More Like this

1. Quantifying Uncertainty in Discrete-Continuous and Skewed Data with Bayesian Deep Learning

   https://doi.org/10.1145/3219819.3219996  

   Vandal, Thomas; Kodra, Evan; Dy, Jennifer; Ganguly, Sangram; Nemani, Ramakrishna; Ganguly, Auroop R. (August 2018, KDD 2018)

   Deep Learning (DL) methods have been transforming computer vision with innovative adaptations to other domains including climate change. For DL to pervade Science and Engineering (S&EE) applications where risk management is a core component, well-characterized uncertainty estimates must accompany predictions. However, S&E observations and model-simulations often follow heavily skewed distributions and are not well modeled with DL approaches, since they usually optimize a Gaussian, or Euclidean, likelihood loss. Recent developments in Bayesian Deep Learning (BDL), which attempts to capture uncertainties from noisy observations, aleatoric, and from unknown model parameters, epistemic, provide us a foundation. Here we present a discrete-continuous BDL model with Gaussian and lognormal likelihoods for uncertainty quantification (UQ). We demonstrate the approach by developing UQ estimates on “DeepSD’‘, a super-resolution based DL model for Statistical Downscaling (SD) in climate applied to precipitation, which follows an extremely skewed distribution. We find that the discrete-continuous models outperform a basic Gaussian distribution in terms of predictive accuracy and uncertainty calibration. Furthermore, we find that the lognormal distribution, which can handle skewed distributions, produces quality uncertainty estimates at the extremes. Such results may be important across S&E, as well as other domains such as finance and economics, where extremes are often of significant interest. Furthermore, to our knowledge, this is the first UQ model in SD where both aleatoric and epistemic uncertainties are characterized. 

   more » « less
2. A Comparison of Zero-Inflated Models for Modern Biomedical Data

   https://doi.org/10.33697/ajur.2025.141  

   Beveridge, Max; Goldstein, Zach; Cheol_Chung, Hee (June 2025, American Journal of Undergraduate Research)

   There has been a growing number of datasets exhibiting an excess of zero values that cannot be adequately modeled using standard probability distributions. For example, microbiome data and single-cell RNA sequencing data consist of count measurements in which the proportion of zeros exceeds what can be captured by standard distributions such as the Poisson or negative binomial, while also requiring appropriate modeling of the nonzero counts. Several models have been proposed to address zero-inflated datasets including the zero-inflated negative binomial, hurdle negative binomial model, and the truncated latent Gaussian copula model. This study aims to compare various models and determine which one performs optimally under different conditions using both simulation studies and real data analyses. We are particularly interested in investigating how dependence among the variables, level of zeroinflation or deflation, and variance of the data affects model selection. KEYWORDS: Zero-InflatedModels; HurdleModels; Truncated Latent Gaussian CopulaModel; Microbiome Data; Gene-Sequencing Data; Zero-Inflation, Negative Binomial; Zero-Deflation 

   more » « less
3. Using a genetic algorithm with histogram-based feature selection in hyperspectral image classification

   https://doi.org/10.1145/3321707.3321748  

   Walton, Neil S.; Sheppard, John W.; Shaw, Joseph A. (July 2019, Proceedings of the Genetic and Evolutionary Computation Conference)

   Optical sensing has the potential to be an important tool in the automated monitoring of food quality. Specifically, hyperspectral imaging has enjoyed success in a variety of tasks ranging from plant species classification to ripeness evaluation in produce. Although effective, hyperspectral imaging is prohibitively expensive to deploy at scale in a retail setting. With this in mind, we develop a method to assist in designing a low-cost multispectral imager for produce monitoring by using a genetic algorithm (GA) that simultaneously selects a subset of informative wavelengths and identifies effective filter bandwidths for such an imager. Instead of selecting the single fittest member of the final population as our solution, we fit a univariate Gaussian mixture model to the histogram of the overall GA population, selecting the wavelengths associated with the peaks of the distributions as our solution. By evaluating the entire population, rather than a single solution, we are also able to specify filter bandwidths by calculating the standard deviations of the Gaussian distributions and computing the full-width at half-maximum values. In our experiments, we find that this novel histogram-based method for feature selection is effective when compared to both the standard GA and partial least squares discriminant analysis. 

   more » « less
4. Toward a 3D physical model of diffusive polymer chains

   https://doi.org/10.3389/fphy.2023.1142004  

   Karsai, Andras; Cassidy, Grace J.; Rajanala, Aradhya P.; Yang, Lixinhao; Kerimoglu, Deniz; Gumbart, James C.; Kim, Harold D.; Goldman, Daniel I. (June 2023, Frontiers in Physics)

   Recent studies in polymer physics have created macro-scale analogs to solute microscopic polymer chains like DNA by inducing diffusive motion on a chain of beads. These bead chains have persistence lengths of O(10) links and undergo diffusive motion under random fluctuations like vibration. We present a bead chain model within a new stochastic forcing system: an air fluidizing bed of granular media. A chain of spherical 6 mm resin beads crimped onto silk thread are buffeted randomly by the multiphase flow of grains and low density rising air “bubbles”. We “thermalize” bead chains of various lengths at different fluidizing airflow rates, while X-ray imaging captures a projection of the chains’ dynamics within the media. With modern 3D printing techniques, we can better represent complex polymers by geometrically varying bead connections and their relative strength, e.g., mimicking the variable stiffness between adjacent nucleotide pairs of DNA. We also develop Discrete Element Method (DEM) simulations to study the 3D motion of the bead chain, where the bead chain is represented by simulated spherical particles connected by linear and angular spring-like bonds. In experiment, we find that the velocity distributions of the beads follow exponential distributions rather than the Gaussian distributions expected from polymers in solution. Through use of the DEM simulation, we find that this difference can likely be attributed to the distributions of the forces imparted onto the chain from the fluidized bed environment. We anticipate expanding this study in the future to explore a wide range of chain composition and confinement geometry, which will provide insights into the physics of large biopolymers. 

   more » « less
5. Computer modelling of close-to-ground tornado wind-fields for different tornado widths

   Kashefizadeh, M.H; Verma, S; Selvam, R.P. (August 2019, Journal of wind engineering and industrial aerodynamics)

   Tangential velocity (Vt) of tornadoes is the major parameter that causes building damage. In-field tornado measurements are less reliable at less than 20 m above ground level (AGL). Laboratory tornado simulators suggest that swirl ratio (S) and radius (ro) are the major tornado parameters that influence the Vt. However, due to scaling problems, the laboratory simulators also report the Vt at greater than 20 m AGL. Well-refined computational fluid dynamics (CFD) models can evaluate the Vt at less than 10 m AGL. However, the CFD models are limited to ro = 1.0 km, and the effect of ro on Vt is not investigated. The aim of this study is to investigate the maximum Vt for different ro close to ground. Simulation results show that increasing ro decreases the maximum Vt with respect to Vro. Moreover, by increasing ro, the corresponding elevation of occurrence of maximum Vt (zmax) will increase. However, for all tornado radii, the zmax is between 20 m and 64 m AGL. In addition, results show that for all ro, the radial Vt profile has two peaks at z < 10 m AGL due to strong shear force close to the ground and at higher elevation the profile transits to Rankine Combined Vortex Model (RCVM). 

   more » « less