Parameter-Free Zeroth-Order Optimization with Ellipsoidal Sampling

Name: Parameter-Free Zeroth-Order Optimization with Ellipsoidal Sampling
Uploaded: 2026-10-16T00:56:38.737593+00:00
Duration: 595 s
Description: POEM-ES is a new parameter-free zeroth-order optimization algorithm that improves upon POEM by using subspace preconditioning and ellipsoidal randomized sampling. It guides exploration along low-rank geometric structures, achieving dimension-reduced convergence rates under low-rank assumptions, making it more efficient for high-dimensional black-box problems.

POEM-ES is a new parameter-free zeroth-order optimization algorithm that improves upon POEM by using subspace preconditioning and ellipsoidal randomized sampling. It guides exploration along low-rank geometric structures, achieving dimension-reduced convergence rates under low-rank assumptions, making it more efficient for high-dimensional black-box problems.

Open the full episode