588 research outputs found
First-principles study of electromechanical and polar properties in perovskite oxides and half-Heusler semiconductors
No full textThis thesis discusses electromechanical and polar properties in two well-known classes of materials, perovskite oxides and half-Heusler compounds, using first-principles calculations. Certain features of the ab initio codes, such as the capability to calculate polarization based on the modern theory of polarization, or to apply a finite electric field, are central to the problems presented in this thesis. Hence these formalisms are discussed, following a brief opening section on the basic methodology of density-functional theory. The first problem presented in this thesis concerns the nonlinear piezoelectric response of ferroelectric PbTiO₃ for the case of a polarization-enhancing electric field applied along the tetragonal axis. The dependence of the c/a ratio on electric field is found to be almost linear in the range up to 500 MV/m, contrary to what expected from Landau-Devonshire theory, but in qualitative agreement with a recent experiment. In the second problem we study the energy landscape and ferroelectric states of double perovskites of the form AA'BB'O₆ in which the atoms on both the A and B sites are arranged in rock-salt order. If a ferroelectric instability occurs, the energy landscape will tend to have minima with the polarization along tetrahedral directions, leading to a rhombohedral phase, or along Cartesian directions, leading to an orthorhombic phase. We are not aware of compounds naturally occurring in this structure, although they might be synthesized experimentally. In the final problem, we use a first-principles rational-design approach to search a large materials family, half-Heusler compounds to identify semiconductors, and then compute their piezoelectric properties. This previously-unrecognized class of piezoelectrics may benefit greatly from calculations such as those presented here. Our work may provide guidance for experimental verification of existing compounds and for the experimental realization of other potential candidates.Ph. D.Includes bibliographical referencesIncludes vitaby Anindya Ro
Bayesian Inference for High-dimensional Time Series with a Directed Acyclic Graphical Structure
No full textIn multivariate time series analysis, understanding the underlying causal relationships among variables is often of interest for various applications. Directed acyclic graphs (DAGs) provide a powerful framework for representing causal dependencies. This paper proposes a novel Bayesian approach for modeling multivariate time series where conditional independencies and causal structure are encoded by a DAG. The proposed model allows structural properties such as stationarity to be easily accommodated, and further does not assume any pre-specified parent-child ordering. Given the application, we further extend the model for matrix-variate time series. We take a Bayesian approach to inference, and a “projection-posterior” based efficient computational algorithm is developed. The posterior convergence properties of the proposed method are established along with two identifiability results for the unrestricted structural equation models. The utility of the proposed method is demonstrated through simulation studies and real data analysis.The authors would like to thank the National Science Foundation for the Collaborative Research Grants DMS-2210280 (for Subhashis Ghosal), DMS-2210281 (for Anindya Roy), and DMS-2210282
(for Arkaprava Roy).https://arxiv.org/html/2503.23563v
Bayesian Inference for High-dimensional Time Series by Latent Process Modeling
No full textTime series data arising in many applications nowadays are high-dimensional. A large number of parameters describe features of these time series. We propose a novel approach to modeling a high-dimensional time series through several independent univariate time series, which are then orthogonally rotated and sparsely linearly transformed. With this approach, any specified intrinsic relations among component time series given by a graphical structure can be maintained at all time snapshots. We call the resulting process an Orthogonally-rotated Univariate Time series (OUT). Key structural properties of time series such as stationarity and causality can be easily accommodated in the OUT model. For Bayesian inference, we put suitable prior distributions on the spectral densities of the independent latent times series, the orthogonal rotation matrix, and the common precision matrix of the component times series at every time point. A likelihood is constructed using the Whittle approximation for univariate latent time series. An efficient Markov Chain Monte Carlo (MCMC) algorithm is developed for posterior computation. We study the convergence of the pseudo-posterior distribution based on the Whittle likelihood for the model's parameters upon developing a new general posterior convergence theorem for pseudo-posteriors. We find that the posterior contraction rate for independent observations essentially prevails in the OUT model under very mild conditions on the temporal dependence described in terms of the smoothness of the corresponding spectral densities. Through a simulation study, we compare the accuracy of estimating the parameters and identifying the graphical structure with other approaches. We apply the proposed methodology to analyze a dataset on different industrial components of the US gross domestic product between 2010 and 2019 and predict future observations.The authors would like to thank the National Science Foundation collaboative research grants DMS-2210280 (Subhashis Ghosal) / 2210281 (Anindya Roy) / 2210282 (Arkaprava Roy).http://arxiv.org/abs/2403.0491
Bayesian Analysis of Singly Imputed Synthetic Data under the Multivariate Normal Model
No full textWe develop appropriate Bayesian procedures to draw inference about the parameters under a multivariate normal model based on synthetic data. We consider two standard forms of synthetic data, generated under plug in sampling method and posterior predictive sampling method. In addition to point estimates of the mean vector and dispersion matrix, Bayesian credible sets for the mean vector and the generalized variance are also provided under both the scenarios. The analysis in the case when some (partial) features are sensitive and need to be hidden is also briey indicatedAbhishek Guin is thankful to the Department of Mathematics and Statistics, UMBC, for providing financial support for his graduate studies. Anindya Roy and Bimal Sinha are thankful to Dr. Tommy Wright, Chief of CSRM/Census Bureau for his support and encouragement. Our sincere thanks are due to Professor Gaurisankar Datta of the University of Georgia for
his critical reading of the manuscript and making some excellent suggestions which led to improved presentation.https://www.banglajol.info/index.php/ijss/article/view/7011
Analysis of the long-read sequencing data using computational tools confirms the presence of 5-methylcytosine in the Saccharomyces cerevisiae genome
No full textModification of DNA bases plays important roles in the epigenetic regulation of eukaryotic gene expression. Among the diferent
types of DNA methylation, 5-methylcytosine (5mC) is common in higher eukaryotes. Although bisulfite sequencing is the established detection method for this modification, newer methods, such as Oxford nanopore sequencing, have been developed as
quick and reliable alternatives. An earlier study using sensitive liquid chromatography tandem mass spectrometry (LC-MS/MS)
indicated the presence of 5mC at very low concentration in Saccharomyces cerevisiae. More recently, a comprehensive study of
the yeast genome found 40 5mC sites using the computational tool Nanopolish on nanopore sequencing output raw data. In the
present study, we are trying to validate the prediction of the 5mC modifications in yeast with Nanopolish and two other nanopore software tools, Tombo and DeepSignal. Using publicly available genome sequencing data, we compared the open-access
computational tools, including Tombo, Nanopolish and DeepSignal, for predicting 5mC. Our results suggest that these tools
are indeed capable of predicting DNA 5mC modifications at a specific location from Oxford nanopore sequencing data. We also
predicted that 5mC present in the S. cerevisiae genome might be located predominantly at the RDN locus of chromosome 1
Flexible Joint Models For Screening Studies
No full textIn this dissertation, we study and develop statistical methods to jointly analyze longitudinal biomarkers with time-to-event outcomes motivated by risk assessment in cancer screening. Cancer screening studies collect longitudinal biopsies to allow the identification of additional longitudinal biomarker measurements for risk stratification. However, these studies present several challenges for current joint modeling approaches. In Chapter 2, we develop a dynamic risk prediction approach that links both continuous and binary biomarkers to the interval-censored precancer outcome with shared high dimensional random effects. A cancer screening dataset shows improved risk stratification compared to univariate joint models. In Chapter 3, we develop a latent health model for at-risk patients that can both deteriorate to case status and improve to low-risk status. We link the change in the health process to a longitudinal biomarker whose trajectory can change based on the event. We see that treating individuals who become risk-free as right censored and ignoring the event's impact on the biomarker trajectory can result in significantly biased risk estimates. In Chapter 4, we compare three common approaches to identify longitudinal biomarkers associated with survival outcomes: joint models, conditional models, and time-dependent Cox models. We use simulations to evaluate how well the methods identify and distinguish biomarkers that are useful for long-term risk assessment and early detection
Lexical speaker identification in TV shows
No full textThe final publication is available at https://link.springer.com/article/10.1007/s11042-014-1940-3International audienceIt is possible to use lexical information extracted from speech transcripts for speaker identification (SID), either on its own or to improve the performance of standard cepstral-based SID systems upon fusion. This was established before typically using isolated speech from single speakers (NIST SRE corpora, parliamentary speeches). On the contrary, this work applies lexical approaches for SID on a different type of data. It uses the REPERE corpus consisting of unsegmented multiparty conversations, mostly debates, discussions and Q&A sessions from TV shows. It is hypothesized that people give out clues to their identity when speaking in such settings which this work aims to exploit. The impact on SID performance of the diarization front-end required to pre-process the unsegmented data is also measured. Four lexical SID approaches are studied in this work, including TFIDF, BM25 and LDA-based topic modeling. Results are analysed in terms of TV shows and speaker roles. Lexical approaches achieve low error rates for certain speaker roles such as anchors and journalists, sometimes lower than a standard cepstral-based Gaussian Supervector-Support Vector Machine (GSV-SVM) system. Also, in certain cases, the lexical system shows modest improvement over the cepstral-based system performance using score-level sum fusion. To highlight the potential of using lexical information not just to improve upon cepstral-based SID systems but as an independent approach in its own right, initial studies on crossmedia SID is briefly reported. Instead of using 2 Anindya Roy et al. speech data as all cepstral systems require, this approach uses Wikipedia texts to train lexical speaker models which are then tested on speech transcripts to identify speakers
Efficiency Tradeoffs in Estimating the Linear Trend Plus Noise Model
No full textThis paper presents the results of a Monte Carlo comparison of feasible GLS estimators of the trend parameter in the linear trend plus noise model, where the noise component may or may not be a unit root process. We include an FGLS estimator that estimates the noise component using a median-unbiased estimator.
Efficient Designs for Cyclic Data Generated by Multivariate Harmonic Mixed Models
No full textThe primary focus of this dissertations is to design efficient data collection schemes (including pooling bio-specimens, subsampling and combining laboratory and monitor data) for longitudinal studies, particularly in medical fields where repeated cyclical patterns over time are expected in the model response. Motivation of the work comes from the need for efficient designs in the expensive medical studies where collection of data at regular frequency may not be feasible due to budget constraints. Throughout the dissertations we use multivariate mixed-effects exponential harmonic models to mimic the cyclical evolution of biomarkers and individual varia-tion. We propose to optimize objective criteria related to efficiency of parameter estimation, such as D-optimality, with respect to pooling and sampling design. Due to analytical intractability, we mostly rely on numerical optimization to devise sensible schemes for finding effective study designs. We apply the optimal design to the BioCycle data. We propose a combined model to use both monitor and laboratory data to improve the efficiency of the design. The laboratory measurements are more accurate than the monitor ones. Because the accuracy of the laboratory measurements is higher than that of the monitor measurements, when we combine the monitor and laboratory measurements, the efficiency is improved compared to using just monitor data. The efficiency of optimal design based on combined measurements close to that of designs with similar cost and using only laboratory measurements, and much higher than those using equally spaced laboratory measurements. We apply the combined design to the EAGeR study. The combined design improves the efficiency compared to using only laboratory measurements for the luteinizing hormone (LH). For designs using pooled bio-specimens, we add a novel aspect by incorporating a fatigue model to the data preparation process. We explicitly model the heteroscedasticity of the error of the pooling volume under pooling design using Hill type models. We study the efficiency for small and large pooling procedures. We investigate the evolution of the relative efficiency over pooling size and discuss the relative gain from pooling in terms of cost. We apply our approach to the EAGeR data. The results indicate definitive improvement that are obtained from pooling designs for longitudinal studies with cyclical mean response
- …
