Zachary James

Hi! I'm a PhD student at Cornell Bowers advised by Joe Guinness, using generative AI and high-performance computing to more efficiently interpolate and forecast geospatial data.

In my free time, I enjoy cooking, making origami, and finding an excuse to look at the shellfish on Rockaway Beach. I will be joining Meta in NYC as a Research Scientist in the fall.

email  /  linkedin  /  github

Research

I am interested in using diffusion models, Gaussian processes, and high-performance computing to analyze geospatial data, particularly with applications to climate science and ecology.

We create an image-to-image diffusion model that incorporates aerosol information into 6-hour forecasts of CAPE, producing more accurate forecasts.

We model the nonstationary parameters of a Gaussian process as functions of space using neural networks, enabling scalable inference on large datasets while maintaining interpretability.

Vecchia Approximation can be highly efficient on GPUs for large spatial datasets. We present memory and synchronization optimizations that improve performance by up to 20x over existing implementations.

Work done with the New York City Department of Emergency Management to improve real-time forecasting using high-resolution meteorological data and scalable spatial models.

We built an image classifier to identify fruit bats in camera trap images to help epidemiologists study the transmission of Nipah virus. This was done by fine-tuning a pretrained vgg16 model on a labeled dataset of images collected in Bangladesh.

A random graph model reveals that the urban-rural divide is a major determinant of opioid abuse in the state of Georgia.

Miscellanea

This website's source code is borrowed from Jon Barron.