knowledge as defined by the tripartite analysis of knowledge. Epistemic visualization is used to analyze the quality of the knowledge of a software agent in an interrogative domain. For our purpose, interrogative domains are hearings, trials, interrogations, personality test or any document source in which the primary content is questions and answers pairs. In this paper, we introduce the Epistemic Structure Es that captures the agent's knowledge and the visualization of that epistemic structure using common visualization techniques.

At Ctest Laboratories we are exploring the notion of automated conversion of the semi-structured text to an epistemic structure suitable for deductive inference. In this paper we will develop an epistemic structured representation for electronic transcripts ofinterrogative domains. We propose that knowledge which is typically not visible to keyword search or string matching, can be readily extracted from the an electronic transcript when it is given an appropriate epistemic structure. We introduce an Epistemic Structure Es and a process for converting a semi-structured transcript from and interrogative domain to Es. In this paper we restrict our discussion and analysis to transcripts that have been stored as semi-structured text. In particular we are interested in any knowledge that can be deduced by an interrogative agent from the content of an electronic transcript. Further we develop the notion of an interrogative agent that relies on epistemic justification as a condition for knowledge.

A graph is chordal if each of its cycles of four or more nodes has a chord, which is an edge joining two nodes that are not adjacent in the cycle. Chordal graphs are also called triangulated graphs. A graph is chordal if and only if it has a perfect elimination ordering. A perfect elimination ordering in a graph is an ordering of the vertices of the graph such that, for each vertex v, v and neighbors of v that occur later than v in the order form a clique. In their paper Rose, Tarjan and Lueker have shown that a perfect elimination ordering of a chordal graph can be found efficiently using an algorithm lexicographic breadth first search. This algorithm maintains a partition of the vertices of the graph into a sequence of sets; initially this sequence consists of a single set with all vertices. The algorithm repeatedly chooses a vertex v from the earliest set in the sequence that contains previously not chosen vertices, and splits each set S of the sequence into two smaller subsets, the first consisting of the neighbors of v in S and the second consisting of the non-neighbors. When this splitting process is done for all the vertices, the sequence of the sets has one vertex per set, and is the reverse of a perfect elimination ordering. In his paper Fanica Gavril finds a maximum clique and a minimum coloring of a chordal graph using this perfect elimination ordering.In this paper, we will give an algorithm for finding a maximum clique of a chordal graph by removing vertices of minimum degree. We will also show how to color a chordal graph with minimum number of colors. We give a linear time algorithm to find a maximum clique of a chordal graph and also will show that the size of a maximum clique of a chordal graph equals the chromatic number of the graph, i.e., we can color the whole graph with the number of colors equal to the size of a maximum clique.

Preparation of academic papers involves not only the creative processes but also the more mechanical tasks such as adjusting the form and style to suit the demands of the publishing journal or conference. Among several packages that help in these rather tedious mechanical tasks, the TEX + LATEX + BibTEX combination is extremely popular. This thesis is about tools that help in the necessary task of citing related work accurately. It focuses on three aspects of this larger bibliography frame work: (i) a survey of existing bibliography formats and tools, (ii) a database view of BibTEX files and functionality that ensues, and (iii) processing references given as free style pieces of text. Numerous tools that ease the citation task have been developed in the last five years. The thesis reviews thoroughly the 65 open source, and freeware tools, and somewhat less thoroughly the 18 commercial tools because of limitations of trial ware. These tools range from small stand-alone utilities of a couple of thousand lines of code to large suites of tools that evolved out of the research work of teams over a few years. Their functionality includes the collection of references and searching the various on-line bibliographies for full details and prepare them for inclusion in the references section typically found at the end of papers. We identify a few voids in functionality, especially dealing with free style references, and contribute new tools. The second focus of the thesis is on the maintenance of bibliographies by individuals. In this context, we contribute several new tools: (i) LoadBibTeX stores bibliographic entries as a MySQL database of BibTEX fields as tables as opposed to storing them as plain text .bib files. (ii) BibSearch allows authors to search the database of BibTEX entries based on multiple keywords that can be matched in multiple fields and the resulting output may be saved as a standard .bib file. (iii) Normalization is a feature incorporated into the above tools to bring about normalization of equivalent BibTEX entries. (iv) Duplicate discovery as a feature of LoadBibTeX detects duplicates in a bibliography database in a reliable way. The third focus of the thesis is on the extraction and conversion of references from free style plain text into bibliographic entries expressed in the formal syntax of BibTEX. Often an author collects references as a file of copied-and-pasted pieces of text. We developed a tool that converts such clippings in free style text to bibliographic entries in BibTEX format. Being free style, author names, titles of papers, names of journals and conferences, page numbers, etc. may not appear in a guaranteed order. Recognition of these fields is driven by heuristics. Our tool provides feedback to the authors with (i) a confidence number indicating the correctness of the recognition of a field, and (ii) a colorized HTML version of the input free style text indicating the results of the translation. An extension of this tool extracts the references section of papers published as PDF and translates them into BibTEX entries. We developed an API as a Java package to allow other developers to incorporate the free style to BibTEX conversion functionality into their applications. As an example, we integrate into Aigaion, a highly effective web-based bibliographic tool, both translating free style references, and extracting references from PDF files.

This paper examines the impact of a multicultural approach on the usability of web and software interface designs. Through the use of an eye-tracking system, the study compares the ability of American users to navigate traditional American and Japanese websites. The ASL R6 eye-tracking system recorded user search latency and the visual scan path in locating specific items on the American and Japanese pages. Experimental results found statistically significant latency values when searching for left- or right-oriented navigation menus. Among the participants, visual observations of scan paths indicated a strong preference for initial movements toward the left. These results demonstrate the importance of manipulating web layouts and navigation menus for American and Japanese users. This paper further discusses the potential strengths resulting from modifications of interface designs to correspond with such cultural search tendencies, and suggestions for further research.

Interaction design patterns provide best-practice solutions to software user interface design problems. Several interaction design pattern collections have been published, but none of them address the details of interaction for software that facilitates musical composition on a personal computer. This paper presents a new interaction design pattern collection for musical applications and explains the pattern creation process, organizational structure, and validation model for the collection. Additionally, this paper surveys general design pattern concepts and user-centered software design practices and illustrates how these two subjects have been merged within the interaction design discipline.

Since the explosion of computer systems and computer networks within the past decade, e-commerce, online banking, and other "internet" oriented applications have risen exponentially. According to Forrester Research Group, online shopping in the US grew 580% from 1998 to 2000 accounting for more than $45 billion in sales [10]. Online Banking Report states there are over 100 million people participating in online banking worldwide, an increase of 80% since 1984. This number is expected to rise to 300 million households by 2012 [3]. These applications rely on secure communications for information passing such as credit card numbers and bank account information. The secure communication is realized through the use of cryptography and security protocols for key exchange, authentication etcetera. These protocols can be attacked, possibly resulting in vital information being compromised. This paper discusses classic methodologies concerning intrusion detection and how they are being applied to security protocols. Three methods are presented for detecting and/or preventing intrusions in security protocols. The first method is a simple method aimed at detecting intrusions from attackers with rudimentary skills. The second method, a modified version of the original model, provides a more formidable defense to the sophisticated attacker. Lastly, this paper discusses the third method, IPSec, and how it provides the best security for detecting intrusions in security protocols. Each method is tested with known attacks and the results are discussed.

The use of the tablet and pen to input the Chinese Characters into computer is no longer a dream. During the last decade, different researchers have tried to find the solution to this problem. Although some results have been achieved, the use of tablet and pen is still not wide spread. In the introduction, I explain the Chinese character elements and their structural composition. Through different methods applied in Chapter 2 of the thesis paper and the presentation of Handwritten Chinese Character Recognition(HCCR)problems, I present the fuzzy and structural approaches in the preclassification phase. The task of connecting the tablet and pen to the recognition of final Chinese character is considered in system design. Due to the complexities of the HCCR problems, this thesis research paper will first concentrate on the methodology by exposing theoretical system selection. My objective is to propose a practical, simple, and feasible method of HCCR and its preliminary implementation for the beginner interested in this problem. The algorithms for stroke extraction by finite automata and character matching by Depth First Searching (DFS) require specific conditions for preparing Dynamic Programming (DP). The next step will be setting up several databases, finding the appropriate tool of the tablet and pen, connecting correctly the tool to the computer,and finishing the postprocessing phase. These tasks are beyond the scope of this thesis.

Metadata is an important aspect in learning resource discovery, reuse of learning resources, within-collection organization of learning resources, retrieval of non-textual media, and interoperability among learning resources. This paper presents an analysis of the challenges in metadata creation field and defines a framework for customizable metadata authoring. Also it presents the CLOMAT prototype (Customizable Learning Objects Metadata Authoring Tool) which was build as a proof-of-concept for the framework presented in this paper. The research concludes with the evaluation results of the prototype. In particular, the paper explores issues associated with the application of standards, the properties of the learning objects, the needs of the users involved in the creation of metadata, the problems that arise in practice. These issues help to characterize the problem space of metadata creation and to explore the possibilities for improvement. As research integrating different technology aspects of metadata management is till in its infancy, I consider this and similar work as early framework for learning object metadata creation and management. These frameworks form the basis for more detailed app oaches to follow.

Wireless sensor networks are being built to facilitate automated information gathering in military, industrial, environmental and surveillance applications. Many such applications of Sensor Networks require desired/pre-fixed QoS (Quality of Service: packetdata delivery within a defined deadline) guarantees and as well as high reliability. These applications demand high packet delivery ratio and are extremely delay sensitive and also demand reliability. However, certain factors limit the ability of the multihop sensor network to achieve the desired goals. These factors include the delay caused by network congestion, or hot regions in the network, limited energy of the sensor networknodes, packet loss due to the collisions and link failure andand mobilitylink failure. In this work, Our paper suggests an energy aware dual-path routing scheme for real timedeadline driven traffic is proposed, which balances node energy utilization, which is energy awareincreases the network lifetime, reducess the routing delay across the network by taking network congestion into account and increases the reliability of the packets reaching the destination by introducing minimal data redundancy. We als This paper also o introduces an adaptive prioritized Medium Access Layer (MAC) specifically to provide a differentiated service model for real time packets. Our intuitive claims areis well supported by simulationby simulation results.

A Computer Vision application to automatically assess the vigor of three-day-old soybean seedlings was developed. The application operates on images of the three day- old seedlings on a brown paper towel background that was acquired with a flatbed scanner. From these images, each seedling is identified and then measured. With the computed lengths a 0 to 1000 vigor index is derived reducing the time necessary to measure the seedlings and increasing the standardization of the seedlingmeasuring step in the seed vigor test. The program extracts the soybeans away from the paper towel background in the image and converts them into various digital representations, including a 1-pixel wide summary structure of each seedling referred to as a skeleton, so that they may be measured. The main challenges addressed through the accomplishments of this research were1) developing methods to segment the soybeans from the background using advanced adaptive thresholding methods, 2)developing methods to classify the soybean seedlings into abnormal and normal categories, 3) using common features of the seedlings to put them into type categories so that the cotyledons, which should not be measured, could be removed from the skeleton, 4) developing a generic skeleton measuring algorithm that will allow the quantification of the complicated skeletons of the soybean seedlings, and5) developing a 0 to 1000 vigor index that indicates the potential for rapid, and uniform development of the seed lot represented by the image and gives a standardized number that can be used to compare the vigor between seed lots.

In 1930, a paper by Frank Plumpton Ramsey entitled "On a Problem of Formal Logic" appeared in the Proceedings of the London Mathematical Society. Although the impetus of this paper was one of mathematical logic, a far reaching combinatorial result was needed by Ramsey to achieve his objective. This combinatorial result became known as Ram \sey's Theorem. One of the combinatorial structures which was developed during the study of Ramsey's Theorem is that of a Ramsey graph. A Ramsey graph, denoted (k,l,n,e), is defined as an undirected graph that contains no cliques of size k, no independent sets of size I, with order n, and size e. Knowledge of Ramsey graphs is useful in the improvement of bounds and sometimes the calculation of exact values for various Ramsey number parameter situations. Straightforward enumeration of (k, I, n, e) Ramsey graphs for larger values of n is intractable with the current computing technology available. In order to produce such graphs, specialized algorithms need to be implemented. This thesis provides the theoretical background developed by Graver and Yackel [GRA68a], expanded upon by Grinstead and Roberts [GRl82a], and generalized by Radziszowski and Kreher [RAD88a, RAD88b] for the implementation of algorithms utilized for the enumeration of various Ram sey graphs. An object oriented graph manipulation package, including the above mentioned Ramsey graph enumeration algorithms, is implemented and documented. This package is utilized for the enumeration of all (3,3), (3,4), (3,5) and (3, 6) graphs. Some (3, 7) and (3, 8) also are calculated. These results duplicate and verify Ramsey graphs previously enumerated during other investigations. [RAD88a, RAD88b] In addition to these results, some newly enumerated (3,8) critical graphs, as well as some newly enumerated (3,9) graphs, including a minimum (3, 9, 26, 52) -graph are presented.

Graphs are, historically, among the most important and useful abstractions in both theoretical and applied computer science. However, due to a recent abundance of scientific data, social data, and machine learning and analysis methods based on graphical models, we are now facing a radical expansion of graph computing into new research areas. Dozens of graph-structured computing systems were published in the last decade, and graph theory itself is flourishing due to major results, such as the graph structure theorem. This dissertation presents several projects at the intersection of graph-structured systems, graph theory, and research practice, which are united by the common theme of vertex sort orders. We first present Sheep, a novel distributed graph partitioning algorithm based on vertex sorting. We show that Sheep finds balanced edge partitions with a low communication volume in near-linear time with excellent distributed scaling properties. Afterwards, we investigate graph systems research practices. We construct a corpus of research papers and identify the most referenced benchmark algorithms and datasets. We discuss systematic biases in these benchmarks, and in particular the hidden performance impact of their ``default'' vertex sort orders. We show this impact is significant by benchmarking a gold standard system, Galois, while controlling the vertex order. Finally, we address the role of synthetic data generators in graph systems research. We identify several features of the popular Kronecker graph model that inhibit its correct use as a standard benchmark. We propose and evaluate changes to the model to make it an appropriate benchmark. This dissertation shows that a graph's input shape and the performance of graph-structured system are inseparable; that a good graph shape (e.g., a partitioning) can be found as quickly as sorting; and that, in general, one must control the input shape to correctly measure a system's performance.

The Distributed Oceanographic Match-Up Service (DOMS) currently under development is a centralized service that allows researchers to easily match in situ and satellite oceanographic data from distributed sources to facilitate satellite calibration, validation, and retrieval algorithm development. The Shipboard Automated Meteorological and Oceanographic System (SAMOS) initiative provides routine access to high-quality marine meteorological and near-surface oceanographic observations from research vessels. SAMOS is one of several endpoints connected into the DOMS network, providing in-situ data for the match-up service. DOMS in-situ endpoints currently use Apache Solr as a backend search engine on each node in the distributed network. While Solr is a high-performance solution that facilitates creation and maintenance of indexed data, it is limited in the sense that its schema is fixed. The property graph model escapes this limitation by removing any prohibiting requirements on the data model, and permitting relationships between data objects. This paper documents the development of the SAMOS Neo4j property graph database including new search possibilities that take advantage of the property graph model, performance comparisons with Apache Solr, and a vision for graph databases as a storage tool for oceanographic data. The integration of the SAMOS Neo4j graph into DOMS is also described. Various data models are explored including spatial-temporal records from SAMOS added to a time tree using Graph Aware technology. This extension provides callable Java procedures within the CYPHER query language that generate in-graph structures used in data retrieval. Neo4j excels at performing relationship and path-based queries, which challenge relational-SQL databases because they require memory intensive joins due to the limitation of their design. Consider a user who wants to find records over several years, but only for specific months. If a traditional database only stores timestamps, this type of query could be complex and likely prohibitively slow. Using the time tree model in a graph, one can specify a path from the root to the data which restricts resolutions to certain time frames (e.g., months). This query can be executed without joins, unions, or other compute-intensive operations, putting Neo4j at a computational advantage to the SQL database alternative. That said, while this advantage may be useful, it should not be interpreted as an advantage to Solr in the context of DOMS. Solr makes use of Apache Lucene indexing at its core, while Neo4j provides its own native schema indexes. Ultimately they each provide unique solutions for data retrieval that are geared for specific tasks. In the DOMS setting it would appear that Solr is the most suitable option, as there seems to be very limited use cases where Neo4j does outperform Solr. This is primarily because the use case as a subsetting tool does not require the flexibility and path-based queries that graph database tools offer. Rather, DOMS nodes are using high performance indexing structures to quickly filter large amounts of raw data that are not deeply connected, a feature of large data sets where graph queries would indeed become useful.

In recent years, numerous technological advancements have revived a new interest in the study of Ballistocardiography (BCG). Ballistocardiography has many potential applications in the health field and has recently been integrated into various home health monitoring devices which allow users to consistently track information about their cardiovascular health. Despite these recent advancements, BCG signals are still very susceptible to noise due to the fact that the sensors must be sensitive enough to measure small mechanical movements of the body, unlike electrocardiography signals which are obtained from the body’s electrical activity. This paper focuses on examining two new techniques for processing BCG signals: Singular Spectrum Analysis and Savitzky-Golay filters. Multiple accelerometers are used to simultaneously collect data from varying locations of the body, and the strength of the BCG signal at these locations are independently analyzed. We determine the necessary parameters to enable each method to effectively and successfully remove noise and other artifacts from BCG data. Our results show that SSA is a powerful technique capable of separating the desired BCG signal from the numerous artifacts more effectively than standard techniques. Also, although BCG signals can be detected throughout the body, the sternum and parietal bone tends to contain the most distinguished signals.

Corfu is a platform for building systems which are extremely scalable, strongly consistent and robust. Unlike other systems which weaken guarantees to provide better performance, we have built Corfu with a resilient fabric tuned and engineered for scalability and strong consistency at its core: the Corfu shared log. On top of the Corfu log, we have built a layer of advanced data services which leverage the properties of the Corfu log. Today, Corfu is already replacing data platforms in commercial products, and the thriving Corfu open source code base enjoys regular contributions from a number of industrial and academic institutions.One of the key properties of Corfu is consistency, a highly desirable property which simplifies programming complex, asynchronous distributed systems by increasing the number of assumptions a programmer can make about how a system will behave. For years, system designers focused on providing the strongest possible guarantees on top of unreliable and even malicious systems. The rise of the Internet and cloud-scale computing, however, shifted the focus of system designers towards scalability. In a rush to meet the needs of cloud-scale workloads, system designers realized that if they weaken the consistency guarantees they provided, they would greatly increase the scalability of their systems. As a result, designers simplified the guarantees provided by their systems and weaker consistency models such as eventual consistency emerged, greatly increasing the burden on developers leading to error-prone applications. Programmers in the cloud era are forced to choose between consistency and scalability.Corfu, the topic of this dissertation, is a platform for scalable consistency. Corfu answers the question: ``If we were to build a distributed system from scratch, taking into consideration both the desire for consistency and the need for scalability, what would it look like?''. The answer lies in the Corfu distributed log. We begin by introducing the Corfu distributed log. Corfu achieves strong consistency by presenting the abstraction of a log --- clients may read from anywhere in the log but they may only append to the end of the log. The ordering of updates on the log is decided by a high throughput sequencer, which we show can handle nearly a million requests per second. The log is scalable as every update to the log is replicated independently, and every client appending to the log merely needs to acquire a token before beginning replication. This means that we can scale the log by merely adding replicas, and our only limit is the rate of requests the sequencer can handle.While building a single distributed log already provides strong consistency and scalability, multiple applications may wish to share the same log. By sharing the same log, updates across multiple applications can be ordered with respect to one another, which form the basic building block for advanced operations such as transactions. This dissertation details two designs for virtualizing the log: \emph{streaming}, which divides the log into streams built using log entries which point to one another, and \emph{\branching}, which virtualizes the log by radically changing how data is replicated in the shared log. \xmakefirstuc{\branching} greatly improves the performance of random reads, and allows applications to exploit locality by placing virtualized logs on a single replica.Efficiently virtualizing the log turns out to be important for implementing distributed objects in Corfu, a convenient and powerful abstraction for interacting with the Corfu distributed log introduced earlier. Rather than reading and appending entries to a log, distributed objects enable programmers to interact with in-memory objects which resemble traditional data structures such as maps, trees and linked lists. Under the covers, the \emph{Corfu runtime}, a library which client applications link to, translates accesses and modifications of in-memory objects into operations on the Corfu distributed log.The Corfu runtime provides rich support for objects. An automated translation process converts plain old Java objects (POJOs) directly into Corfu objects through both runtime and compile-time transformation of code. This allows programmers to quickly adapt existing code to run on top of Corfu. The Corfu runtime also provides strong support for transactions, which enables multiple applications to read and modify objects without relaxing consistency guarantees. We show that with stream materialization, Corfu can support storing large amounts of state while supporting strong consistency and transactions.Finally, we describe our experience in both writing new applications and adapting existing applications to Corfu. We start by building an adapter for Apache Zookeeper clients to run on top of Corfu and then describe the implementation of Silver, a new distributed file system which leverages the power of the Corfu log. We then conclude by describing our efforts to retrofit a large and complex commerical application: a software defined network (SDN) switch controller, and detail how the strong transaction model and rich object interface greatly reduce the burden on distributed system programmers.Overall, Corfu promises to change how data platforms are built. Instead of forcing developers to compromise, Corfu offers programmers the best of both worlds. This dissertation traces the development and evolution of Corfu over five years and through eight publications. As a testament to the robustness of the original Corfu log design, the core log API remains mostly unchanged from the early Corfu papers, even though Corfu is now a public open source project with contributors from multiple academic and industrial institutions.

Individuals affected by severe motor impairment lose the ability to move and interact effectively. The severity of motor impairment varies across affected individuals, and conditions like LIS and ALS can fluctuate and deteriorate over time, sometimes within the same day. Despite the increasing availability of augmentative and assistive technologies, existing solutions do not easily adapt to these changing capabilities. In this paper, we present Eyehome 3.0, an extensible multimodal input system which builds user interfaces that adapt to the user via recurrent neural networks. In particular, eye gaze data is used to inform the layout of these user interfaces. We also present a framework for easily developing additional layouts and applications for the system, and how additional devices can be incorporated. We discuss how our participatory design work with Amado, a person with LIS, helped drive the development of the system.

Multi-task learning (MTL) has recently contributed to learning better representations in service of various natural language (NLP) tasks. MTL aims at improving the performance of a primary task by jointly training on a secondary task. This paper introduces to the field of deep recurrent neural networks the concept of automated tasks, which exploit the sequential nature of the original input data, as secondary tasks in an MTL model. Specifically, we explore next word prediction, next character prediction, and missing word completion as potential automated tasks. Our results show that training on a primary task in parallel with a secondary automated task improves both the convergence speed and accuracy for the primary task. Furthermore, we suggest two methods for augmenting an existing network with automated tasks and establish better-than-baseline performance in topic prediction, sentiment analysis, and hashtag recommendation. Finally, we show that the MTL models can perform well on datasets like Twitter that are small and colloquial by nature. We claim that because every sequential dataset has associated automated tasks, automated MTL can be generalized to learn better representations for any sequential neural network model.

Density-based clustering algorithms such as DBSCAN have been widely used for spatial knowledge discovery as they offer several key advantages compared to other clustering algorithms. They can discover clusters with arbitrary shapes, are robust to noise and do not require prior knowledge (or estimation) of the number of clusters. The idea of using a scan circle centered at each point with a search radius Eps to find at least MinPts points as a criterion for deriving local density is easily understandable and sufficient for exploring isotropic spatial point patterns. However, there are many cases that cannot be adequately captured this way, particularly if they involve linear features or shapes with a continuously changing density such as a spiral. In such cases, DBSCAN tends to either create an increasing number of small clusters or add noise points into large clusters. Therefore, in this paper, we propose a novel anisotropic density-based clustering algorithm (ADCN). To motivate our work, we introduce synthetic and real-world cases that cannot be sufficiently handled by DBSCAN (and OPTICS). We then present our clustering algorithm and test it with a wide range of cases. We demonstrate that our algorithm can perform as equally well as DBSCAN in cases that do not explicitly benefit from an anisotropic perspective and that it outperforms DBSCAN in cases that do. We show that our approach has the same time complexity as DBSCAN and OPTICS, namely O(n log n) when using a spatial index and O(n 2 ) otherwise. We provide an implementation and test the runtime over multiple cases. Finally, we apply DBSCAN, OPTICS, and ADCN to the task of extracting urban areas of interest (AOI) from geotagged photos in six cities. Visual comparison shows that, comparing to DBSCAN and OPTICS, ADCN is inclined to extract AOIs with linear shapes which follow the underline road networks. ADCN also turns out to connect clusters when the spatial distribution of them shows similar directions.

Nash equilibrium is the central solution concept in Game Theory. Since Nash's original paper in 1951, it has found countless applications in modeling strategic behavior of traders in markets, (human) drivers and (electronic) routers in congested networks, nations in nuclear disarmament negotiations, and more.A decade ago, the relevance of this solution concept was called into question by computer scientists~\cite{DGP, CDT}, who proved (under appropriate complexity assumptions) that {\em computing} a Nash equilibrium is an intractable problem. And if centralized, specially designed algorithms cannot find Nash equilibria, why should we expect distributed, selfish agents to converge to one?The remaining hope was that at least approximate Nash equilibria can be efficiently computed.Understanding whether there is an efficient algorithm for {\em approximate Nash equilibrium} has been the central open problem in this field for the past decade.In this thesis, we provide strong evidence that even finding an approximate Nash equilibrium is intractable.We prove several intractability theorems for different settings (two-player games and many-player games) and models (computational complexity, query complexity, and communication complexity).In particular, our main result is that under a plausible and natural complexity assumption (``Exponential Time Hypothesis for \PPAD''), there is no polynomial-time algorithm for finding an approximate Nash equilibrium in two-player games.The problem of approximate Nash equilibrium in a two-player game poses a unique technical challenge: it is a member of the class \PPAD, which captures the complexity of several fundamental total problems, i.e. problems that always have a solution; and it also admits a quasipolynomial ($\approx n^{\log n}$) time algorithm. Either property alone is believed to place this problem far below \NP-hard problems in the complexity hierarchy; having both simultaneously places it just above \P, at what can be called the frontier of intractability. Indeed, the tools we develop in this thesis to advance on this frontier are useful for proving hardness of approximation of several other important problems whose complexity lies between \P~and \NP:\begin{description}\item [Brouwer's fixed point]Given a continuous function $f$ mapping a compact convex set to itself, Brouwer's fixed point theorem guarantees that $f$ has a fixed point, i.e. $x$ such that $f(x) = x$.Our intractability result holds for the relaxed problem of finding an approximate fixed point, i.e. $x$ such that $f(x) \approx x$.\item [Market equilibrium]Market equilibrium is a vector of prices and allocations where the supply meets the demand for each good.%We consider the Arrow-Debreu model where agents are both sellers and buyers of goods. Our intractability result holds for the relaxed problem of finding an approximate market equilibrium, where the supply of each good approximately meets the demand.\item [CourseMatch (A-CEEI)]Approximate Competitive Equilibrium from Equal Income (A-CEEI) is the economic principle underlying CourseMatch, a system for fair allocation of classes to students (currently in use at Wharton, University of Pennsylvania). \item [Densest $k$-subgraph]Our intractability result holds for the following relaxation of the $k$-Clique problem: given a graph containing a $k$-clique, the algorithm has to find a subgraph over $k$ vertices that is ``almost a clique'', i.e. most of the edges are present.\item [Community detection]We consider a well-studied model of communities in social networks, where each member of the community is friends with a large fraction of the community, and each non-member is only friends with a small fraction of the community.\item [VC dimension and Littlestone dimension]The Vapnik-Chervonenkis (VC) dimension is a fundamental measure in learning theory that captures the complexity of a binary concept class. Similarly, the Littlestone dimension is a measure of complexity of online learning. \item [Signaling in zero-sum games]We consider a fundamental problem in signaling, where an informed signaler reveals private information about the payoffs in a two-player zero-sum game, with the goal of helping one of the players.

Technology guides the practice of scientific inquiry. In the biological sciences, DNA sequencing has encouraged the commingling of traditional experimental biology and computer science. In this thesis, I describe computational and biochemical methods to advance nanopore DNA sequencing technology. Nanopore DNA sequencing is an single-molecule technology that shows promise in the area of read lengths, instrument portability, and, as shown in this work, chemical modification detection. A nanopore sequencing device contains a nanometer-sized pore that separates two electrolyte buffer reservoirs. A voltage potential is applied across the nanopore and the device records the ionic current through pore. As DNA polymers translocate through the pore they modulate the ionic current by partially obstructing the pore in a sequence-dependent manner. Commercial sequencing devices contain arrays of pores that are individually controlled.The first chapter of this work provides a technical background and practical application of the technology. In the following chapters I describe new algorithms that improve the utility of the technology. In Chapter 2 I describe a hidden Markov model for the nanopore ionic current. I describe how multiple model topologies were implemented, one of which included modeling the time domain. I describe how a hierarchical Dirichlet process can be used to model new non-canonical (modified) bases affording a HMM-HDP model. In chapter \ref{chap:detecting_modifications} I show how the model can be trained to detect multiple chemical modifications on synthetic DNA and genomic DNA samples. In the final section of this chapter I show how the model was incorporated into a cloud-based pipeline that allows for horizontal scaling of the datasets.The last section of the paper describes a biochemical method to re-read DNA sequences using a nanopore and a helicase enzyme. Multiple biochemical techniques have been used to increase the accuracy of nanopore sequencing; re-reading the same strand allows for stochastic errors to be differentially placed in multiple passes.

In recent years, Social Network Service (SNS) is a novel, popular way to make friends andconvey information online. Therefore, the analysis of network data has attracted a lot ofattention. It is an area that is rapidly growing, both with Statistics and Computer Science.This paper rst provides a summary of statistical methods used in network data analysis,including basic denitions, measurements, and descriptive statistics. We then introduce theExponential Random Graph Model to t network data. Secondly, we dig into a more specicarea of network analysis: Community Detection. We discuss two dierent methods to explorethe community stucture, one is Louvain algorithm and the other is Mixed MembershipStochastic Blockmodels. After that, we combine the community identication with a two-stage user similarity algorithm to build a friend recommendation method. In the empiricalstudy section, we apply this method to a real-world dataset and evaluate its performacethrough specic measurements.

Modern HPC applications pose high demands on I/O performance and storage capability. The emerging non-volatile memory (NVM) techniques, such as Phase Change Memory and STT-RAM, offer low-latency, high bandwidth, and persistence for HPC applications. However, the existing I/O stack, including OS, high level library, I/O middleware, and applications, are designed and optimized based on an assumption of disk-based storage. To effectively use NVM, we must re-examine the existing I/O sub-system to properly integrate NVM into it. Using NVM as a fast storage, the previous assumption on the inferior performance of storage (e.g., hard drive) is not valid any more. The performance problem caused by slow storage may be mitigated; the existing mechanisms to narrow the performance gap between storage and CPU may be unnecessary and result in large overhead. Thus fully understanding of the impact of introducing NVM into the HPC software stack demands a thorough performance study. In this paper, we analyze and model the performance of I/O intensive HPC applications with NVM as a block device. We study the performance from three perspectives: (1) the impact of NVM on the performance of traditional page caches; (2) a performance comparison between MPI individual I/O and POSIX I/O; and (3) the impact of NVM on the performance of collective I/O. We reveal the diminishing effects of page caches, ignorable performance difference between MPI individual I/O and POSIX I/O, and performance disadvantage of collective I/O on NVM due to unnecessary data shuffling. We model the performance of MPI collective I/O and study the complex interaction between data shuffling, storage performance, and I/O access patterns. Extensive experiments have been conducted to verify our analysis.Keywords: NVM, page cache, MPI I/O, Collective I/O

Wi-Fi Direct is a new technology which establishes Wi-Fi peer to peer (P2P) connection between devices at the network layer. It is very similar to Bluetooth which could connect different devices without Wi-Fi or LTE network, but it is faster than Bluetooth. Systems above Android 4.0 provide the Wi-Fi Direct Application Programmer Interface (API) for Android devices to communicate in the same Wi-Fi Direct group. But currently Multi-Group communication (communication between the groups) can not be implemented directly by using Wi-Fi Direct API on Android system. This paper proposes a new topology for Wi-Fi Direct Multi-Group communication, and a new device Access Point is introduced as an intermediary to connect different Wi-Fi Direct Groups. If a Client (Client1) in one group wants to communicate with another Client (Client2) in another group, Client1 will first send the information to its Group Owner (GO1), and then GO1 will send the information to the Access Point. The Access Point can determine which Group Owner the Client2 connects with through the routing table information, and it will send the packets to the target Group owner (GO2). Finally, GO2 will send the packets to destination Client2, which finishes a complete Wi-Fi Direct Multi-Group communication.

Smart Objects [1] from games like The Sims are used to create Smart Terrain. Smart Terrain allows for the terrain to hold the variables a non-player character (NPC) requires to decide where to move. Probabilistic Smart Terrain [2] builds on Smart Terrain by adding the probability that given Smart Objects may not meet a need 100% of the time. This paper will cover the implementation of the Probabilistic Smart Terrain algorithm in a Unity 2D game. The game is tile based and operates in real-time. A map flooding algorithm was designed using Raycast objects within the Unity environment. A playable game was created in which the Probabilistic Smart Terrain algorithm is used to control a NPC that patrols the area guarding three treasure objects. The terrain holds the distances to each coin while the Probabilistic Smart Terrain algorithm is used to calculate the probability of where to find the player. This causes the NPC to move to the most probable direction of finding the player. This implementation will allow actual players to evaluate the apparent intelligence of the Probabilistic Smart Terrain algorithm.