Home▸Machine Learning & Neural Networks▸CMA-ES: Covariance Matrix Adaptation Evolution Strategy

CMA-ES: Covariance Matrix Adaptation Evolution Strategy

Watch CMA-ES search a 3D loss landscape: a population of candidates sampled from an evolving Gaussian, its mean and covariance ellipse adapting generation by generation toward the optimum.

Machine Learning & Neural Networks3DAdvanced60 FPS📱 Mobile-adapted⇄ 2D version

CMA-ES is one of the most effective derivative-free optimizers for black-box hyperparameter and continuous-parameter search: it needs no gradients, just a fitness score per candidate. This simulator runs the real update equations — weighted-mean recombination, rank-μ covariance adaptation, and cumulative step-size control — over a live 3D loss landscape (Sphere, Rosenbrock or Rastrigin), rendering every sampled candidate as a point on the surface and the evolving Gaussian's 1-σ ellipse on the base plane so you can watch the search distribution stretch, rotate and shrink toward the optimum generation by generation.

⚙ Under the hood

Watch CMA-ES search a 3D loss landscape: a population of candidates sampled from an evolving Gaussian, with its mean, step size and covariance ellipse adapting generation by generation toward the optimum on Sphere, Rosenbrock or Rastrigin surfaces.

cma-esevolution-strategyhyperparameter-optimizationblack-box-optimizationgaussian-processmachine-learning

3D · Three.js / WebGL renderer · 60 FPS target · runs fully client-side, no install