Homeomorphism Methods
One change of coordinates, three uses: move from a simple ball to a feasible set, then explore prediction recovery, optimization, and constrained sampling.
Read the visual guideResearch notes
Blogs
Learning, optimization, and generation through geometry. Concrete examples, visual explanations, and a closer look at when the methods help.
Hard-Constrained Machine Learning
From constrained prediction to constrained generation: use a budget triangle to compare penalties, projection, and feasible coordinates, then follow constraints into diffusion and flow sampling.
Read the visual guide