Lazy learning¶

A machine-learning strategy that postpones generalization from stored training examples until a prediction query arrives.

Version: v2 · 2026-09-19 · History
Domain-specific #: 5281
Domain group: Applied Sciences & Engineering
Origin domain: Computer Science & Software Engineering
Subdomain: Machine Learning → Computer Science & Software Engineering

Core Idea¶

Nearest-neighbor and case-based learners shift computation and model choice to query time, enabling rapid updates but increasing storage and latency and making distance and local neighborhood choices constitutive. The system retains labeled cases, retrieves those relevant to a new query and constructs a local prediction or explanation only for that query. The abstraction is therefore identified by a declared carrier, a transformation or constraint over that carrier, and an invariant that tells an analyst whether the named structure is genuinely present.

How would you explain it like I'm…

Look-It-Up Learning

Imagine you keep a big box of photos of animals with their names written on the back. When someone shows you a new animal, you find the photos that look most like it and use their names. You don't figure anything out ahead of time — you only think when someone asks. That's lazy learning.

Remember Examples, Decide Later

Lazy learning is a way for computers to learn by saving examples instead of building a summary first. When a new question comes in, the computer finds the saved examples most similar to it and uses them to make a guess. Adding new examples is quick, because you just store them. But it needs lots of memory, and each answer takes longer because the searching happens at question time. How you decide what counts as 'similar' matters a lot.

Query-Time Case Learning

Lazy learning, used by nearest-neighbor and case-based methods, postpones the real work until a query arrives. Instead of training a general model, the system keeps labeled cases, retrieves the ones closest to the new input, and builds a local prediction for that input only. Contrast this with eager learning, which builds a model before seeing any queries. Lazy learning updates easily—just add cases—but costs more memory and slower answers. Because predictions depend on neighbors, the choice of distance measure and neighborhood size is part of what defines the model.

Lazy learning covers nearest-neighbor and case-based methods that defer generalization and model choice to query time. The system retains labeled instances; for each query it applies a similarity or retrieval rule, selects a local neighborhood, weights its members, and produces a local prediction or explanation. Because no global model is fitted, incorporating new data is cheap, but memory grows with the stored cases and query latency rises, often motivating indexing structures. The distance metric, neighborhood size and weighting scheme are not incidental tuning knobs but constitutive of what the model predicts. A proper specification therefore states the task and data, stored representation, retrieval rule, neighborhood and weighting, prediction rule, update policy, indexing, latency and memory, and the validation split and calibration.

Scope of Application¶

Lazy learning belongs to machine learning and is useful where the analyst can specify the typed machine learning carrier, defining objects and relations, parameters, conventions, evidence, boundary cases, and comparison targets, then evaluate the task and data, stored representation, similarity or retrieval rule, query-time neighborhood and weighting, prediction rule, update policy, indexing, latency and memory, validation split and calibration are explicit. The scope is broad within that domain but bounded by the need for the task and data, stored representation, similarity or retrieval rule, query-time neighborhood and weighting, prediction rule, update policy, indexing, latency and memory, validation split and calibration are explicit. The entry records a descriptive analytical identity; practical use requires the governing domain's evidence, standards, and safety obligations.

Clarity¶

The abstraction clarifies a crowded vocabulary by making the task and data, stored representation, similarity or retrieval rule, query-time neighborhood and weighting, prediction rule, update policy, indexing, latency and memory, validation split and calibration are explicit the center of the account. A claim should name the carrier, the governing operation or relation, the applicable assumptions, and the recognition test.

Manages Complexity¶

Without the abstraction, an analyst must reason directly over many local details: the carrier roles, admissibility assumptions, competing conventions, derived invariants, boundary cases, and proof or validation obligations specific to Lazy learning. Lazy learning compresses them into the roles in the structural signature. That compression permits comparison across instances without erasing the variables that determine validity. It also exposes which details may be varied safely and which are constitutive.

Abstract Reasoning¶

Identify the carrier. State what the elements, states, objects, or observations are: the typed machine learning carrier, defining objects and relations, parameters, conventions, evidence, boundary cases, and comparison targets. Reject examples whose alleged carrier belongs to a different problem.
Lock the constitutive rule. Express the task and data, stored representation, similarity or retrieval rule, query-time neighborhood and weighting, prediction rule, update policy, indexing, latency and memory, validation split and calibration are explicit independently of one notation or implementation.

Knowledge Transfer¶

Knowledge transfers strongly among subfields of machine learning because they reuse the typed machine learning carrier, defining objects and relations, parameters, conventions, evidence, boundary cases, and comparison targets, The system retains labeled cases, retrieves those relevant to a new query and constructs a local prediction or explanation only for that query., and type the carrier, state every parameter and convention in the definition, test that the task and data, stored representation, similarity or retrieval rule, query-time neighborhood and weighting, prediction rule, update policy, indexing, latency and memory, validation split and calibration are explicit, compare the nearest accepted identity, and report counterexamples, uncertainty, and limiting cases.

Relationships to Other Abstractions¶

Current abstraction Lazy learning Domain-specific

Parents (1) — outgoing typed relations

Lazy learning is a kind of Lazy Evaluation Prime

It remains its own entry because its identity is fixed by the task and data, stored representation, similarity or retrieval rule, query-time neighborhood and weighting, prediction rule, update policy, indexing, latency and memory, validation split and calibration.

Hierarchy paths (2) — routes to 2 parentless roots

Lazy learning → is a kind of Lazy Evaluation → is a kind of Postponement → presupposes Optionality → presupposes Reversibility and Irreversibility

Show alternative path (1)

Neighborhood in Abstraction Space¶

Lazy learning sits in a crowded region of the domain-specific corpus (8^th percentile for distinctiveness): several abstractions share nearly its structure, so a description that fits it tends to fit its neighbors too.

Family — Machine Learning Models & Representations (25 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-25