Article
Geosciences, Multidisciplinary
Andrea Diana, Elvira Romano, Antonio Irpino
Summary: Recently, there has been an increasing interest in distribution-free prediction in the fields of machine learning and statistics. This study introduces an inductive conformal prediction strategy specifically designed for spatiofunctional data, using two different regression models to make predictions and proposing novel measures of non-conformity and prediction bands for the functional response variable.
SPATIAL STATISTICS
(2023)
Article
Statistics & Probability
Huiying Mao, Ryan Martin, Brian J. J. Reich
Summary: This article presents a new model-free nonparametric spatial prediction approach based on conformal prediction, which is applicable to complex spatial statistics problems. Numerical experiments demonstrate that the proposed prediction method has higher efficiency and validity for large datasets in various spatial settings.
JOURNAL OF THE AMERICAN STATISTICAL ASSOCIATION
(2022)
Article
Chemistry, Physical
R. Remme, T. Kaczun, M. Scheurer, A. Dreuw, F. A. Hamprecht
Summary: Orbital-free density functional theory (OF-DFT) is a promising method for computing ground state molecular properties at minimal cost. In this study, we propose a deep neural network architecture called KineticNet to learn the kinetic energy functional from ground truth provided by the more expensive Kohn-Sham density functional theory. KineticNet achieves chemical accuracy for learned functionals across different input densities and geometries of small molecules, and even demonstrates OF-DFT density optimization with chemical accuracy for two-electron systems.
JOURNAL OF CHEMICAL PHYSICS
(2023)
Article
Engineering, Biomedical
Jaehyoung Hong, Hyonho Chun
Summary: Advancements in wearable devices have made it possible to track healthcare outcomes such as blood pressure and heart rate over time, allowing for the prediction of health risks and personalized medicine. To accurately predict individual responses, it is important to account for the variation among subjects. Sharing information through a mixed effect model can improve predictions, but when there are multiple patterns in the data, sharing information across all patients may dilute signals.
BIOMEDICAL SIGNAL PROCESSING AND CONTROL
(2023)
Article
Computer Science, Interdisciplinary Applications
Martial Amovin-Assagba, Irene Gannaz, Julien Jacques
Summary: In an industrial setting, the activity of sensors is recorded frequently. This study proposes a contaminated mixture model for automatically detecting abnormal measurement behavior in multivariate functional data sets. The model achieves high performance and flexibility in detecting outliers.
COMPUTATIONAL STATISTICS & DATA ANALYSIS
(2022)
Article
Biochemical Research Methods
Zachary R. McCaw, Hugues Aschard, Hanna Julienne
Summary: This study introduces a Gaussian mixture model for dealing with missing data and develops an R package for handling such data. The results indicate that this model is more effective in recovering true cluster assignments and provides accurate assessment of cluster assignment uncertainty.
BMC BIOINFORMATICS
(2022)
Article
Computer Science, Artificial Intelligence
Younghwan Jeon, Ganguk Hwang
Summary: This paper addresses the data association problem and proposes a Bayesian approach based on a mixture of Gaussian Processes (GPs) to adapt to changing observations. Experimental results and theoretical analysis demonstrate the effectiveness of the proposed method.
PATTERN RECOGNITION
(2022)
Article
Statistics & Probability
Jacopo Diquigiovanni, Matteo Fontana, Simone Vantini
Summary: Motivated by the need for prediction sets in a general regression framework, this study proposes a set of conformal predictors that can generate valid or exact multivariate simultaneous prediction bands under the assumption of exchangeable regression pairs. The method is investigated and analyzed through simulations and a real-world application in urban mobility.
JOURNAL OF MULTIVARIATE ANALYSIS
(2022)
Article
Statistics & Probability
Arun K. Kuchibhotla, Richard A. Berk
Summary: This paper focuses on the accuracy of risk forecasts for a probation and parole department in the United States, considering the use of statistical learning risk methods. The study demonstrates that uncertainty measures using nested conformal prediction sets can differ significantly from standard uncertainty measures and provides a modified approach called the localized conformal method. The comparison and recommendations made in this paper are novel.
ANNALS OF APPLIED STATISTICS
(2023)
Review
Chemistry, Multidisciplinary
Zhen Tao, Qi Yu, Saswata Roy, Sharon Hammes-Schiffer
Summary: This article introduces the application of the NEO framework to direct dynamics simulations of chemical reactions such as proton, hydride, and proton-coupled electron transfer reactions. The NEO method can effectively include zero-point energy, density delocalization, and anharmonicity, and can be used for simulations of reactions in ground and excited states.
ACCOUNTS OF CHEMICAL RESEARCH
(2021)
Article
Biotechnology & Applied Microbiology
Vassilis Alimisis, Georgios Gennis, Konstantinos Touloupas, Christos Dimas, Nikolaos Uzunoglu, Paul P. Sotiriadis
Summary: This paper introduces a new analog front-end classification system for epileptic seizure prediction in embedded devices, combining an analog feature extractor with an analog Gaussian mixture model-based binary classifier, providing high sensitivity and low power consumption.
BIOENGINEERING-BASEL
(2022)
Article
Computer Science, Artificial Intelligence
Leonardo Cella, Ryan Martin
Summary: This paper addresses the fundamental problem of prediction in statistics, proposing the use of probabilistic predictors to quantify the uncertainty of future observations. The concept of validity is introduced, and its behavioral and statistical implications are explored. The study shows that valid probabilistic predictors must be imprecise, avoid sure loss, and have desirable frequentist error rate control properties.
INTERNATIONAL JOURNAL OF APPROXIMATE REASONING
(2022)
Article
Mathematics, Interdisciplinary Applications
Terrance D. Savitsky, Matthew R. Williams
Summary: The U.S. Bureau of Labor Statistics publishes monthly employment totals for all U.S. counties, but does not consider the dependence among counties. This study proposes a joint modeling approach for employment time series, treating them as indexed noisy functions. The application is one of the first in the U.S. Federal Statistical System to address heterogenous seasonality patterns among a collection of time series.
Article
Mathematical & Computational Biology
Tanya P. Garcia, Layla Parast
Summary: This study introduces a novel nonparametric estimator for estimating cumulative risk in scenarios where genetic mutation status is unknown, showing improved prediction accuracy by incorporating covariate information and dynamic landmark prediction. The estimator is unbiased and more accurate compared to methods that ignore covariate information and landmarking. Applying this method to a study on Huntington disease mortality, dynamic survival prediction curves are developed incorporating gender and familial genetic information.
Article
Computer Science, Information Systems
Feng Wang, Fanshu Liao, Yixuan Li, Hui Wang
Summary: Dynamic multi-objective optimization problems have attracted attention in recent years. Predicting the movement of the Pareto set or Pareto front in changing environments is a challenging problem. A new prediction method, MOEA/D-GMM, incorporating the Gaussian Mixture Model into the MOEA/D framework, shows better performance in tracking the new Pareto set compared to existing strategies, especially when historical information quality and non-linear changes affect the results.
INFORMATION SCIENCES
(2021)