SRC 2008

Home | Program | Speakers | Hotels | Travel Info | Contact | Other Conferences

Abstracts for the 2008 Spring Research Conference


Speaker: Peter Bickel
Title: Low effective dimension in models or data: a key to high dimensional inference?
Abstract: Theoretical analysis seems to suggest that standard problems such as estimating a function of high dimensional variables with noisy data (regression or classification) should be impossible without detailed detailed knowledge or absurdly large amounts of data.Nevertheless, algorithms to perform classification of images or other high dimensional objects are remarkably successful. The generally held explanation is the presence of sparsity/low dimensional structure. I'll discuss analytically and with examples why this may be right.


Speaker: Howard Burkom
Title: Bridging the Gap Between Statistical Research and Public Health Practice in Infectious Disease Surveillance
Abstract: Recent years have seen considerable research in adaptation of methods in statistics, machine learning, data mining, and related fields for the evolving application of advanced disease surveillance. This presentation addresses a mismatch between some developed methodology and the needs of the health monitoring community. In preparation for a recent roundtable discussion convened by the U.S. Medicine Institute, practicing epidemiologists were polled regarding the utility of their automated systems. Responses indicated that users are getting benefits from these systems, but often not the benefits conceived by developers. Another finding was that "situational awareness" is meaningful and multifaceted, varying according to the objectives and purview of the monitoring institutions. Many of these facets bring statistical challenges. Meeting these challenges requires technology adaptations and combinations, but only after surveillance needs and the associated data environments are well understood. This talk will elaborate, enumerate some of these challenges, and provide illustration using one example drawn from a surveillance dataset.


Speaker: Ching-Shui Cheng
Title: Multi-stratum Fractional Factorial Designs
Abstract: There has been a lot of recent work on industrial experiments with multiple error terms including, e.g., split-plot designs, blocked split-plot designs, strip-plot designs, etc. In this talk, I will present a general and unified approach to the selection and construction of such designs.


Speaker: Tirthankar Dasgupta
Title:Sequential Minimum Energy Designs for Synthesis of Nanostructures
Abstract: The talk will discuss the development of an experimental design methodology, tailor-made to address the unique phenomena associated with nanostructure synthesis. A sequential space filling design called Sequential Minimum Energy Design (SMED) is proposed for exploring best process conditions for synthesis of nanowires. The SMED is a novel approach to generate designs that are model independent, can quickly carve out regions with no observable nanostructure morphology, allow for the exploration of complex response surfaces, and can be used for sequential experimentation. A unique feature of this technique lies in the fact that it originates from a combination of statistical theory and fundamental laws of physics. The basic idea has been developed into a practically implementable algorithm for deterministic functions, and guidelines for choosing the parameters of the design have been proposed. Performance of the algorithm has been studied using experimental data on nanowire synthesis as well as standard two dimensional and higher dimensional test functions. A modification of the algorithm based on non-parametric smoothing has been proposed for random functions.


Speaker: Xinwei Deng
Title: A Statistical Approach to Quantifying the Elastic Deformation of Nanomaterials
Abstract: Accurate estimation of elastic modulus of certain nanomaterials such as Zinc Oxide nanobelt is important in many applications. A recently proposed approach was to estimate elastic modulus from a force-deflection model based on the continuous scan of a nanobelt using an Atomic Force Microscope tip at different contact forces. However, the nanobelt may have some initial bending and it may shift or deform during measurement leading to bias in the estimation. In this work we propose a statistical model to account for these various possible errors. The proposed approach can automatically detect and remove the systematic errors and therefore, can give an accurate and precise estimate of the elastic modulus. The advantages of the approach are demonstrated through the application on several data sets.


Speaker: Marco Ferreira
Title: Gaussian multiscale spatio-temporal models
Abstract: We develop a new class of multiscale spatio-temporal models for Gaussian areal data. Our framework decomposes the spatio-temporal observations and underlying process into several scales of resolution. Under this decomposition the model evolves the multiscale coefficients through time with structural state-space equations. The multiscale decomposition considered here, which includes wavelet decompositions as particular case, is able to accommodate irregular grids and heteroscedastic errors. The multiscale spatio-temporal framework we develop has several salient attributes. First, the multiscale decomposition leads to an extremely efficient divide-and-conquer estimation algorithm. Second, the multiscale coefficients have an interpretation of their own; thus, the multiscale spatio-temporal framework may offer new insight on understudied multiscale aspects of spatio-temporal observations. Finally, deterministic relationships between different resolution levels are automatically respected for both observations, the latent process, and the estimated latent process. We illustrate the use of our multiscale framework with an analysis of a spatio-temporal dataset on agriculture production in the state of Espirito Santo, Brazil (Joint work with Scott Holan and Adelmo Bertolde)


Speaker: James Glimm
Title: Quantified Margins of Uncertainty: Error bars for Simulation Based Design
Abstract: For a number of reasons, advanced engineering design is often based, in the first instance, on simulation. The Edisonian method of build and test is losing in a contest of practicality, some times due to time to market pressures, sometimes due to cost issues, and sometimes due to the infeasibility of experimental design methods. A clear sign of the importance given to simulation based design decisions is the increasing importance assigned to simulation error bars. Campaigns for verification and validation (V&V) are now commonplace, and test for whether the mathematical equations have been solved correctly numerically (verification) and whether the mathematical equations accurately describe the physical problem to be solved (validation). Following the V&V campaigns is an uncertainty quantification (UQ) assessment, which attempts to determine error bars for simulations, as due to numerical approximations, physical modeling approximations, or limited or inaccurate data. Finally, there is a concern with quantified margins of uncertainty, which are a translation of the traditional engineering safety margins into this modern simulation based technology. These attempt to assure design robustness in terms of a distance between a "safe" design point and possible failed design points, with allowances for all above mentioned uncertainties and errors. Plainly, these issues raise many statistical issues, in addition to the obvious physical, numerical, and engineering ones. Design of experiments, principal component analysis, Monte Carlo methods, Bayesian methods and ANOVA come to mind. Stochastic methods, such as stochastic partial differential equations, arise naturally. Sometimes, standard methods are sufficient, but for difficult problems it is very likely that extensions and even far reaching extensions will be required. In this talk we will present the statistical concepts which underlie modern simulation error analysis, and we will illustrate some of the difficulties by drawing on the presenter's experience in dealing with these issues. As a member of a scram jet design team in collaboration with the Stanford Turbulence Center, we will draw on examples, such as for turbulent combustion, where new ideas from statistics may be needed.


Speaker: Carol Gotway Crawford
Title: Linking Environmental and Health Data from Different Spatial Scales: A Case Study from Florida's Environmental Public Health Tracking Initiative
Abstract: The Centers for Disease Control and Prevention (CDC) created the Environmental Public Health Tracking (EPHT) Program to integrate hazard monitoring, exposure, and health effects surveillance into a cohesive tracking network. Few new data are being collected in the EPHT effort. The emphasis has been, and likely will continue to be, on the synthesis of existing environmental and health data systems. Part of Florida's effort to move toward implementation of EPHT is to develop models of the spatial and temporal association between myocardial infarctions and ambient ozone levels in Florida. Existing data were obtained from Florida's Agency for Health Care Administration, Florida's Department of Environmental Protection, the U.S. Census Bureau, and CDC's Behavioral Risk Factor Surveillance System. In this presentation, we highlight the opportunities and challenges associated with combining disparate spatial data for EPHT analyses. We compare the results from two different approaches to data linkage, focusing on the need to account for spatial scale and the support of spatial data in the analysis.


Speaker: Martha Grover Gallivan
Title: An Experimental Design Approach to Process Design
Abstract: When designing a new process, one rarely has a perfect model, but in the case of nanoscale systems, there may be several candidate models with unknown coefficients, and none will be the "correct" one. Also, technology and specifications for products such as microelectronics and pharmaceuticals change quickly, not leaving enough time to implement a more accurate model once it has been built and validated. However, mechanistic understanding will be encoded in the candidate models, such that they can be useful in process design, especially over a subset of the experimental space. One way to approach this problem is to use the set of candidate models, along with the available experimental data, to design the next experiment. By designing experiments to improve the candidate models' prediction variance at the predicted optimal operating point of a process, one can better understand the underlying phenomena and apply this knowledge to improve the candidate models. We apply concepts for model selection and spatial statistics to experimental design, for the microstructure design of metal oxides by a chemical vapor deposition process.


Speaker: Mike Hamada
Title: Bayesian Assessment of Repairable System Reliability and Availability
Abstract: This talk presents the reliability assessment of repairable systems using failure count and failure time data using a Bayesian framework. With this framework, we can analyze these data with standard repairable system models and naturally handle situations that require hierarchical models. Using the analysis results, we can then evaluate current reliability and other performance criteria of repairable systems. We also consider availability, which accounts for the time to make repairs, and show how simulation simplifies this evaluation.


Speaker: Dave Higdon
Title: Inference from combining detailed computer simulations and experimental data
Abstract: Inference regarding complex physical systems (e.g. subsurface aquifers, charged particle accelerators, shock physics) is typically plagued by a lack of information available from relevant, experimental data. What data is available is usually limited and informs inderctly about the phenomena of interest. However, when the physical system is amenable to computer simulation, these simulations can be combined with experimental observations to give useful information regarding calibration parameters, prediction uncertainty, and model inadequacy. This talk describes a framework for carrying out such simulation-based predictive investigations which involves experimental design, sensitivity analysis, response surface modeling, parameter estimation and accounting for systematic discrepancies between the simulation output and experimental data. An application from cosmology will be used as the context for this talk.


Speaker: John Langford
Title: Learning without the Loss
Abstract: In many natural situations, you can probe the loss (or reward) for one action, but you do not know the loss of other actions. This problem is simpler and more tractable than reinforcement learning, but still substantially harder than supervised learning because it has an inherent exploration component. I will discuss two algorithms for this setting: (1) Epoch-greedy, which is a very simple method for trading off between exploration and exploitation. (2) Offset Tree, which is a method for reducing this problem to binary classification.


Speaker: William Li
Title: Analyzing Supersaturated Designs via Model Selection Methods - Do They Work?
Abstract: In the area of supersaturated designs, most attention has been given on the construction of the efficient supersaturated designs in the literature. On the analysis of supersaturated designs, while many traditional model selection methods are applicable, it has been warned in the literature that these methods should only be used with caution. We compare several model selection methods, including all-subset, stepwise, LASSO, and sparse sliced inverse regression. The simulation results show that probability of picking true effects depends on the chose supersaturated design, the correlation structure of the design, and the magnitude of the coefficients. We conclude with the analysis of a real application of the supersaturated designs in the financial services industry.


Speaker: C. Devon Lin
Title: A Flexible Method for Constructing Designs for Computer Experiments
Abstract: It is becoming increasingly popular to perform scientific experiments on computer simulators since rapid growth in computer power has made it possible to study complex physical phenomena that might otherwise be too time-consuming, expensive, or impossible to observe. In many situations, the dimensionality of the inputs to the computer simulators can be very large. In others, a large simulation of the complex phenomena may be conducted, which requires a new approach to design of experiments. In this talk, I will introduce methods for constructing a rich class of Latin hypercubes of flexible run size with appealing projection and space-filling properties. The class includes many orthogonal Latin hypercubes that are not available in the literature, as well as nearly-orthogonal Latin hypercubes, and cascading Latin hypercubes. This is joint work with Derek Bingham, Randy Sitter and Boxin Tang at Simon Fraser University.


Speaker: Jason Loeppky
Title: Design of Computer Experiments: Does size matter?
Abstract: In recent years, virtual experiments implemented by a complex computer code or mathematical model are supplementing or even replacing physical experiments. The computer code mathematically describes the relationship between several input variables and one or more output variables. Often the computer models in question can be computationally demanding. Thus, direct evaluation of the code for optimization or validation is not possible in general. The general strategy is to build a statistical model to act a surrogate or an emulator of the true code. A long used rule of thumb for sample size takes a runs size that is 10 times the number of active dimensions. In this talk we investigate this rule of thumb for a variety of problems encountered in practice. In some cases we will show that increasing the sample size has a large effect on prediction quality and in other cases increasing the sample size has little to no effect. These issues will be demonstrated using a model for polar ice caps and a model for the ligand activation of a G-protien in yeast.


Speaker: Jesus Lopez-Fidalgo
Title: Optimal designs for models with potential censoring
Abstract: Designing an experiment for a real life problem may involve new and complex situations. As motivation a medical problem of finding an experimental design to predict cardiopulmonary morbidity after lung resection with standardized exercise oximetry is considered. A degree of complexity appears when an experimental unit cannot complete the assigned experimental condition, e.g. the prescribed exercise time in the medical example. Thus, the explanatory variable has to be considered as potentially censored. This presentation deals with optimal design theory for models with potential censoring either on independent or dependent variables. On the one hand optimal approximate designs when an independent variable might be censored are considered. The problem is which design should be applied to obtain at the end of the experimentation an optimal approximate design when the censored distribution function is assumed known in advance. On the other hand optimal experimental design theory is adapted to a particular Cox Regression problem. The failure time is modelled according to a probability distribution depending on some explanatory variables through a linear model. In both cases equivalent theorems and algorithms are provided in order to calculate optimal designs. Some examples illustrate these approaches for D-optimality.


Speaker: Michael Luvalle
Title: The application of statistical kinetics to reliability: status and current research problems
Abstract: Statistical kinetics is the use of statistical methods to extract kinetic information from degradation and failure data. While there are plainly advantages to being able to model the internal process causing degradation and failure, few statisticians have applied the ideas since their introduction in the mid 1980’s. Part of the reason is the need to be familiar with some of the mathematics of chemical kinetics and physics, and part is that perhaps only 10-20% of new material system environment combinations require the effort. This talk will be in two parts, in the first part, I will address the 1st problem by describing “statistical kinetics light” particularly for analysis of failure time data consisting of: (1) Five structural models based on a 1step expansion of the kinetics around the accelerated life model. (2) A way to generate families of failure time models, amenable to standard maximum likelihood analysis, based on extending an approach by Meeker to include link functions similar to those used in GLM. (3)A simplified computational approach to identifying parts of the parameter space in alternative kinetic models that are unidentifiable from current experiments. In the second part I will discuss current research problems in applying statistical kinetic models to degradation data.


Speaker: David Mease
Title: Evidence Contrary to the Statistical View of Boosting
Abstract: The statistical perspective on boosting algorithms focuses on optimization, drawing parallels with maximum likelihood estimation for logistic regression. In this talk we present empirical evidence that raises questions about this view. Although the statistical perspective provides a theoretical framework within which it is possible to derive theorems and create new algorithms for general contexts, we show that there remain many unanswered important questions. Furthermore, we provide examples that reveal crucial flaws in the many practical suggestions and new algorithms that are derived from the statistical view. We examine experiments using simple simulation models to illustrate some of these flaws and their practical consequences. This is joint work with Abraham Wyner at the University of Pennsylvania.


Speaker: Robert Mee
Title: Supersaturated Designs: Our Results Aren't Significant
Abstract: Whether using forward selection or all-subsets regression, it is common to select models from supersaturated designs that explain a very large percentage of the total variation in a response. The naïve p-values one sees for the selected model can persuade the user that in fact the included factors are clearly active. We show how permutation procedures may be used to more appropriately ascertain statistical significance both for the overall model and for individual coefficients. We illustrate the methods for several examples. We also show how the power for detecting an active effect decreases as the number of factors in the supersaturated design increases.


Speaker: Max Morris
Title: Data-Driven Nonstationary Modeling of Deterministic Computer Models
Abstract: “Spatial” Gaussian stochastic processes are the basis for much of the statistical methodology recently developed for analyzing data from deterministic computer models. Unless substantial prior information is available about the computer model (or function), applications are typically based on stationary processes, or on processes with stationary variance structure and very simple spatial structure in the mean (e.g. polynomial). When the computer model has characteristics that are not typical of stationary process realizations, this can often result in undesirable behavior in the point predictors of unobserved model outputs and/or output prediction intervals that undercover their targets. In this talk, we introduce a family of function prediction procedures that take the general form of traditional conditional/posterior predictors based on stationary models, but that have the flexibility to behave in a nonstationary way according to the observed data. Demonstration based on test functions shows that these methods can result in predictions of substantially smaller root-mean-squared-error and closer-to-nominal prediction interval coverage than either Bayes or Empirical Bayes predictions based on stationary processes.


Speaker: Shane Reese
Title: A Computational Approach for the Identification of Pollution Source Directions
Abstract: Pollution source apportionment (PSA) is the practice of identifying and describing pollution sources and their contributions. PSA frequently requires the identification of source directions, often as a post-analysis check to ensure that the contribution estimates are reasonable. In this talk, we develop a method of identifying source directions which solves the inverse problem for a particle dispersion model. MCMC is used to evaluate the complex relationship among observed pollutant concentrations, available meteorological information, and unknown source direction parameters. The method is flexible enough to identify multiple source directions for cases in which a species or source type of interest is emitted at more than one location, and Reversible Jump MCMC is used to evaluate the appropriate number of sources. The approach is demonstrated on the St. Louis EPA supersite airshed, demonstrating promising results in identifying known emitters of a variety of pollutants.


Speaker: Henry Rolka
Title: Recent Background and Influences on the Direction of Research in Public Health Biosurveillance
Abstract: Biosurveillance for Public Health has been a focus of recent legislative and policy initiatives such as the Pandemic and All Hazards Preparedness Act (PAHPA) and Homeland Security Presidential Directives (HSPDs). The consequences of these initiatives produce increased technical requirements for data analysis, information science and communicating biosurveillance results involving the characterization of uncertainty. In addition to traditional analytic epidemiological studies, data and information processing is used to establish a “common operating picture” and “situational awareness” for public health on an ongoing bases and especially to be used in responding to emergencies. These evolving public health operational requirements augment the complexity of practical biosurveillance. This talk will provide a brief background and history with some examples of challenging components and areas of potential research.


Speaker: Galit Shmueli
Title: Statistical Challenges in Modern Biosurveillance
Abstract: Modern biosurveillance is the monitoring of a wide-range of pre-diagnostic and diagnostic data for the purpose of enhancing the ability of the public health infrastructure to detect, investigate, and respond to disease outbreaks. Statistical control charts have been a central tool in classic disease surveillance and have also migrated into modern biosurveillance. However, the new types of data monitored, the processes underlying the time series derived from these data, and the application context all deviate from the industrial setting for which these tools were originally designed. Assumptions of normality, independence, and stationarity are typically violated in syndromic time series; target values of process parameters are time-dependent and hard to define; data labeling is ambiguous in the sense that outbreak periods are not clearly defied or known. Additional challenges arise such as multiplicity in several dimensions, performance evaluation, and practical system usage and requirements. Our focus is mainly on the monitoring of time series for early alerting of anomalies to stimulate investigation of potential outbreaks, with a brief summary of methods to detect significant spatial and spatiotemporal case clusters. We discuss the different statistical challenges in monitoring modern biosurveillance data, describe the current state of monitoring in the field, and survey the most recent biosurveillance literature.


Speaker: David Steinberg
Title: Orthogonal Nearly Latin Hypercube Designs
Abstract: Latin Hypercube (LHC) designs are one of the most popular choices for experiments run on computer simulators. As first proposed by McKay, Beckman and Conover in 1979, LHC designs guarantee that input factor settings are uniformly spread for each single factor, but rely on “random mating” to achieve good spread in high dimensions. In experiments with many factors, some pairs of factors typically have moderately high correlations and a number of schemes have been proposed to reduce the correlations. Steinberg and Lin derived a construction scheme that gives n-run LHC designs, with close to n orthogonal columns, but for very limited values of n. Here we extend that approach to a broad set of sample sizes. We are able to preserve the perfect orthogonality, but not the LHC property. However, the designs are "nearly" LHC's, in the sense that the univariate projection of each factor is close to uniform. The construction scheme relies on projection properties of Plackett-Burman designs and proves to be useful in the original class of designs, as well, for avoiding poorly covered low-dimensional projections. This is joint work with Dennis Lin.


Speaker: Ingo Steinwart
Title: Estimating Conditional Quantiles with Support Vector Machines
Abstract: We first recall a recently proposed support vector machine (SVM) formulation for the problem of estimating conditional quantiles. Since this SVM is based on the so-called pinball loss we then investigate this loss and its quantitative relation to conditional quantiles in detail. Finally, we apply these findings to describe the learning performance of the corresponding SVM.


Speaker: John Stufken
Title: On determining support points of locally optimal designs for nonlinear models
Abstract: We develop new tools for identifying the support points of locally optimal designs for nonlinear models. In contrast to the commonly used geometric approach, we use an approach based on algebraic tools.The general results can be applied to often used models, including logistic, probit, double exponential and double reciprocal models for binary data, a loglinear Poisson regression model for count data, and the Michaelis-Menten model. The approach works both with constrained and unconstrained design regions and is relatively easy to implement. This is joint work with Min Yang, University of Missouri-Columbia.


Speaker: Agus Sudjianto
Title: Statistical Methods for Fighting Financial Crimes
Abstract: Financial crimes affect millions of people every year, and financial institutions must employ methods to protect themselves and their customers. Fraud and money laundering are two common types of crimes, and there has been extensive research to develop algorithms to detect these crimes. Detection algorithms face a large set of challenges: financial crimes are rare events, which leads to severely imbalanced classes; criminals deliberately attempt to conceal the nature of their actions, resulting in severe class overlapping; and they quickly change strategies over time in order to trick current detection techniques. In some cases, legal constraints and investigation delays make it impossible to actually verify suspected crimes in a timely manner, resulting in class mislabeling. In addition, the volume and complexity of financial data requires not only algorithm effectiveness, but also execution and retraining efficiency. We discuss some of the classic statistical techniques that have been applied, as well as more recent machine learning and data mining algorithms. Our focus is to introduce the subject and to provide a survey of broad classes of methodologies accompanied by illustrative examples of applications. When required by practical applications, we also suggest some improvements to the existing methodologies.


Speaker: Agus Sudjianto
Title: Credit Portfolio Modeling in Retail Banking and Its Statistical Challenge
Abstract: Credit risk modeling is a very active research topic in computational finance; in particular, for the purpose of modeling corporate default. In retail banking, however, the subject has received less attention. The purpose of this talk is to introduce the challenge of modeling the credit risk of very large and heterogeneous credit portfolios in retail banking from a statistical perspective. Various existing approaches will be presented accompanied by examples. Problems encountered in practice will be presented to stimulate further research.


Speaker: Boxin Tang
Title: Existence and construction of two-level orthogonal arrays for estimating main effects and some specified two-factor interactions
Abstract: This paper considers two-level orthogonal arrays that allow joint estimation of all main effects and a set of prespecified two-factor interactions. We obtain some theoretical results, which provide a simple characterization of when such designs exist and how to construct them if they do exist. General as well as concrete applications of the results are discussed.


Speaker: Martina Vandebroek
Title: Design issues in Stated Preference Experiments
Abstract: In a stated preference or conjoint experiment respondents evaluate a number of products that are defined by their underlying characteristics. The resulting data yields information on the importance that respondents attach to the different characteristics, also called the part-worths. With this information, one can forecast consumer demand for new products. Data can be collected in various ways. In a discrete choice experiment several choice sets consisting of a number of products are presented to the respondents, these are then asked to choose their preferred product from each choice set. Alternatively, respondents can be asked to rank the alternatives in the choice set in a decreasing or increasing order of utility which is then called a rank order experiment. When respondents are asked to rate the alternative products, the experiment is called a rating-based conjoint experiment. The optimal design of a stated preference experiment consists of choosing the appropriate alternatives and of grouping the alternatives in choice sets such that the information gathered about the part-worths is maximized. In this presentation an overview will be given of the statistical models that are used to analyze stated preference data and of the design issues involved. Special attention will be given to the problem of assessing accurately the marginal rate of substitution which measures the consumer's willingness to give up an attribute of a product in exchange for another attribute. The marginal rate of substitution can be obtained by taking the ratio of two part-worths which leads to specific design problems.


Speaker: Lance Waller
Title: Spatial pattern and process: Assessing spatial performance of spatial models
Abstract: The topic is the use of spatial statistics to evaluate spatial simulation models with examples (primarily) from disease ecology, but I'll link these to more general aspects of linking mathematical models and statistical inference.


Speaker: Shuchun Wang
Title: Semiparametric Modeling of Macroeconomic Drivers to Credit Performance
Abstract: A central issue in credit risk management is the identification of the driving forces of credit performance. Empirical approaches often shed little light on the causal relationship among state variables whereas theoretical approaches often fail to capture the typical characteristics observed in practice. In this work, we present a semiparametric structural model to bridge this gap. The model explains how macro-economy changes may affect credit performance. With rigorous econometric analysis, we show how this model helps us understand the complex relationship between state economy and credit performance using a real data example. [Joint work with Dr. Ming Yuan, Ph.D.]


Speaker: Zhong Lin Wang
Title: Nanogenerators: from designed nanomaterials synthesis to energy harvesting
Abstract: Developing novel technologies for wireless nanodevices and nanosystems are of critical importance for in-situ, real-time and implantable biosensing, environmental monitoring, defense technology and even personal electronics. It is highly desired for wireless devices and even required for implanted biomedical devices to be self-powered without using battery. Therefore, it is essential to explore innovative nanotechnologies for converting mechanical energy (such as body movement, muscle stretching), vibration energy (such as acoustic/ultrasonic wave), and hydraulic energy (such as body fluid and blood flow) into electric energy that will be used to power nanodevices without using battery. We have demonstrated an innovative approach for converting nano-scale mechanical energy into electric energy by piezoelectric zinc oxide nanowire (NW) arrays. By deflecting the aligned NWs using an array of conductive atomic force microscopy (AFM) tips in contact mode, the energy that was first created by the deflection force and later converted into electricity by piezoelectric effect has been measured for demonstrating nano-scale power generator. The operation mechanism of the electric generator relies on the unique coupling of piezoelectric and semiconducting dual properties of ZnO as well as the elegant rectifying function of the Schottky barrier formed between the metal tip and the NW. Based on this mechanism, we have recently developed DC nanogenerator driven by ultrasonic wave in bio-fluid, which is a gigantic step towards applications in practice. This presentation will introduce the basic principle of the nanogenerator and the key role played by designed growth of NW arrays for determining the performance of the nanogenerators.


Speaker: Bill Woodall
Title: Research Issues and Ideas on Health-Related Surveillance
Abstract: In this overview paper, some of the surveillance methods and metrics used in health-related applications are described and contrasted with those used in industrial practice. Many of the health-related methods are based on the concepts and methods of statistical process control. Public health data often include spatial information as well as temporal information, and in this and other regards, public health applications could be considered more challenging than industrial applications. Avenues of research into various topics in health-related monitoring are suggested.


Speaker: Dave Woods
Title: Experiments in blocks for a non-normal response via generalised estimating equations
Abstract: Many industrial experiments measure a response that cannot be adequately described by a linear model with normally distributed errors. An example is an experiment in aeronautics to investigate the cracking of bearing coatings where a binary response was observed, success (no cracking) or failure (cracked). A further complication which often occurs in practice is the need to run the experiment in blocks, for example, to account for different operators or batches of experimental units. To produce more efficient experiments, block effects are often included in the model for the response. When the block effects can be considered as nuisance variables, a marginal (or population averaged) model may be appropriate, where the effect of individual blocks are not explicitly modelled. We discuss block designs for experiments where the response is described by a marginal model fitted using Generalised Estimating Equations (GEEs). GEEs are an extension of Generalised Linear Models (GLMs) that incorporate a correlation structure between experiment units in the same block; the marginal response for each observation follows an appropriate GLM. This talk will describe some design strategies for such models in an industrial context.


Speaker: Huaiqing Wu
Title: Analysis of Window-Observation Recurrence Data
Abstract: Many systems experience recurrent events. Recurrence data are collected to analyze quantities of interest, such as the mean cumulative number of events. Methods of analysis are available for recurrence data with left and/or right censoring. Due to practical constraints, however, recurrence data are sometimes recorded only in windows. Between the windows, there are gaps over which the process cannot be observed. We extend existing statistical methods, both nonparametric and parametric, to window-observation recurrence data. The nonparametric estimator requires minimum assumptions, but will be inconsistent if the size of the risk set is not positive over the entire period of interest. There is no such difficulty when using a parametric model for the recurrence data. For cases in which the size of the risk set is zero for some periods of time, we propose and compare two alternative hybrid estimators. The methods are illustrated with two example applications.(This is joint work with Jianying Zuo and William Q. Meeker).


Speaker: Aijun Zhang
Title: Recent Advances in Default Risk Modeling Based on Vintage Data
Abstract: Vintage data commonly arise in financial risk management, especially in today's retail banking. They have two special features: triangular structure and dual-time coordinates. Existing credit risk models cannot be applied in this situation. In this paper, we propose a dual-time intensity model for analyzing the default risk, which is decomposable into maturation, macroeconomic and vintage effects. [Joint work with Dr. Agus Sudjianto, Ph.D. and Dr. Vijay Nair, Ph.D.]