A numerical scheme to simulate time-domain motion responses of multibody floating systems has been successfully proposed. This scheme is integrated into a time-domain simulation tool, with fully coupled hydrodynamic coefficients obtained from the hydrodynamic software - WAMIT which solves the Boundary Value Problem (BVP). The equations of motion are transformed into standard state-space format, using the constant coefficient approximation and the impulse response function method. Thus the Ordinary Differential Equation (ODE) solvers in MATLAB can be directly employed. The time-domain responses of a single spar at sea are initially obtained. The optimal Linear Quadratic Regulator (LQR) controller is further applied to this single spar, by assuming that the Dynamic Positioning (DP) system can provide the optimized thruster forces. Various factors that affect the controlling efficiency, e.g., the time steps ∆τ and ∆t, the weighting factors(Q,R), are further investigated in detail. Next, a two-body floating system is studied. The response amplitude operators (RAOs) of each body are calculated and compared with the single body case. Then the effects of the body-to-body interaction coefficients on the time-domain responses are further investigated. Moreover, the mean drift force is incorporated in the DP system to further mitigate the motion responses of each body. Finally, this tool is extended to a three-body floating system, with the relative motions between them derived.

<p>Latent variable models are widely used in modern data science for both statistic and dynamic data. This thesis focuses on large-scale latent variable models formulated for time series data and static network data. The former refers to the state space model for dynamic systems, which models the evolution of latent state variables and the relationship between the latent state variables and observations. The latter refers to a network decoder model, which map a large network into a low-dimensional space of latent embedding vectors. Both problems can be solved by adaptive stochastic gradient Markov chain Monte Carlo (MCMC), which allows us to simulate the latent variables and estimate the model parameters in a simultaneous manner and thus facilitates the down-stream statistical inference from the data. </p> <p><br></p> <p>For the state space model, its challenge is on inference for high-dimensional, large scale and long series data. The existing algorithms, such as particle filter or sequential importance sampler, do not scale well to the dimension of the system and the sample size of the dataset, and often suffers from the sample degeneracy issue for long series data. To address the issue, the thesis proposes the stochastic approximation Langevinized ensemble Kalman filter (SA-LEnKF) for jointly estimating the states and unknown parameters of the dynamic system, where the parameters are estimated on the fly based on the state variables simulated by the LEnKF under the framework of stochastic approximation MCMC. Under mild conditions, we prove its consistency in parameter estimation and ergodicity in state variable simulations. The proposed algorithm can be used in uncertainty quantification for long series, large scale, and high-dimensional dynamic systems. Numerical results on simulated datasets and large real-world datasets indicate its superiority over the existing algorithms, and its great potential in statistical analysis of complex dynamic systems encountered in modern data science. </p> <p><br></p> <p>For the network embedding problem, an appropriate embedding dimension is hard to determine under the theoretical framework of the existing methods, where the embedding dimension is often considered as a tunable hyperparameter or a choice of common practice. The thesis proposes a novel network embedding method with a built-in mechanism for embedding dimension selection. The basic idea is to treat the embedding vectors as the latent inputs for a deep neural network (DNN) model. Then by an adaptive stochastic gradient MCMC algorithm, we can simulate of the embedding vectors and estimate the parameters of the DNN model in a simultaneous manner. By the theory of sparse deep learning, the embedding dimension can be determined via imposing an appropriate sparsity penalty on the DNN model. Experiments on real-world networks show that our method can perform dimension selection in network embedding and meanwhile preserve network structures. </p> <p><br></p>

Business to Business (B2B) firms spend significant resources managing close relationships with their customers, yet there is limited understanding of how the customers perceive the relationship based on the customer management efforts initiated by the firm. Specifically, studies on how firms communicate different values to B2B customers and how they perceive the values the firm offers by consistently evaluating the direct marketing communication which ultimately affect their buying behaviors have been largely overlooked. Typically, the direct marketing communication efforts are geared towards explicitly featuring economic values or relational values. To implement an effective communication strategy catering to customers’ preferences, firms should understand how these organizational marketing communications dynamically influence the perceived importance of different values offered by the firm. Therefore, using data from a Fortune 500 B2B service firm and employing a content analysis and a robust econometric model, we find that (i) the effect of economic and relational marketing communication on customer purchase behavior vary by customers and change overtime (ii) the latent stock variable of direct marketing communication affect the customer purchase behaviors and (iii) the evolution of customers’ perceived importance can be recovered using the transaction data. Overall, we provide a marketing resource reallocation strategy that enables marketers to customize marketing communication and improve a firm’s financial performance.

The study was aimed to provide information on local biomass development during crop growth using ground based optical sensors and to incorporate the local crop status to a crop growth simulation model to improve understanding on inherent variability of crop field. The experiment was conducted in a farmer’s field located near Princeton in Caldwell County, Western Kentucky. Data collection on soil, crop and weather variables was carried out in the farm from 2006 December to 2008 October. During this period corn (Zea mays L.) and winter wheat (Triticum sp) were grown in the field. A 450 m long representative transect across the field consisting of 45 locations each separated by 10 m was selected for the study. Soil water content was measured in a biweekly interval during crop growth from these locations. Measurements on crop growth parameters such as plant height, tiller count, biomass and grain yield were able to show spatial variability in crop biomass and grain yield production. Crop reflectance measured at important crop growth stages. Soil water sensing capacitance probe was site specifically calibrated for each soil depth in each location. Various vegetation indices were calculated as proxy variables of crop growth. Inherent soil properties such as soil texture and elevation were found playing a major role in influencing spatial variability in crop yield mainly by affecting soil water storage. Temporal persistence of spatial patterns in soil water storage was not observed. Optimum spatial correlation structure was observed between crop growth parameters and optical sensor measurements collected early in the season and aggregated at 2*2 m2 sampling area. NDVI, soil texture, soil water storage and different crop growth parameters were helpful in explaining the spatial processes that influence grain yield and biomass using state space analysis. DSSAT was fairly sensitive to reflect site specific inputs on soil variability in crop production.

<p>By using AR/ARX-models on data generated by a nonlinear differential equation system representing a model for the cell-cycle control system in budding yeast, the interactions among proteins and thereby also to some extent the genes, are sought. A method consisting of graphical analysis of differences between estimates from two local linear models seems to make it possible to separate a set of linear equations from the nonlinear system. By comparing the properties of the estimations in the linear equations a set of approximate equations corresponding well to the real ones are found. </p><p>A NARX model is tested on the same system to see whether it is possible to find the dependencies in one of the nonlinear differential equations. This approach did, for the choice of model, not work.</p>

Most macroeconomic activity series such as Swedish GDP growth are collected quarterly while an important proportion of time series are recorded at a higher frequency. Thus, policy and business decision makers are often confront with the problems of forecasting and assessing current business and economy state via incomplete statistical data due to publication lags. In this paper, we survey a few general methods and examine different models for mixed frequency issues. We mainly compare mixed data sampling regression (MIDAS) and state space dynamic factor model (SS-DFM) by the comparison experiments forecasting Swedish GDP growth with various economic indicators. We find that single-indicator MIDAS is a wise choice when the explanatory variable is coincident with the target series; that an AR term enables MIDAS more promising since it considers autoregressive behaviour of the target series and makes the dynamic construction more flexible; that SS-DFM and M-MIDAS are the most outstanding models and M-MIDAS dominates undoubtedly at short horizons up to 6 months, whereas SS-DFM is more reliable at long predictive horizons. And finally we conclude that there is no perfect winner because each model can dominate in a special situation.

By using AR/ARX-models on data generated by a nonlinear differential equation system representing a model for the cell-cycle control system in budding yeast, the interactions among proteins and thereby also to some extent the genes, are sought. A method consisting of graphical analysis of differences between estimates from two local linear models seems to make it possible to separate a set of linear equations from the nonlinear system. By comparing the properties of the estimations in the linear equations a set of approximate equations corresponding well to the real ones are found. A NARX model is tested on the same system to see whether it is possible to find the dependencies in one of the nonlinear differential equations. This approach did, for the choice of model, not work.

Spatial variability of soil properties complicates the understanding of water and solute transport at the field scale. This study evaluated the impact of land use, soil surface roughness, and rainfall characteristics on water transport and Br- leaching under field conditions by means of a new experimental design employing scale-dependent treatment distribution. On a transect with two land use systems, i.e., cropland and grassland, rainfall intensity and the time delay between Br- application and subsequent rainfall were arranged in a periodically repetitive pattern at two different scales. Both scales were distinct from the scale of surface roughness as described by elevation variance. Nests of tensiometers and suction probes were installed at 1-m intervals along the transect to monitor matric potentials and Br- concentrations at different depths, respectively. After rainfall simulation, soil samples were collected at every 0.5 m horizontal distance in 10 cm vertical increments down to 1 m depth for Br- analysis. Soil Br- concentration was more evenly distributed with soil depth and leached deeper in grassland than cropland, owing to vertically continuous macropores that supported preferential flow. Frequency-domain analysis and autoregressive state-space approach revealed that the dominant factors controlling Br- leaching varied with depth. In shallow layers, land use was the main driving force for Br- distribution. Beyond that, the spatial pattern of Br- was mostly affected by rainfall characteristics. Below 40 cm, the horizontal distribution of Br- was dominated by soil texture and to a smaller extent by rainfall intensity. Bromide concentrations obtained from soil solution samples that were collected through suction probes showed similar results with respect to the influence of rainfall intensity. The spatial variation scale of temporal matric potential change varied with both time and depth, corresponding to different boundary condition scales. Matric potential change in some cases, reflected the impact of soil properties other than the boundary conditions investigated, such as hydraulic conductivity, contributing to the scale-variant behavior of Br- leaching. These findings suggest the applicability of scale-dependent treatment distribution in designing field experiments and also hold important implications for agricultural management and hydrological modelling.

This dissertation studies various issues related to regime switching and DSGE models. The methods developed are used to study U.S. business cycles. Chapter one considers and derives the limit distributions of likelihood ratio based tests for Markov regime switching in multiple parameters in the context of a general class of nonlinear models. The analysis simultaneously addresses three difficulties: (1) some nuisance parameters are unidentified under the null hypothesis, (2) the null hypothesis yields a local optimum, and (3) the conditional regime probabilities follow stochastic processes that can only be represented recursively. When applied to US quarterly real GDP growth rates, the tests suggest strong evidence favoring the regime switching specification over a range of sample periods. Chapter two develops a modified likelihood ratio (MLR) test to detect regime switching in state space models. I apply the filtering algorithm introduced in Gordon and Smith (1988) to construct a modified likelihood function under the alternative hypothesis of two regimes and I extend the analysis in Chapter one to establish the asymptotic distribution of the MLR statistic under the null hypothesis of a single regime. I also apply the test to a simple model of the U.S. unemployment rate. This contribution is the first to develop a test based on the likelihood ratio principle to detect regime switching in state space models. The final chapter estimates a search and matching model of the aggregate labor market with sticky price and staggered wage negotiation. It starts with a partial equilibrium search and matching model and expands into a general equilibrium model with sticky price and staggered wage. I study the quantitative implications of the model. The results show that (1) the price stickiness and staggered wage structure are quantitatively important for the search and matching model of the aggregate labor market; (2) relatively high outside option payments to the workers, such as unemployment insurance payments, are needed to match the data; and (3) workers have lower bargaining power relative to firms, which contrasts with the assumption in the literature that workers and firms share equally the surplus generated from their employment relationship.

abstract: Transfer learning refers to statistical machine learning methods that integrate the knowledge of one domain (source domain) and the data of another domain (target domain) in an appropriate way, in order to develop a model for the target domain that is better than a model using the data of the target domain alone. Transfer learning emerged because classic machine learning, when used to model different domains, has to take on one of two mechanical approaches. That is, it will either assume the data distributions of the different domains to be the same and thereby developing one model that fits all, or develop one model for each domain independently. Transfer learning, on the other hand, aims to mitigate the limitations of the two approaches by accounting for both the similarity and specificity of related domains. The objective of my dissertation research is to develop new transfer learning methods and demonstrate the utility of the methods in real-world applications. Specifically, in my methodological development, I focus on two different transfer learning scenarios: spatial transfer learning across different domains and temporal transfer learning along time in the same domain. Furthermore, I apply the proposed spatial transfer learning approach to modeling of degenerate biological systems.Degeneracy is a well-known characteristic, widely-existing in many biological systems, and contributes to the heterogeneity, complexity, and robustness of biological systems. In particular, I study the application of one degenerate biological system which is to use transcription factor (TF) binding sites to predict gene expression across multiple cell lines. Also, I apply the proposed temporal transfer learning approach to change detection of dynamic network data. Change detection is a classic research area in Statistical Process Control (SPC), but change detection in network data has been limited studied. I integrate the temporal transfer learning method called the Network State Space Model (NSSM) and SPC and formulate the problem of change detection from dynamic networks into a covariance monitoring problem. I demonstrate the performance of the NSSM in change detection of dynamic social networks. / Dissertation/Thesis / Doctoral Dissertation Industrial Engineering 2015