Have a UROP opening you would like to submit?
Please fill out the form.
Informative Ensemble Learning to Discover Dynamics from Data
12: Earth, Atmospheric and Planetary Sciences
Sai Ravela, email@example.com
We are interested in learning Neural Dynamical Systems, dynamical systems described in part by neural networks to model physical processes using their observed noisy data from sensors or as simplified models from simulations of complex models. We are particularly interested in developing such systems as models of natural hazards -- hurricanes, floods, and other perils -- but other idealized systems are also of interest. Recent UROP work describes one approach, e.g., arXiv:2008.09915, and award-winning solutions to 12.S592's PSETs, e.g., arXiv:2008.05590, describe others :) In this project, you will develop neural dynamical systems using an entirely new way to train neural networks to learn dynamics from simulated and observed time series with non-stationary statistics. You could be interested in theoretical insights or methodological development or application.
Experience with programming in python and Matlab would be a great asset. Knowledge of Linear Algebra, Stochastic Processes, or Optimization will be fantastic. This project is well suited for rising juniors or seniors and is particularly well suited for a segway into an undergraduate thesis and MENG. You are ready to commit at least 10 hours a week to the project, which includes attending 1x1 and group meetings over zoom, and seminars.