Interactive Notes
Interactive technical notes on machine learning, physics, and the math in between.
-
Autodiff as pushforward and pullback
Tangents, cotangents, and what composes: jvp and vjp as exact pushforward and pullback on real and complex space, worked out down to an exterior derivative in a few lines of JAX.
autodiff · differential-geometry · jax · machine-learning -
Tutorial: Generative models as transport
Generative models, from change of variables, continuous flows, and the fixed-path trick behind diffusion models to flow matching, physics applications, and few-step distillation.
generative-models · diffusion-models · machine-learning · physics · tutorial -
Analytic bijections for smooth and interpretable normalizing flows
Smooth, closed-form-invertible scalar bijections on the whole real line: building blocks for normalizing flows that keep smoothness, an unbounded domain, a cheap exact inverse, and local expressivity all at once.
normalizing-flows · machine-learning · generative-models · physics