Seminars are held weekly on Wednesdays, from 3-4pm central time in Hardin NW 49. Please attend in person if possible.
If you need a zoom link, please fill out the form at the bottom of this page to receive an invitation.
September 8 - Dean Dustin (UNL)
Title: Stability as an objective criterion for prior selection (in linear models).
Abstract: Choosing the penalty term in a shrinkage method is equivalent to choosing a prior. We modify standard shrinkage methods so the penalties are adaptive i.e., depend on the data. Then we propose stability criterion for the methods and evaluate how well each method performs under perturbations of the data. Through a series of computed examples, we show how can be done. We also interpret what this seems to mean since conventional asymptotic criteria such as oracle properties are not used. We present theorems showing that the oracle property is common among penalties that satisfy basic regularity conditions. Given the lack of uniqueness of the oracle property, we argue that a finite sample check, such as stability, is necessary to distinguish between methods. Finally, we propose a method for allowing the data to choose a penalty that is optimized for prediction using a genetic algorithm.
September 15 - Alumni Career Panel
Panelists: Jason Adams (Sandia), Ella Burnham (Gustavus Adolphus College), Jinyu Li (Microsoft), Sayli Pokal (Zoetis)
September 22 & 29 - No Seminar
October 6 - Andy Poppick (Carleton College)
Title: Observation-Based Simulations of Humidity and Temperature Using Quantile Regression
Abstract: The impacts of heat events depend on both temperature and humidity (e.g., humid heat waves are deadlier to people and dry heat waves create a greater fire risk). One of the primary information sources for studying changes in the distribution of climate variables (such as the bivariate distribution of temperature and humidity) is data from Atmosphere-Ocean General Circulation Models (GCMs). However, it is well understood that GCMs do not perfectly reproduce the distribution of the observed climate and therefore may be insufficient on their own if one needs a realistic simulation of the future climate (e.g., for use in a climate impacts study), implying a need for methods that combine information from historical observations with GCM output to produce better calibrated simulations. We present an observation-based, conditional quantile mapping approach to this problem in the context of temperature and humidity simulations. A temperature simulation is produced by transforming observations to include projected changes in mean and temporal covariance structure from a GCM. Next, a humidity simulation is produced by transforming observations to account for projected changes in the conditional humidity distribution given temperature. We use the Community Earth System Model Large Ensemble (CESM1-LE) to estimate projected changes in summertime temperature and humidity over the Continental United States, and create future simulations using station observations from the Global Summary of the Day. We find e.g. that CESM1-LE projects increases in the risk of high dew point on historically hot days and, in comparison with raw CESM1-LE output, our observation-based simulation largely projects smaller changes in the future risk of high humidity on days with historically warm temperatures. No prior knowledge of climate science will be assumed for this talk.
October 13 - Whitney Huang (Clemson University)
Title: Estimating Concurrent Climate Extremes: A Conditional Approach
Abstract: Simultaneous concurrence of extreme values across multiple climate variables can result in large societal and environmental impacts. Therefore, there is growing interest in understanding these concurrent extremes. In many applications, not only the frequency but also the magnitude of concurrent extremes are of interest. One way to approach this problem is to study the distribution of one climate variable given that another is extreme. In this work we develop a statistical framework for estimating bivariate concurrent extremes via a conditional approach, where univariate extreme value modeling is combined with dependence modeling of the conditional tail distribution using techniques from quantile regression and extreme value analysis to quantify concurrent extremes. We focus on the distribution of daily wind speed conditioned on daily precipitation taking its seasonal maximum. The Canadian Regional Climate Model large ensemble is used to assess the performance of the proposed framework both via a simulation study with specified dependence structure and via an analysis of the climate model-simulated dependence structure.
October 20 - Ved Piyush (UNL - Statistics)
Title: Automatic Image Captioning using Convolutional and Recurrent Neural Networks.
Abstract: Automatic image captioning is the process of generating a descriptive text description for an image. Image captioning is one of the few applications of deep neural networks where we work with image and text data simultaneously. This captioning model can be trained using standard backpropagation techniques such as Stochastic Gradient Descent (SGD). I trained this model on the MS-COCO dataset with real-world images of humans, animals, vehicles, etc., in various situations and surroundings. For training purposes, I use about 30,000 images which have five human annotations each. The trained captioning model is composed of a Convolutional Neural Network (CNN) to extract features from the image and a Long Short Term Memory Model (LSTM) to extract features from the text description of the image. The goal of the learning problem is to use these visual and textual features to predict a caption as close to the ground truth human caption as possible. To make the model more interpretable, I leverage the work of Xu et al. to visualize where in the image the model fixes its gaze to predict the words in the generated caption. A potential use case of the captioning model would be to create an application that can describe what is happening in a video frame by frame.
October 27 - Mark Risser (Lawrence Berkeley National Lab)
Title: Detecting changes in spatial extremes with attribution to both anthropogenic influences and natural modes of climate variability
Abstract: The gridding of daily accumulated precipitation--especially extremes--from ground-based station observations is problematic due to the fractal nature of precipitation, and therefore estimates of long period return values and their changes based on such gridded daily data sets are generally underestimated. To address this issue, we present a method for deriving “probabilistic” high-resolution data sets specifically designed to characterize the climatological properties of extreme precipitation by applying spatial statistical analyses to the extreme statistics of daily precipitation. Our methodology is appropriate for heterogeneous spatial domains and furthermore is scalable to an arbitrarily large network of weather stations. An important application of our approach is the detection of both natural and anthropogenically-induced changes in the climatology of extreme precipitation, for which we develop a robust statistical technique to identify significant pointwise changes while carefully controlling the rate of false positives. All uncertainty quantification is based on resampling methods, and we utilize supercomputing to quickly analyze, conduct inference, and detect seasonal changes in extremes for a network of approximately 5000 weather stations from the Global Historical Climatology Network over the contiguous United States. Human-induced climate change generally results in larger and more frequent extreme events, although there are also important areas where the opposite is true.
November 3 - Won Chang (University of Cincinnati)
Title: A Bayesian Solution to Inverse Problem for Circadian Cycles
Abstract: Most organisms exhibit various endogenous oscillating behaviors which provide crucial information as to how the internal biochemical processes are connected and regulated. Understanding the molecular mechanisms behind these oscillators requires interdisciplinary efforts combining both biological and computer experiments, as the latter can complement the former by simulating perturbed conditions with higher resolution. Harmonizing the two types of experiment, however, poses significant statistical challenges due to identifiability issues, numerical instability, and ill behavior in high dimension. This article devises a new Bayesian calibration framework for oscillating biochemical models. The proposed Bayesian model is estimated using an advanced MCMC which can efficiently infer the parameter values that match the simulated and observed oscillatory processes. Also proposed is an approach to sensitivity analysis approach based on the intervention posterior. This approach measures the influence of individual parameters on the target process by utilizing the obtained MCMC samples as a computational tool. The proposed framework is illustrated with circadian oscillations observed in a filamentous fungus, Neurospora crassa.
November 10 - Mike Hayes (UNL - SNR)
Title: Drought Chasing…in the Context of a Changing Climate
Abstract: It has been said that societies will manage climate variability and change in the same way they manage droughts. Given that perspective, how are the nation and the world dealing with drought risk management and meeting the challenge of facing droughts in the future? This presentation will provide a current assessment of the status of drought risk management, emphasizing the issues of monitoring and prediction, planning, and mitigation strategies. The presentation will also address several research projects that Dr. Hayes is involved with regarding drought and climate-related impacts, and how these can be relevant in the changing climate. If societies can be better prepared for drought, other natural hazards, and extreme events, their preparedness for the many potential challenges climate change will create will be improved as well.
November 17 - Vamsi Manthena (UNL - Statistics)
Title: Combining multi-type data to improve multicategory trait prediction
Abstract: Modern plant breeding programs collect several secondary types of data such as weather, image data, and secondary trait data. Genomic data is high-dimensional and often over-crowds smaller data types when combined together to explain the response variable. There is a need to develop methods that can combine different data types of differing sizes successfully to improve prediction. In this work, we develop a new three-step statistical method to predict multicategory traits by combining three data types — genomic, weather, and secondary trait and address the various challenges in this problem. We compare our method with various standard classifiers using real data.
November 24 - Thanksgiving - No Seminar
December 1 - Pedro Cesar de Oliveira Ribeiro (visiting PhD scholar)
Title: Hybrid Prediction of Biomass Sorghum using Climatic Data via Combining Ability Models
Abstract: The biomass of hybrid sorghum is used in the generation of bioenergy and its development requires three lines (A, B, and R). The development of hybrid sorghum involves numerous steps such as field trials in multi-locations to validate their performance to provide recommendations of the experimental hybrids. The breeding programs are interested in developing superior hybrids for a wide range of environmental conditions; however, the cost of field-testing is expensive and time-consuming. Genomic Selection (GS) is a powerful tool that allows breeders to predict the performance of new hybrids in yet to observe untested environments. GS models coupled with the use of environmental covariables (ECs) have the potential to enhance selection accuracy in breeding programs. The goals of this study were to: i) evaluate GS models to predict the biomass of untested sorghum hybrids in tested/untested environments, using a historical dataset of the Embrapa’s breeding program; ii) compare the effect of modeling different environmental kinship matrices with ECs; iii) identify mega-environments for sorghum hybrids based only on ECs. Evaluations were conducted using seven different models including main effects of environments; general combining ability (GCA) and specific combining ability (SCA) terms, interactions between genetic effects (GCA and SCA) with environments and ECs for different environmental kinship matrices based on ECs.
December 8 & 15 - Finals Prep - No Seminar
Seminar Zoom Link Request
(if you are a Stats Grad Student, the Zoom link can be found in your calendar invite)