Christopher Krapu | Searcharxiv

arXiv2024

Deep autoregressive modeling for land use land cover

Land use / land cover (LULC) modeling is a challenging task due to long-range dependencies between geographic features and distinct spatial patterns related to topography, ecology, and human development. We identify a close connection between modeling of spatial patterns of land use and the task of image inpainting from computer vision and conduct a study of a modified PixelCNN architecture with approximately 19 million parameters for modeling LULC. In comparison with a benchmark spatial statistical model, we find that the former is capable of capturing much richer spatial correlation patterns such as roads and water bodies but does not produce a calibrated predictive distribution, suggesting the need for additional tuning. We find evidence of predictive underdispersion with regard to important ecologically-relevant land use statistics such as patch count and adjacency which can be ameliorated to some extent by manipulating sampling variability.

cs.CV↗

arXiv2023

Flexible hierarchical risk modeling for large insurance data via NumPyro

Data analysis and individual policy-level modeling for insurance involves handling large data sets with strong spatiotemporal correlations, non-Gaussian distributions, and complex hierarchical structures. In this research, we demonstrate that by utilizing gradient-based Markov chain Monte Carlo (MCMC) techniques accelerated by graphics processing units, the trade-off between complex model structures and scalability for inference is overcome at the million-record size. By implementing our model in NumPyro, we leverage its built-in MCMC capabilities to fit a model with multiple sophisticated components such as latent conditional autoregression and spline-based exposure adjustment, achieving an 8.8x speedup compared to CPU-based implementations. We apply this model to a case study of 2.6 million individual policy-level claim count records for automobile insurance from Brazil in 2011. We illustrate how this modeling approach significantly advances current risk assessment processes for numerous, closely related outcomes. The code and data are available at https://github.com/ckrapu/bayes-at-scale.

stat.AP↗