> ML_ALGORITHM // COVARIANCE-MATRIX-ADAPTATION-CMA-ES_v1.0
Covariance Matrix Adaptation Evolution Strategy (CMA-ES)
The premier derivative-free black-box continuous optimization algorithm that models local curvature by continuously updating a full covariance matrix.
Evolution Strategiesevolutionary-searchhigh-intrinsicsmall (<1k)
Back to All AlgorithmsComputational Complexity
Training Complexity:O(generations * (lambda * fitness + d^2)) where d < 1000
Inference Complexity:O(1)
Hardware Profile
CPU Friendly:Yes
Requires GPU:No
Memory Footprint:low
Interpretability & Data
Interpretability Tier:high-intrinsic
Training Data Needs:small (<1k)
Interpretability Assessment
Adapted covariance matrix directly recovers the inverse Hessian of the local objective landscape.
Suitable Tasks & Supported Modalities
Suitable Tasks:
continuous blackbox optimizationreinforcement learning policy searchhyperparameter tuning
Supported Modalities:
tabular
Implementing Libraries
Foundational Literature
Common Pitfalls & Warnings
- O(d^2) covariance matrix operations and memory make standard CMA-ES intractable for deep neural networks with >10,000 parameters