When
Where
Speaker: Edward McDugald, Program in Applied Mathematics
Title: Attention-based reconstruction of high-resolution tsunami waves from sparse tsunameter networks
Abstract: The Senseiver is an attention-based neural network designed for sparse sensing tasks in a variety of physical contexts. In this talk, I will discuss recent experiments with the Senseiver to reconstruct full-field tsunami waves given incredibly sparse (~2-12 observation points) measurements. I will provide a high-level overview of the Senseiver architecture, as well as a summary of recent ML-based solutions for tsunami forecasting. I will then discuss tsunami forecasting methods relying solely on tsunameters (ie, buoys in the ocean that measure wave height), and suggest how the Senseiver can be deployed to solve the sparse sensing problem in the tsunami data assimilation method.
Followed by an organizational meeting.