State-of-the-art speech-recognition systems can successfully perform simple tasks in real-time on most computers, when the tasks are performed in controlled and noiseree environments. However, current algorithms and processors are not yet powerful enough for real-time large-vocabulary conversational speech recognition in noisy, real-world environments. Parallel processing can improve the real-time performance of speech recognition systems and increase their applicability, and developing an effective approach to parallelization is especially important given the recent trend toward multi-core processor design. In this dissertation, we introduce methods for parallelizing a single-pass across-word n-gram lexical-tree based Viterbi recognizer, which is the most popular architecture for Viterbi-based large vocabulary continuous speech recognition. We parallelize two different open-source implementations of such a recognizer, one developed at Mississippi State University and the other developed at Rheinisch-Westfalische Technische Hochschule University in Germany. We describe three methods for parallelization. The first, called parallel fast likelihood computation, parallelizes likelihood computations by decomposing mixtures among CPU cores, so that each core computes the likelihood of the set of mixtures allocated to it. A second method, lexical-tree division, parallelizes the search management component of a speech recognizer by dividing the lexical tree among the cores. A third and alternative method for parallelizing the search-management component of a speech recognizer, called lexical-tree copies decomposition, dynamically distributes the active lexical-tree copies among the cores. All parallelization methods were tested on two and four cores of an Intel Core2 Quad processor and significantly improved real-time performance. Several challenges for parallelizing a lexical-tree based Viterbi speech recognizer are also identified and discussed.

No description available.

No description available.

No description available.

No description available.

In today's data-driven society, data holds immense value and is sought after across various domains, including the realm of science. Materials science, in particular, relies heavily on data acquisition and analysis to further advancements in the field. In this study, data for an alloy database is generated through high-throughput calculations, serving as a valuable resource for investigating the effects of vacancy concentration on the structural and mechanical properties of (TM)0.5Al0.5N1-x random alloys. The alloys, comprised of Ta or Nb (TM), exhibit promising potential for diverse applications such as cutting tool, electrical and optical devices, etc. To accurately represent the average behavior of real random alloys, special quasirandom structures (SQS) are utilized, and the short-range order (SRO) parameters are analyzed using the ATAT code. A two-level computational approach with different convergence criteria is used to investigate the influence of vacancy concentration on alloys. This approach utilizes the high throughput toolkit (httk) software package in conjunction with VASP calculations. The calculations are performed on supercomputer resources provided by the National Academic Infrastructure for Supercomputing in Sweden (NAISS). The elastic constants of the alloys are calculated using the httk software, providing insights into their mechanical properties. The findings highlight the substantial influence of vacancy concentration on the structural and mechanical behavior of (TM)0.5Al0.5N1-x random alloys.

The development of chiral organometallics for asymmetric synthesis is a topic of significant research in the recent past. The most studied in this class are the chiral organolithium reagents with many reported examples. The primary focus of our research is the development of C_α-chiral Grignard reagents, where the metal bearing α-carbon is the sole source of chirality. Examples of such Grignard reagents are rare owing to the problems associated with their synthesis, and their low configurational stability. We have studied these problems in three different modules of this project. Reactions of 1-magnesio-2,2-diphenyl-cyclopropylcarbonitrile with carbon electrophiles are first attempted in order to expand the utility of this configurationally stable C_α-chiral Grignard reagent in asymmetric synthesis. This reagent has been shown to be non-reactive towards carbon electrophiles at low temperatures. Consequently, we attempt to enhance the reactivity of this compound through two different approaches, Lewis-base activation and the "ate-complex" generation. The Magnesium/Halogen (Mg/X) exchange reactions have been shown to be extremely useful in the synthesis of complex Aryl, alkenyl (sp²) and alkynyl (sp) Grignard reagents. Examples of Mg/X exchange reactions of Alkyl (sp³) halides are, however, rare. Even more rare are such examples with secondary and tertiary alkyl halides, justifying the relative paucity of chiral Grignard reagents. In this module of our project, we study the Mg/X exchange reactions on secondary alkyl halides possessing a γ-hydroxyl group, as an internal activator for such Mg/X exchange reactions. Enantiomerization pathways of chiral organolithium compounds have been widely studied. However, few such studies have been performed on chiral Grignard reagents. In this module of the project, we studied the solvent assisted enantiomerization mechanism of the C_α-chiral 1-magnesio-2,2-diphenyl-cyclopropylcarbonitrile. Rate constant for the enantiomerization of this compound was measured in three different ethereal solvents to study the effect of solvent on the configurational stability. Finally, the order of the enantiomerization process with respect to [Et₂O] was studied in order to predict the mechanism of this process in Et₂O solvent. Our kinetic studies on the enantiomerization process provided us with a definitive picture for the enantiomerization of the C_α-chiral 1-magnesio-2,2-diphenyl-cyclopropylcarbonitrile, where solvation of the Grignard reagent preceded an ion-pair separation step which eventually lead to enantiomerization of the Grignard species. However, the precise structure of all the involved solvated intermediates could not be determined as kinetics was not able to distinguish between these intermediates. We next performed computational calculations to study the effect of solvation on the analogous 1-magnesio-cyclopropylcarbonitrile in order to address the unanswered questions from our kinetic studies. / Ph. D.

Numerical information is ubiquitous in modern life. The prevalence of numerical information in the marketplace necessitates understanding how consumers handle and interpret that information, for both theoretical and practical reasons. Past research has largely focused on consumers’ encoding of numbers, calculative limitations, and usage of heuristics. This dissertation will contribute to this burgeoning literature in several ways. First, I identify a general tendency in how consumers calculate ratios based on an intuitive model of division. Specifically, consumers tend to divide larger numbers by smaller numbers. The intuitive model of division has marketing implications for both consumers’ evaluations of quantity offers and sensitivities to promotions. Next, I examine how consumers draw inferences from distributional information. In contrast to the assumption that consumers utilize means to assess central tendency, I demonstrate that consumers use the modal response to judge what is typical, with implications for consumers’ inferences about product ratings and other social distributions. / PHD / Numerical information is ubiquitous in modern life. The prevalence of numerical information in the marketplace necessitates understanding how consumers handle and interpret that information, for both theoretical and practical reasons. Past research has largely focused on how consumers’ mentally perceive numbers, how difficult it is to engage in calculation, and usage of mental shortcuts. This dissertation will contribute to this burgeoning literature in several ways. First, I identify a general tendency in how consumers calculate ratios based on an intuitive model of division. Specifically, consumers tend to divide larger numbers by smaller numbers. The intuitive model of division has marketing implications for both consumers’ evaluations of quantity offers and sensitivities to promotions. Next, I examine how consumers draw inferences from distributional information. In contrast to the assumption that consumers utilize means to assess central tendency, I demonstrate that consumers use the modal response to judge what is typical, with implications for consumers’ inferences about product ratings and other social distributions.

The amino acid sequence of a protein encodes its folding, the reaction by which a peptide self-assembles into its native functional shape. A folded protein will then go on to carry on biological functions such as ligand binding, signaling, mechanical functions, or biomolecular catalysis. While much experimental work has been done to elucidate protein structure and conformational dynamics, Molecular Dynamics (MD) simulations have been necessary to provide atomic-level details of biomolecular dynamics. A challenge in using MD is the the large computational cost required to reach biologically relevant timescales when integration steps are typically limited to a few femtoseconds. To address this challenge, specialized hardware such as the ANTON supercomputer or distributed computing platforms like Folding@home can be used to collect milliseconds of aggregate trajectory data. From these datasets, kinetic network models called Markov state models (MSMs) can be constructed to infer long timescale dynamics from ensembles of short trajectories. These models can be analyzed in a human interpretable way and make physics-based connections to experimental observables. This dissertation describes how we have used MD simulations and MSMs to model protein folding reactions and protein-protein binding reactions to better understand mutation effects on our systems of interest. The first chapter of this dissertation describes MD simulations FOXO1 FKH domain folding, which we used MSMs to characterize at atomic resolution. To predict how mutations found in diffuse large B-cell lymphoma (DLBCL) cell lines effect protein stability, we developed an MSM-based hydrophobic free energy of transfer (HT) model to estimate mutation effects. Our HT model results agree better with experiment than other state-of-the-art computational methods. Chapter two describes how we have used approximately 43000 relative binding free energy calculations via the expanded ensemble (EE) method to perform in silico site saturation mutagenesis on miniprotein binders to the highly conserved influenza A H1 hemagglutinin stem region (HA2) de novo designed by the Baker [66]. These miniproteins were selected through an exhaustive design process with iterations of computational design, high throughput affinity screens, and site saturation mutagenesis. We compare our EE SSM method with inferred relative affinities from Chevalier et al.[66], as well as with the state-of-the-art Rosetta method Flex ddG. While Flex ddG predictions are more accurate on average, they are highly conservative. In contrast, EE predictions can better classify stabilizing and destabilizing mutations. We also use a Shannon entropy based method to identify residue positions that are more susceptible to mutation. This work suggests that simulation-based free energy methods can provide predictive information for in silico affinity maturation of designed miniproteins, with many feasible improvements to the efficiency and accuracy within reach. In the final chapter, we atttempt to model the complete binding reactions of the 6 miniproteins mentioned above. We used unbiased simulations to build standard msms and , in combination with biased simulations, multiensemble markov models (MEMMs) of binding for each wild type and affinity matured pair. The unbiased MSMs show that the affinity matured miniproteins prefer different bound states than the wild type miniproteins. Additionally, they provide physically realistic k_on_s and a macroscopic 3 state pathway through an encounter complex. We characterize each of those states and use an contact map based structural similarity index measure (SSIM) and residue-wise Kullback-Leibler divergence method to better understand the differences in the bound states between affinity matured and wild type construct. Interestingly, while our biased simulations do see unbinding transitions, in estimating the MEMM, they overweight the unbinding reaction and unbound state, leading to models that do not make physical sense. This demonstrates that more sensitive enhances sampling techniques may be necessary for building MEMMs. The final two chapters of this dissertation present new methodologies for computational protein design, making great strides towards a dynamic understanding of how proteins bind their targets and how mutations effect those reactions. / Chemistry

No description available.