Meta-learning is a machine learning paradigm in which the learning algorithm is itself optimized using experience — typically measured as performance across a distribution of related tasks or training episodes. The system learns an induction bias, initialization, update rule, or representation that enables faster or more sample-efficient learning on new tasks. It is characterized by two nested loops: an inner loop that learns task-specific parameters from task data, and an outer loop that updates the meta-parameters based on performance across the task distribution. The mechanism of persistence is through published algorithms implemented in software and sustained as practice within the ML community. [formal: meta-lectio | substrate: mind | horizon: hours | explicit: yes | epoch: 0.89]
Accepted ontology entry
meta-learning
Meta-learning is a machine learning paradigm in which the learning algorithm is itself optimized using experience — typically measured as performance across a distribution of related tasks or training episodes. The system learns an inducti…
Definition
Why it is in scope
A human-made machine learning approach where a system is designed to learn how to learn — the map of techniques that use experience across multiple tasks or training episodes to improve the efficiency or effectiveness of subsequent learning, rather than learning a single fixed mapping from inputs to outputs.
Names and aliases
- meta-learningen · CANONICAL
Relations from this entry
- cmrhajc2b015x8aehc613gx7cINSTANCE_OF →
meta-learning is a specific kind of learning — learning how to learn, adapting learning processes themselves. A competent speaker would call meta-learning 'a kind of learning'. Nearest kind: learning.
- cmrg0scos00ef2a1nklfvbk7xINSTANCE_OF →
meta-learning is a specific kind of machine learning where models learn to learn — the meta-learning algorithm adapts the learning process itself. A competent speaker would call it a type of machine learning. The nearest kind is machine learning (no intermediate rung found in the ladder).
Relations to this entry
No accepted relations in this direction.
Record identity
- Created
- Aug 9, 2026, 4:35 AM UTC
- Content hash
- cde9d930104578815519bf8b1e9ad88b73ed18d3d79fde4df9ec803da6360bb9