With about 3.6 million adults in the United States having visual impairment or blind- ness, assistive technology is essential to give these people grocery shopping independance. This thesis presents a new method to detect and localize nutrition facts tables (NFTs) on mobile devices more quickly and from less-ideal inputs than before. The method is a drop- in replacement for an existing NFT analysis pipeline and utilizes multiple image analysis methods which exploit various properties of standard NFTs.In testing, this method performs very well with no false-positives and 42% total recall. These results are ideal for real-world application where inputs are analyzed as quickly as possible. Additionally, this new method exposes many possibilities for future improvement.

The Dairy industry requires accurate detection of oestrus(heat) in dairy cows to maximise output of the animals. Traditionally this is a process dependant on human observation and interpretation of the various signs of heat. Many areas of the dairy industry can be automated, however the detection of oestrus is an area that still requires human experts. This thesis investigates the application of Machine Learning classification techniques, on dairy cow milking data provided by the Livestock Improvement Corporation, to predict oestrus. The usefulness of various ensemble learning algorithms such as Bagging and Boosting are explored as well as specific skewed data techniques. An empirical study into the effectiveness of classifiers designed to target skewed data is included as a significant part of the investigation. Roughly Balanced Bagging and the novel Under Bagging classifiers are explored in considerable detail and found to perform quite favourably over the SMOTE technique for the datasets selected. This study uses non-dairy, commonplace, Machine Learning datasets; many of which are found in the UCI Machine Learning Repository.

This thesis describes the experimental test results and computer simulation investigations which were conducted to verify the existing theory of skewed disk forced response predictions. The experimental tests were conducted on a horizontal flexible shaft rotor system supported in two hydrodynamic journal bearings. The computer simulations were conducted with a program that uses a matrix transfer method to get the desired solution. The agreement between experiment and simulation is very good for most skewed disk response characteristics. The influence of measurement errors and operation condition uncertainties are discussed.In the first part of this study, the dynamic behavior of experimental investigations focused on two different skewed disk designs which were mounted at midspan, 1/3 span and 2/3 span of the shaft. The two skewed disks were designed to allow a fine angle adjustment of the desired skew angle. The two designs are (a) the angle tiltable disk and (b) the couple unbalanced mass disk. The experimental results are shown to be close to the theoretical predictions of other authors.In the second part of this study, an existing computer program was used to simulate the experimental test rotor. The results give excellent qualitative agreement although there are some differences in quantitative analysis comparisons. The forced response characteristics of the computer simulation match the experimental results. It has been shown that by using the approximate linear skewed disk model, it is possible to get similar results to the experimental tests for similar disk skew conditions. / Master of Science

In this thesis, we propose a circuit design technique that reduces the energy utilized by any logic circuit for computation. We achieve this, by reducing the voltage swing on the circuit without greatly compromising the speed of operation and keeping in mind the noise margin constraints. Our technique involves the use of head or tail transistors that provide a Vth drop in the voltage swing. We choose to use head or tail transistors on alternate logic levels providing us with an option of driver stage, based on the noise margin of the subsequent stage. We demonstrate the working of this concept on inverter chains, to prove the correctness as well as the ability of the reduced voltage swing circuits to drive subsequent stages. We also discuss the implementation of this technique on basic gates and simple combinational circuits. We then show detailed experiments on a larger circuit, in this case a Kogge-Stone parallel prefix adder. We will discuss the overheads involved in the design and methods to partially overcome these by the use of selectively skewed gates and application of forward body bias. Finally we implement the same design using a different technology to demonstrate the scalability of the technique. / text

Due to the financial turmoil in recent years, risk management has become an important issue, investors would like to be fully-prepared to cope with financial crisis before it happen. This research uses the Fama and French three-factor and the U.S. Dollar Index (USDX) as an exchange rate variations indicator to capture the international relations. It constitutes a four-factor model to analyze the S&P100 stock returns changes, and we introduce the skewed-t distribution to simulate the distribution of stock returns and capture the characteristics of skewness and kurtosis. We use cluster analysis to cluster the sample companies by their risk characteristics. And then we observe the explanatory power of each risk factor. The study shows that the S&P100 stocks are subjected to the market premium, and the scale effect is smaller than others. ¡@¡@ At last, in accordance with the GARCH-Skewed-t model to simulate the average, variance, skewness and kurtosis of each cluster. We track the long-term performance of each parameter which are used to observe the unusual changes before financial crisis. The empirical results show that the skewness parameter has perfect warning for financial turmoil. The cluster with warning ability is affected by B/M ratio effect and exchange rate changes. Among the case, the cluster has the best early warning effect when it's influenced by the exchange rate indicator. It displays that by adding an exchange rate risk indicator into the multi-factor model, we will have a better clustering result. It means that the skewness parameter of cluster with influence of exchange rate indicator can be used to observe financial turmoil, which can in turns, be used as an early warning system to determine the occurrence of extreme events.

The classic nonparametric confidence intervals for a difference or ratio of medians assume that the distributions of the response variable or the log-transformed response variable have identical shapes in each population. Asymptotic distribution-free confidence intervals for a difference and ratio of medians are proposed which do not require identically shaped distributions. The new asymptotic methods are easy to compute and simulation results show that they perform well in small samples.

The objective of the research presented in this thesis was to study and optimize the transverse deck reinforcement for a skewed concrete bridge deck supported by Hybrid Composite Beams (HCB's).  An HCB consists of a Glass Fiber Reinforced Polymer outer shell, a concrete arch, and high strength seven wire steel strands running along the bottom to tie the ends of the concrete arch together.  The remaining space within the shell is filled with foam.  The concrete arch does not need to be cast until the beam is in place, making the HCB very light during shipping.  This lowers construction costs and time since more beams can be transported per truck and smaller cranes can be used.  HCB's are quite flexible, so AASHTO LRFD's design model for bridge decks, as a one-way slab continuous over rigid supports, might not apply well to the HCB's deck design. A skewed three HCB girder bridge with a reinforced concrete deck and end diaphragms was built in the laboratory at Virginia Tech.  Concentrated loads were applied at locations chosen to maximize the negative and positive moments in the deck in the transverse direction.  The tests revealed that the transverse reinforcement was more than adequate under service loads. An Abaqus model was created to further study the behavior of the bridge and to help create future design recommendations.  The model revealed that the HCB bridge was behaving more like a stiffened plate at the middle section of the bridge, indicating that the flexibility of the girders needed to be considered. / Master of Science

Cars and other automobiles are used for transportation every single day all over the world. Almost ninety percent of the households have access to at least one car. Along with this, the chance of getting into an accident has also risen. Our objective is to deal with the case of parking. Everyone must have felt anxiety about the distances while sitting inside the car. Oneway or the other way, it always leads to property damage, etc. It will be difficult for people who spend lavish money on their vehicles to see the damage. In our project, we designed a system that aids drivers with information about distances between the vehicle and a wall or obstacle. It will be even more helpful if the driver has an idea of the distance between the wall and the vehicle by visual representation not as a text because of natural instinct. A hardware setup is fixed to the walls of the garage after analyzing all the possible scenarios including skewed parking which only guides the driver. This system uses sensors like ultrasonic to get information about the distances and an LED strip to guide the driver about the distances. Depending on the distances the LED will indicate either red or green and also the number of glowing LEDs. The two signals to the LED are pre-calibrated after considering the garage spacing. This system can effectively remove the chance of accidents while parking a vehicle in the garage.

Clustering is the process of finding underlying group structure in data. As the scale of data collection continues to grow, this “big data” phenomenon results in more complex data structures. These data structures are not always compatible with traditional clustering methods, making their use problematic. This thesis presents methodology for analyzing samples of four-way and higher data, examples of these more complex data types. These data structures consist of samples of continuous data arranged in multidimensional arrays. A large emphasis is placed on clustering this data using mixture models that leverage tensor-variate distributions to model the data. Parameter estimation for all these methods are based on the expectation-maximization algorithm. Both simulated and real data are used for illustration. / Thesis / Doctor of Science (PhD)

No description available.