Teaching
Teaching
In the ChatGPT era, I think the most valuable part of a classroom is the people in it, the peers and colleagues you get to think alongside. So in the lectures I help teach, I try to lean on participation and discussion rather than one-way delivery.
Scientific machine learning: a mini lecture series
I taught several lectures within Physics-Informed Machine Learning at UC San Diego, introducing the core ideas of scientific machine learning. The full lectures are on YouTube; here is a quick tour of what each one covers. These are recordings of live classes, so student questions and discussion are woven throughout. In the participation-first spirit above, they are real teaching sessions rather than polished videos.
Watch the full series on YouTubeNeural ODEs
How treating a network's depth as continuous time turns it into a dynamical system that an ODE solver integrates and the adjoint method trains efficiently.
Jump to a highlight
Neural operators
Learning maps between function spaces rather than between fixed vectors, so one model transfers across discretizations. Builds up to DeepONet and the Fourier Neural Operator.
Jump to a highlight
Jump to a highlight
Physics-informed neural networks
Embedding a PDE's residual directly in the training loss so a network can solve or identify differential equations with little or no data, and where the approach tends to break down.
Jump to a highlight
Foundation models
What foundation models are, how pretraining at scale produces general-purpose representations, and what their rise means for scientific computing.
Jump to a highlight