Skip to main content

Module 07: CMA-ES: The Crown Jewel - Learning the Shape of the Search

CMA-ES is the gold standard for black-box continuous optimization. It learns the covariance structure of the fitness landscape, adapting not just step size but the entire shape of the search distribution. We build it piece by piece.

Learning Objectives​

  1. Understand the covariance matrix and its role in adaptation
  2. Implement the rank-one update (evolution path)
  3. Implement the rank-mu update (population information)
  4. Code cumulative step-size adaptation (CSA)
  5. Use pycma for production optimization

Concept Explanation​

Coming soon.

Code Examples​

Coming soon.

Exercises​

Coming soon.

Milestone Checklist​

  • Built CMA-ES from scratch using only NumPy
  • Understand rank-one and rank-mu updates
  • Can tune CMA-ES hyperparameters (sigma0, popsize)
  • Used pycma on a real optimization problem

Was this page helpful?