Meta-Feature Extraction
Updated 2026-08-10
INTRODUCTION
English translation pending.
CORE DEFINITION
A technique from meta-learning and automated machine learning. Meta-features describe properties of the data rather than of individual observations, such as feature correlations, class imbalance, signal-to-noise ratio, and distributional skew. The core proposition is that problems have shapes, and a shape predicts which model families and hyperparameters tend to work, so prior tasks can inform new ones. The key qualification is that the mapping holds only when the new task resembles the training tasks.
SCAFFOLDING EFFECT
Reduce cognitive load
- Problem profiling: compute dataset-level properties before selecting a model family. - Prior retrieval: find historically similar tasks and see what worked there. - Cold-start shortcut: use the profile to skip exhaustive search on a new task.
Anchor fast decisions
Model performance depends on structural properties of the data, such as linearity, dimensionality, and class balance, rather than on the specific values. Compressing a dataset into a small vector of such properties creates a representation that can be compared across tasks, so performance observed on similar past tasks transfers as a prior. This converts model selection from blind search into informed retrieval.
MINIMUM ACTION
In progress 0/1Practice this model in one real situation:
account_treeGenealogyexpand_more
menu_bookReferencesexpand_more
Source support: Explicit
- en.wikipedia.orghttps://en.wikipedia.org/wiki/Meta-learning_(computer_scienceverified
PRIVATE NOTES · Only visible to you
SAVED Q&A
ENTRY Q&A · Private saving available
Ask with a clear boundary
thinkingmodels answers from published entry context only.
Your question is sent to thinkingmodels. The answer uses public entry context only.
RELATED MODELS