← all writing

Convex optimization

math

In progress

Nothing written up yet. This is a placeholder for notes I am actively taking.

Working through the Stanford convex optimization textbook (Boyd & Vandenberghe). Also using this as a resource.

What I want out of it:

  • convex sets and convex functions — what actually makes a problem convex, rather than pattern-matching on the form
  • Lagrangian duality, and why the dual is always convex even when the primal is not
  • KKT conditions, stated properly rather than recalled as a slogan
  • gradient descent, Newton’s method, and interior point methods, and why the choice depends on problem structure
  • where the convexity assumption breaks in deep learning, and what survives anyway

The reason this sits under foundations rather than a curiosity: nearly everything in what models optimize is a non-convex problem attacked with tools built for convex ones, and I would rather understand the gap than route around it.