Covariance Matrix Adaptation Evolution Strategy
Algorithms AIMech:CovarianceMatrixAdaptationEvolutionStrategy DRAFT
Aliases: CMA-ES
An evolution strategy for continuous optimization that samples a multivariate normal search distribution and adapts its mean, covariance and global step size from ranked offspring and accumulated selected-step paths.
Identity Notes & Concept Scope
This record describes CMA-ES through the 2001 weighted-recombination algorithm and its author-supplied errata. That algorithm’s covariance update is based on a rank-one evolution-path outer product; later rank-mu, active covariance, restart and limited-memory variants are not silently imported. The search distribution is not a predictive model of objective values. Derandomized adaptation does not make offspring sampling deterministic. No normalized identity match was found across all 442 pinned AIO labels/scoped synonyms and 270 preceding native IDs/names/aliases. Full-checkout fixed-string searches included ignored and hidden files in worktree and main, returning 19 and 19 lines respectively. These were prior candidate leads, source discussions and neighboring DEHB/FABOLAS method mentions. Eight exact AIO CLASS nodes with scoped synonyms and outgoing edges and seven native neighboring mechanisms were inspected. Absence is bounded to this snapshot and semantic neighborhood.
Claims & Grounded Evidence
1 Claim
Mechanism
#method
Initialize a search mean, positive global step size, identity covariance and zero evolution paths. Sample a generation of Gaussian offspring with covariance equal to the global step size squared times the covariance matrix. Rank offspring by the minimization objective and recombine the selected best offspring with positive normalized weights to set the next mean. Accumulate a normalized selected mean displacement into a covariance evolution path. In the inspected 2001 procedure, blend the old covariance with the outer product of that path. Accumulate a separate path with covariance scaling removed and adjust the global step size exponentially according to its length relative to the expected length of a standard normal vector. Obtain the covariance axes and scales by eigendecomposition for subsequent sampling and repeat under the chosen stopping criterion. No derivatives of the objective are required.
⚠️ Curated Limitations & Negative Results
- This is a source-specific account of the 2001 algorithm, not an exact implementation contract for all contemporary CMA-ES software. Its positive recombination weights, rank-one covariance update and cumulative step-size rule are distinguished from later extensions. The author errata qualify the default weight formula when the selected-parent count exceeds half the offspring population and revise a suggested ordering of cumulation parameters.
- Initial mean and step size matter. Weak, noisy or selection-irrelevant feedback and numerical conditioning can impair covariance adaptation or collapse short search directions. Full covariance storage grows quadratically with dimension; eigendecomposition and expensive function evaluations add computational costs.
- Constraint handling and termination require explicit choices. The paper discusses resampling feasible offspring but cautions about boundary optima. Ranking and distribution adaptation do not guarantee a global optimum, monotonic improvement of every generation, or a calibrated inverse-Hessian estimate. No benchmark speedup or author-code execution is asserted.
Artificial Intelligence Ontology (AIO) Grounding
Explicit AIO Snapshot Assessment: No Match
Rationale: No inspected AIO target captures this general optimization procedure or a defining component. The assessment distinguishes evolutionary language models, learning/training tasks and predictive models from the native algorithm; it is bounded to this pinned vocabulary.
Search Scope: All 442 pinned AIO CLASS identities and scoped synonyms, all 270 preceding maintained identities, ignored/hidden checkout text, and semantic inspection of eight exact AIO targets plus seven neighboring native mechanisms.
| Examined AIO Candidate | Rejection Reason |
|---|---|
AIO:EvolutionaryLLM
|
The CLASS definition is a large language model with evolutionary optimization. Its broad related synonym evolutionary algorithms does not make a general continuous optimizer a language model or an exact vocabulary match. |
AIO:TrainingStrategy
|
The pinned definition concerns preprocessing/training of machine-learning models. General black-box continuous optimization is not inherently model training; an optional application does not establish a necessary mapping. |
AIO:MachineLearningTask
|
The target denotes building methods that learn from data. These algorithms can optimize arbitrary evaluated functions without a learning task, so shared optimization vocabulary alone is insufficient. |
AIO:MetaLearning
|
Adapting a search distribution or optimizer state does not require learning from metadata about machine-learning experiments as in the pinned definition. |
AIO:ActiveLearning
|
Candidate evaluation is not the defined task of querying labels for data points. |
AIO:ReinforcementLearning
|
A possible policy-search application does not turn the general optimizer into the pinned learning task; no environment/reward-policy interface is inherent to the full identity. |
AIO:RandomForest
|
Population members and search distributions are not a predictive decision-tree ensemble; no random-forest surrogate is required. |
AIO:EvaluationBias
|
This is a bias concept, not the optimizer. Possible evaluation or sampling problems are limitations rather than identity mappings. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
method-source
|
Nikolaus Hansen, Andreas Ostermeier
|
primary paper
full_text
|
sha256:423664eab4f6…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T12:18:29Z
Updated: 2026-10-11T12:18:29Z