Bridging rigorous statistical science and real-world impact
Bridging rigorous statistical science and real-world impact
The Spatial Statistics Group develops advanced statistical and machine learning methods to tackle some of today’s most pressing geospatial and human dynamics challenges. Our work bridges gaps in data, resolution, and uncertainty to deliver actionable insights for real-world applications; from global building intelligence to population modeling.
Join a team advancing Bayesian reasoning
Join a team advancing Bayesian reasoning
Our team of researchers support approximately $2.5M of program funds across three primary sponsors. Key research disciplines are Statistics, Mathematics, Computer Science, Data Science, and other related degree fields. Research disciplines include Bayesian modeling and Inference, Machine Learning, Artificial Intelligence, and Applied Statistical Research. Core skill sets include probabilistic programming, data fusion and multiscale modeling, functional data analysis, AI/ML workflow automation, scientific software development and reproducibility.
Capabilities
Bayesian Modeling and Inference
Bayesian Modeling and Inference
Application: Global Building Intelligence
Our team develops complex, large-scale Bayesian hierarchical models to faithfully characterize underlying phenomena across varying levels of data sparsity. By translating both the physical and policy-driven aspects of a system into statistical requirements, our methodologies not only capture the dynamics present within observed data but also provide robust tools for estimation and imputation of missing information. In the context of Global Building Intelligence, we leverage social, physical, and locational attributes of observable buildings to construct scalable models that describe building characteristics across the world.
Our team develops complex, large-scale Bayesian hierarchical models to faithfully characterize underlying phenomena across varying levels of data sparsity. By translating both the physical and policy-driven aspects of a system into statistical requirements, our methodologies not only capture the dynamics present within observed data but also provide robust tools for estimation and imputation of missing information. In the context of Global Building Intelligence, we leverage social, physical, and locational attributes of observable buildings to construct scalable models that describe building characteristics across the world.
Uncertainty Quantification for AI/ML
Uncertainty Quantification for AI/ML
Application: Damage Parameter Estimation in Structural Health Dynamics
Our team advances the field of uncertainty quantification for artificial intelligence (AI) and machine learning (ML) models by developing explainable and consistent hypothesis tests for key parameters of interest. In the context of structural health dynamics, we train machine learning models to characterize the effects of external force loads on structures, particularly focusing on damage occurring at material boundaries. Using Spectral-Normalized Neural Gaussian Processes, we derive implicit distributions over critical parameters and leverage these distributions to construct hypothesis tests that quantify the extent of structural damage.
Our team advances the field of uncertainty quantification for artificial intelligence (AI) and machine learning (ML) models by developing explainable and consistent hypothesis tests for key parameters of interest. In the context of structural health dynamics, we train machine learning models to characterize the effects of external force loads on structures, particularly focusing on damage occurring at material boundaries. Using Spectral-Normalized Neural Gaussian Processes, we derive implicit distributions over critical parameters and leverage these distributions to construct hypothesis tests that quantify the extent of structural damage.
Latent Analysis for Trajectory Recovery
Latent Analysis for Trajectory Recovery
Application: Trajectory Reconstruction from Unlabeled Geolocations
Our team develops Bayesian stochastic differential equation (SDE) methodologies to infer trajectories from unordered spatiotemporal data. By employing a hierarchical modeling framework, we learn the parameters governing the drift and diffusion components of an Itô integral solution, enabling robust trajectory reconstruction. This capability supports a wide range of applications, including disaster recovery, traffic pattern analysis, flood dynamics, and population migration modeling.
Our team develops Bayesian stochastic differential equation (SDE) methodologies to infer trajectories from unordered spatiotemporal data. By employing a hierarchical modeling framework, we learn the parameters governing the drift and diffusion components of an Itô integral solution, enabling robust trajectory reconstruction. This capability supports a wide range of applications, including disaster recovery, traffic pattern analysis, flood dynamics, and population migration modeling.
Probabilistic Time Series Modeling
Probabilistic Time Series Modeling
Application: Hourly Building Occupancy Dynamics
Our team designs scalable Bayesian models to estimate hourly building occupancy patterns using data sources such as Google Popular Times, mobility datasets, and operational schedules. By modeling occupancy as a probabilistic distribution over time, we capture not only expected activity levels but also the associated uncertainty and variation across different place types and cultural contexts. This capability enables a wide range of applications, including disaster response planning, energy demand forecasting, urban infrastructure optimization, and land use validation.
Our team designs scalable Bayesian models to estimate hourly building occupancy patterns using data sources such as Google Popular Times, mobility datasets, and operational schedules. By modeling occupancy as a probabilistic distribution over time, we capture not only expected activity levels but also the associated uncertainty and variation across different place types and cultural contexts. This capability enables a wide range of applications, including disaster response planning, energy demand forecasting, urban infrastructure optimization, and land use validation.
Contact
Justin Jacobs, Spatial Statistics Group Leader
View Justin's Staff Profile
Justin Jacobs, Spatial Statistics Group Leader