Home / Curated / Algorithms / Locally Linear Embedding

Locally Linear Embedding

Algorithms AIMech:LocallyLinearEmbedding DRAFT

Aliases: LLE

A nonlinear dimensionality-reduction algorithm that learns sparse, sum-to-one linear reconstructions of each point from nearby points and computes embedding coordinates by minimizing reconstruction error with those same weights.

Identity Notes & Concept Scope LLE preserves local reconstruction relations rather than explicitly fitting all-pairs geodesic distances. It is distinct from Isomap, a linear PCA projection, Laplacian Eigenmaps, and a learned neural embedding layer. The expanded 2003 paper elaborates the same LLE identity; optional extensions are not automatically separate aliases. All 442 pinned AIO labels/scoped synonyms and 109 preceding native identities were compared, together with hidden/ignored-inclusive searches of both checkouts; source citations and pending frontier mentions were distinguished from existing concept identities.

Claims & Grounded Evidence

2 Claims

Mechanism #method-scope

Select neighbors, solve a constrained least-squares reconstruction for each input with weights zero outside its neighborhood and summing to one, then keep the weights fixed while minimizing embedded reconstruction error under centering and covariance constraints. The final step uses the smallest eigenmodes of (I − W)^T(I − W), excluding the constant translation mode. Weight regularization is needed for singular or poorly conditioned local fits.

supports Source: Nonlinear Dimensionality Reduction by Locally Linear Embedding · Locator: Equations (1)–(3), Figure 2 and notes 5–9, printed pp. 2323–2326 / PDF pp. 1–4

Limitation #local-fit-and-extension-limits

The expanded LLE paper explicitly distinguishes similarity-transform invariance from shear invariance, discusses neighborhood and connectivity failures, conditions singular Gram matrices with a weight penalty, and separates the fixed-data embedding from optional mappings to unseen points.

supports Source: Think Globally, Fit Locally: Unsupervised Learning of Low Dimensional Manifolds · Locator: Section 2 and Figure 4, pp. 122–125 / PDF pp. 4–7; Sections 4.1–4.3, pp. 131–135 / PDF pp. 13–17; Section 6, p. 141 / PDF p. 23

⚠️ Curated Limitations & Negative Results

Neighborhoods must adequately represent local structure; excessive neighborhood size can violate local linearity and sparse or disconnected neighborhoods require separate treatment.
The optimized weights are invariant to translation, rotation and homogeneous scaling of a neighborhood, not to arbitrary affine shears.
Global minimization of the stated quadratic embedding objective does not establish faithful recovery of an arbitrary manifold or a unique coordinate system when eigenspaces are degenerate.
The core algorithm embeds its supplied point set; applying it to new points requires an additional mapping or recomputation.

Artificial Intelligence Ontology (AIO) Grounding

Mapped AIO Classes

AIO Target Class	Mapping Relation	Mapping Rationale
`AIO:DimensionalityReduction` Dimensionality Reduction	related_match	A specific unsupervised embedding algorithm used for dimensionality reduction. The pinned target inherits the machine-learning-task hierarchy; this association does not assert algorithm/task identity or subclass membership. RepresentationLearning in this pin specifically denotes a deep neural network, and EmbeddingLayer denotes an integer-index lookup layer, neither of which defines this method.

Evidence Sources & Primary Literature

2 Sources

Source ID	Publication / Document	Type / Access	Reference & Checksum
`original-paper`	Nonlinear Dimensionality Reduction by Locally Linear Embedding Sam T. Roweis, Lawrence K. Saul	primary paper full_text	ID: doi:10.1126/science.290.5500.2323 sha256:45f23283c050…
`expanded-method`	Think Globally, Fit Locally: Unsupervised Learning of Low Dimensional Manifolds Lawrence K. Saul, Sam T. Roweis	primary paper full_text	ID: https://www.jmlr.org/papers/v4/saul03a.html sha256:399b54bc2ff2…

Curation Provenance

Curator: Codex

Method: ai_assisted

Model: GPT-6

Created: 2026-10-11T05:11:14Z

Updated: 2026-10-11T05:12:37Z

Source File: data/curated/LocallyLinearEmbedding.yaml