Bio/CV:
I am a petrologist working at the intersection of geochemistry and machine learning. I am fascinated by records of Earth's magmatism and volcanism stored in minerals and melt inclusions, from magma storage and transport prior to eruption to the timescales of magmatic processes. A central thread in my research is making these records more accessible and reproducible. I develop open-source tools like PyIRoGlass for quantifying volatiles in melt inclusions and glasses with Bayesian inference and mineralML for probabilistic mineral classification with neural networks using transfer learning. I have applied these tools to examining arc rocks from Volcán de Fuego and the Cascades to gabbroic xenoliths from Iceland. I enjoy thinking about how statistical and computational tools can make petrologic interpretations more open, reproducible, and uncertainty-aware. More recently, I have been applying machine learning to multi-parameter geophysical records from active volcanoes to detect and forecast eruptions. When I am not thinking about rocks, I enjoy hiking, rowing, fermenting things, and playing with sound.
Personal Website: sarahshi.github.io
Education and Work:
Data Science Fellow in the Geoinformatics Research Group, Lamont-Doherty Earth Observatory, Columbia University
MPhil in Earth Science, Euretta J. Kellett Fellow, University of Cambridge
B.A. in Earth and Environmental Science, Columbia University
Advisor: Penny Wieser
Role:
