Artificial intelligence.

Deep Learning is an increasingly important subdomain of arti cial intelligence.
Deep Learning architectures, arti cial neural networks characterized by having both
a large breadth of neurons and a large depth of layers, bene ts from training on Big
Data. The size and complexity of the model combined with the size of the training
data makes the training procedure very computationally and temporally expensive.
Accelerating the training procedure of Deep Learning using cluster computers faces
many challenges ranging from distributed optimizers to the large communication overhead
speci c to a system with o the shelf networking components. In this thesis, we
present a novel synchronous data parallel distributed Deep Learning implementation
on HPCC Systems, a cluster computer system. We discuss research that has been
conducted on the distribution and parallelization of Deep Learning, as well as the
concerns relating to cluster environments. Additionally, we provide case studies that
evaluate and validate our implementation.

A bio-inspired robotic underwater vessel was developed to test the effect of
fin morphology on the propulsive performance of caudal fin. The robotic vessel, called The
Bullet Fish, features a cylindrical body with a hemisphere at the forward section and a
conical body at the stern. The vessel uses an oscillating caudal fin for thrust generation.
The robotic vessel was tested in a recirculating flume for seven different caudal fins that
range different bio-inspired forms and aspect ratios. The experiments were performed at
four different flow velocities and two flapping frequencies: 0.5 and 1.0 Hz. We found that
for 1 Hz flapping frequency that in general as the aspect-ratio decreases both thrust
production tends and power decrease resulting in a better propulsive efficiency for aspect
ratios between 0.9 and 1.0. A less uniform trend was found for 0.5 Hz, where our data
suggest multiple efficiency peaks. Additional experiments on the robotic model could help
understand the propulsion aquatic locomotion and help the design of bio-inspired
underwater vehicles.

Modern cancerous tumor diagnostics is nearly impossible without invasive
methods, such as biopsy, that may require involved surgical procedures. In recent years
some work has been done to develop alternative non-invasive methods of medical
diagnostics. For this purpose, the data obtained from an ultrasound image of the body crosssection,
has been analyzed using statistical models, including Rayleigh, Rice, Nakagami,
and K statistical distributions. The homodyned-K (H-K) distribution has been found to be
a good statistical tool to analyze the envelope and/or the intensity of backscattered signal
in ultrasound tissue characterization. However, its use has usually been limited due to the
fact that its probability density function (PDF) is not available in closed-form. In this work
we present a novel closed-form representation for the H-K distribution. In addition, we propose using the first order approximation of the H-K distribution, the I-K distribution
that has a closed-form, for the ultrasound tissue characterization applications. More
specifically, we show that some tissue conditions that cause the backscattered signal to
have low effective density values, can be successfully modeled by the I-K PDF. We
introduce the concept of using H-K PDF-based and I-K PDF-based entropies as additional
tools for characterization of ultrasonic breast tissue images. The entropy may be used as a
goodness of fit measure that allows to select a better-fitting statistical model for a specific
data set. In addition, the values of the entropies as well as the values of the statistical
distribution parameters, allow for more accurate classification of tumors.

Object recognition is imperfect; often incomplete processing or deprived
information yield misperceptions (i.e., misidentification) of objects. While quickly
rectified and typically benign, instances of such errors can produce dangerous
consequences (e.g., police shootings). Through a series of experiments, this study
examined the competitive process of multiple object interpretations (candidates) during
the earlier stages of object recognition process using a lexical decision task paradigm.
Participants encountered low-pass filtered objects that were previously demonstrated to
evoke multiple responses: a highly frequented interpretation (“primary candidates”) and a
lesser frequented interpretation (“secondary candidates”). When objects were presented
without context, no facilitative effects were observed for primary candidates. However,
secondary candidates demonstrated evidence for being actively suppressed.

With a focus on dynamics and control, an aquatic quadcopter with optical wireless communications is modeled, designed, constructed, and tested. Optical transmitter and receiver circuitry is designed and discussed. By utilization of the small angle assumption, the nonlinear dynamics of quadcopter movement are linearized around an equilibrium state of zero motion. The set of equations are then tentatively employed beyond limit of the small angle assumption, as this work represents an initial explorative study. Specific constraints are enforced on the thrust output of all four rotors to reduce the multiple-input multiple-output quadcopter dynamics to a set of single-input single-output systems. Root locus and step response plots are used to analyze the roll and pitch rotations of the quadcopter. Ultimately a proportional integral derivative based control system is designed to control the pitch and roll. The vehicle’s yaw rate is similarly studied to develop a proportional controller. The prototype is then implemented via an I2C network of Arduino microcontrollers and supporting hardware.

Image Processing and Computer Vision solutions have become commodities
for software developers, thanks to the growing availability of Application Program-
ming Interfaces (APIs) that encapsulate rich functionality, powered by advanced al-
gorithms. To understand and create an e cient method to process faces in images
by computers, one must understand how the human visual system processes them.
Face processing by computers has been an active research area for about 50
years now. Face detection has become a commodity and is now incorporated into
simple devices such as digital cameras and smartphones.
An iOS app was implemented in Objective-C using Microsoft Cognitive Ser-
vices APIs, as a tool for human vision and face processing research. Experimental
work on image compression, upside-down orientation, the Thatcher e ect, negative
inversion, high frequency, facial artifacts, caricatures and image degradation were
completed on the Radboud and 10k US Adult Faces Databases along with other
images.

A self-adaptive software is developed to predict the stock market. It’s Stock
Prediction Engine functions autonomously when its skill-set suffices to achieve its goal,
and it includes human-in-the-loop when it recognizes conditions benefiting from more
complex, expert human intervention. Key to the system is a module that decides of
human participation. It works by monitoring three mental states unobtrusively and in real
time with Electroencephalography (EEG). The mental states are drawn from the
Opportunity-Willingness-Capability (OWC) model. This research demonstrates that the
three mental states are predictive of whether the Human Computer Interaction System
functions better autonomously (human with low scores on opportunity and/or
willingness, capability) or with the human-in-the-loop, with willingness carrying the
largest predictive power. This transdisciplinary software engineering research
exemplifies the next step of self-adaptive systems in which human and computer benefit from optimized autonomous and cooperative interactions, and in which neural inputs
allow for unobtrusive pre-interactions.

Vehicular Ad hoc NETworks (VANETs) are a subclass of Mobile Ad hoc NETworks
and represent a relatively new and very active field of research. VANETs will enable in
the near future applications that will dramatically improve roadway safety and traffic
efficiency. There is a need to increase traffic efficiency as the gap between the traveled
and the physical lane miles keeps increasing. The Dynamic Traffic Assignment problem
tries to dynamically distribute vehicles efficiently on the road network and in accordance
with their origins and destinations. We present a novel dynamic decentralized and
infrastructure-less algorithm to alleviate traffic congestions on road networks and to fill
the void left by current algorithms which are either static, centralized, or require
infrastructure. The algorithm follows an online approach that seeks stochastic user
equilibrium and assigns traffic as it evolves in real time, without prior knowledge of the traffic demand or the schedule of the cars that will enter the road network in the future.
The Reverse Online Algorithm for the Dynamic Traffic Assignment inspired by Ant
Colony Optimization for VANETs follows a metaheuristic approach that uses reports from
other vehicles to update the vehicle’s perceived view of the road network and change route
if necessary. To alleviate the broadcast storm spontaneous clusters are created around
traffic incidents and a threshold system based on the level of congestion is used to limit
the number of incidents to be reported. Simulation results for the algorithm show a great
improvement on travel time over routing based on shortest distance. As the VANET
transceivers have a limited range, that would limit messages to reach at most 1,000 meters,
we present a modified version of this algorithm that uses a rebroadcasting scheme. This
rebroadcasting scheme has been successfully tested on roadways with segments of up to
4,000 meters. This is accomplished for the case of traffic flowing in a single direction on
the roads. It is anticipated that future simulations will show further improvement when
traffic in the other direction is introduced and vehicles travelling in that direction are
allowed to use a store carry and forward mechanism.

The 1990s ushered in what historian Susan Stryker describes as “a tremendous burst of new transgender activism” in the United States. Concomitantly, the success of Star Trek: The Next Generation led to a renaissance of US science fiction television. This dissertation asks, what is the relation between transgender (trans) politics and US science fiction (sf) television from 1990 to the present? The theoretical framework is Trans/Elemental feminism, a new paradigm developed in the dissertation. The method is multiperspectival cultural studies, which considers how the production, content, and reception of media texts and their metatexts collectively determine the texts’ meaning. The data include trade articles about the television industry; published interviews with producers; 3,175 hours of televisual content; commercial advertisements for television programs; films, novels, and webisodes (Web episodes) in selected media franchises; professional reviews; online discussion boards; fan fiction; and fan videos.

Sentiment analysis of tweets is an application of mining Twitter, and is growing
in popularity as a means of determining public opinion. Machine learning algorithms
are used to perform sentiment analysis; however, data quality issues such as high dimensionality, class imbalance or noise may negatively impact classifier performance.
Machine learning techniques exist for targeting these problems, but have not been
applied to this domain, or have not been studied in detail. In this thesis we discuss
research that has been conducted on tweet sentiment classification, its accompanying
data concerns, and methods of addressing these concerns. We test the impact
of feature selection, data sampling and ensemble techniques in an effort to improve
classifier performance. We also evaluate the combination of feature selection and
ensemble techniques and examine the effects of high dimensionality when combining
multiple types of features. Additionally, we provide strategies and insights for
potential avenues of future work.