Dr. David Greenberg

I am interested in what machine learning (ML) can do for the geosciences. Beyond simple problems of predicting, completing or classifying data, I believe ML has significant untapped potential for improving how numerical simulations are developed and used. In particular, parameter tuning, parametrization development, uncertainty quantification and data assimilation are all tasks where a hybrid approach combining the data-efficiency and flexibility of ML with the inductive biases of physics, chemistry and biology can be immensely helpful. Conversely, I think the geosciences also have a great deal to teach ML, and I’m excited by the conceptual and technical challenges raised by the sheer scale of geoscientific datasets and simulations.