In this thesis, we address two major problems in Generative Adversarial Networks (GAN), an important sub-field in deep learning. The first problem that we address is the instability in the training process that happens in many real-world problems and the second problem that we address is the lack of a good evaluation metric for the performance of GAN algorithms. To understand and address the first problem, three approaches are developed. Namely, we introduce randomness to the training process; we investigate various normalization methods; most importantly we develop a better parameter initialization strategy to help stabilize training. In the randomness techniques part of the thesis, we developed two randomness approaches, namely the addition of gradient noise and the batch random flipping of the results from the discrimination section of a GAN. In the normalization part of the thesis, we compared the performances of the z-score transform, the min-max normalization, affine transformations and batch normalization. In the most novel and important part of this thesis, we developed techniques to initialize the GAN generator section with parameters that can produce a uniform distribution on the range of the training data. As far as we are aware, this seemingly simple idea has not yet appeared in the extant literature, and the empirical results we obtain on 2-dimensional synthetic data show marked improvement. As to better evaluation metrics, we demonstrate a simple yet effective way to evaluate the effectiveness of the generator using a novel "overlap loss".

In this thesis, we address two major problems in Generative Adversarial Networks (GAN), an important sub-field in deep learning. The first problem that we address is the instability in the training process that happens in many real-world problems and the second problem that we address is the lack of a good evaluation metric for the performance of GAN algorithms. To understand and address the first problem, three approaches are developed. Namely, we introduce randomness to the training process; we investigate various normalization methods; most importantly we develop a better parameter initialization strategy to help stabilize training. In the randomness techniques part of the thesis, we developed two randomness approaches, namely the addition of gradient noise and the batch random flipping of the results from the discrimination section of a GAN. In the normalization part of the thesis, we compared the performances of the z-score transform, the min-max normalization, affine transformations and batch normalization. In the most novel and important part of this thesis, we developed techniques to initialize the GAN generator section with parameters that can produce a uniform distribution on the range of the training data. As far as we are aware, this seemingly simple idea has not yet appeared in the extant literature, and the empirical results we obtain on 2-dimensional synthetic data show marked improvement. As to better evaluation metrics, we demonstrate a simple yet effective way to evaluate the effectiveness of the generator using a novel "overlap loss".

This thesis presents the design of a step-up DC-DC converter with on-chip coupled inductors. Circuit theory of DC-DC converters in general is presented, after which a mathematical model of a step up converter is developed. A circuit implementation optimized from results of the mathematical model follows. For a completely integrated step-up converter, the inductor size is reduced by increasing the frequency of operation and using a circuit topology that employs coupled inductors. Spiral inductors are also studied to achieve maximum quality factor and inductance. A fast PWM control system is used to regulate the high-frequency converter. The fabrication was done in standard TSMC 0.18-$\mu$m digital CMOS process for four circuits, including one with a conventional topology and the others with a coupled inductor topology with varying inductor geometries. Measurement results from a fabricated prototype have been presented, demonstrating the functionality of the four circuits with coupled inductors on the fabricated chip and the improvement of the coupled solution over the conventional design. It is demonstrated that the circuits with coupled inductors have a significant improvement in performance based on conversion ratio and efficiency. Finally, the design process is evaluated and recommendations are made for future work. Furthermore, a new self-oscillating and robust control system is proposed that enables simpler and more efficient regulation for high-frequency converters such as one developed for this thesis.

Master of Science / Department of Animal Sciences and Industry / James S. Drouillard / Ruminal characteristics and feedlot performance were measured for steers adapted to a high-grain diet using a traditional 22-d step-up program (Control) and for steers adapted over 10 d with the aid of M. elsdenii NCIMB 41125 fresh culture administered as an oral gavage (Fresh), an oral gavage of rehydrated lyophilized culture (Rehyd), or Rehyd combined with lyophilized culture powder administered daily as a ration top-dress (Rehyd+Daily). Yearling steers (n = 435; initial BW = 408 ± 5 kg) were blocked by weight and randomly allocated to 64 concrete surfaced pens with 7 steers/pen and 16 pens/treatment. Step-up diets contained 40, 30, and 20% corn silage, and were followed by a finishing diet containing 10% corn silage and 90% concentrate. Ruminal fluid was obtained from a subset of steers by rumenocentesis 26 h after their first feeding, and inoculated into culture tubes containing lactate medium to determine differences in capacity for lactate metabolism over 24 h. A radiofrequency pH bolus was placed in the reticulorumen of 32 steers, that measured ruminal pH every 10 min for 124 d. Steers were fed once daily ad libitum for 156 d, then weighed, and transported 450 km to a commercial abattoir for harvest. HCW and incidence of liver abscesses were determined at harvest and carcass traits were evaluated after 36 h of refrigeration. No differences were detected for feedlot performance (P > 0.20), liver abscesses (P = 0.45), or carcass traits (P > 0.20). Capacity for lactate utilization was increased with all forms of M. elsdenii, as evidenced by increases in optical density (absorbance) of M. elsdenii cultures, disappearance of lactate, and increase in butyrate production (P < 0.01). Steers on Rehyd and Rehyd+Daily treatments spent less time between pH thresholds of 5.6 > pH > 5.2 (P < 0.01); 5.2 > pH > 5.0 (P < 0.01); and pH < 5.0 (P < 0.01) than Controls throughout the finishing period. In conclusion, steers dosed with M. elsdenii can be stepped up to finishing diets in 10 d with no adverse effects on performance. Ruminal fluid containing fresh cultures or freeze-dried and rehydrated cultures of M. elsdenii NCIMB 41125 were equally effective in metabolizing lactic acid. Lyophilized M. elsdenii resulted in less time below important pH thresholds, but no further benefit of daily administration was realized in this experiment.

This thesis deals with the design and realization of a DC/DC converter, which allows power conversion from 10-52 V input voltage to 10-52 V output voltage at a maximum output current of 3 A. The converter can communicate with the superior system via the CAN. In the first chapter I deal with the available integrated circuits of DC/DC converters, from which I choose one for my application. In the second chapter, I propose a block scheme of the converter, which includes the requirements from the assignment as well as my additional ones. In the third chapter I deal with the design of an electronic circuit and with the calculation of control loops. In the fourth chapter I propose firmware for the STM32 MCU, which controls the device and communicates with the superior system. In the last chapter the DC/DC converter is tested.

No description available.

A boost converter is one of the most efficient techniques to step up DC input voltage to a higher needed DC output voltage. The boost converter has many possible applications, such as in a photovoltaic system, hybrid car and battery charger. The proposed prototype in this report is a proof of concept that a Two-Phase Boost Converter is a possible improvement topology to offer higher efficiency without compromising any advantages readily offered by a basic boost. The prototype is designed to be able to handle up to 200 watts of output power with an input of 36 volts and an output of 48 volts. This paper goes through step-by-step the calculation, design, build and test of a Two-Phase Boost Converter. Calculations found in this paper were done on Mathcad and the simulations were done on LTSpice and Pspice. These include converter’s efficiency and other measures of converter’s performance. Advantages, disadvantages as well as possible improvements of the proposed topology will be presented. Data collected and analyzed from the prototype were done on a bench test, not through an actual application.

The importance of multiplicity adjustment has gained wide recognition in modern scientific research. Without it, there will be too many spurious results and reproducibility becomes an issue; with it, if overtly conservative, discoveries will be made more difficult. In the current literature on repeated testing of multiple hypotheses, Bonferroni-based methods are still the main vehicle carrying the bulk of multiplicity adjustment. There is room for power improvement by suitably utilizing both hypothesis-wise and analysis- wise dependencies. This research will contribute to the development of a natural group-sequential extension of the classical stepwise multiple testing procedures, such as Dunnett’s stepdown and Hochberg’s step-up procedures. It is shown that the proposed group-sequential procedures strongly control the familywise error rate while being more powerful than the recently developed class of group-sequential Bonferroni-Holm’s procedures. Particularly in this research, a convexity property is discovered for the distribution of the maxima of pairwise null P-values with the underlying test statistics having distributions such as bivariate normal, t, Gamma, F, or Archimedean copulas. Such property renders itself for an immediate use in improving Holm’s procedure by incorporating pairwise dependencies of P-values. The improved Holm’s procedure, as all stepdown multiple testing procedures, can also be naturally extended to group-sequential setting. / Statistics

This dissertation presents the derivation, analysis and application issues of advanced topologies with coupled inductors. The proposed innovative solutions can achieve significant performance improvement compared to the state-of-the-art technology. New applications call for high-efficiency high step-up DC-DC converters. The basic topologies suffer from extreme duty ratios and severe rectifier reverse recovery. Utilizing coupled inductor is a simple solution to avoid extreme duty ratios, but the leakage inductance associated with the coupled inductor induces severe voltage stress and loss. An innovative solution is proposed featuring with efficient leakage energy recovery and alleviated rectifier reverse recovery. Impressive efficiency improvement is achieved with a simple topology structure. The coupled inductor switching cell is identified. Topology variations and evaluations are also addressed. The concept that utilizes coupled inductors to alleviate rectifier reverse recovery is then extended, and new topologies suitable for other applications are generated. The proposed concept is demonstrated to solve the severe rectifier reverse recovery that occurs in continuous current mode (CCM) boost converters. Significant profile reduction and power density improvement can be achieved in front-end CCM power factor correction (PFC) boost converters, which are the overwhelmingly choice for use in telecommunications and server applications. This dissertation also proposes topologies to realize the single-stage parallel PFC by utilizing coupled inductors. Compared to the state-of-the-art single-stage PFC converters, the proposed topologies introduce a new power flow pattern that minimizes the bulk-capacitor voltage stress and the switch current stress. / Ph. D.

This thesis introduces a new photovoltaic (PV) system architecture employing low voltage parallel-connected PV panels interfaced to a high voltage regulated DC bus of a three-phase grid-tied inverter. The concept provides several improvements over existing technologies in terms of cost, safety, reliability, and modularity. A novel resonant mode DC-DC boost converter topology is proposed to enable the PV modules to deliver power to the fixed DC bus. The topology offers high step-up capabilities and a nearly constant efficiency over a wide operating range. A reduced sensor maximum power point tracking (MPPT) controller is developed for the converter to maximize energy harvesting of the PV panels. The reduced sensor algorithm can be generally applied to the class of converters employing pulse frequency modulation control. A ZigBee wireless communication system is implemented to provide advanced control, monitoring and protection features. A testbench for a low cost 500 $W$ smart microconverter is designed and implemented, demonstrating the viability of the system architecture.