LargeVis
Algorithms AIMech:LargeVis DRAFT
A dimensionality-reduction and graph-visualization algorithm that constructs an approximate weighted nearest-neighbor graph and learns low-dimensional coordinates with sampled attractive-edge and repulsive-pair likelihood terms.
Identity Notes & Concept Scope
LargeVis combines a specific approximate-neighbor construction with a probabilistic layout objective. It is distinct from t-SNE’s normalized KL objective, UMAP’s locally scaled graph weights, LINE graph representations, and generic nearest-neighbor search. All 442 pinned AIO labels/scoped synonyms and 114 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
1 Claim
Mechanism
#method-scope
Initialize approximate neighbors with random projection trees and refine candidate neighbors through neighbors-of-neighbors exploration. Assign locally normalized Gaussian weights with a perplexity-controlled bandwidth and symmetrize them. Model edge likelihood using a decreasing function of embedded distance, sample positive edges in proportion to their weights, draw negative endpoints with degree-to-the-three-quarters weighting, and optimize the sampled log objective using asynchronous stochastic gradients.
⚠️ Curated Limitations & Negative Results
- Graph construction is approximate; the reported near-perfect neighbor recovery on selected datasets is not a guarantee for arbitrary input.
- The sampled objective in equation (6) replaces the all-negative-pairs expression in equation (5); it is not asserted to be an unbiased evaluation of every original negative term.
- The linear layout-cost statement assumes fixed embedding dimension and negative-sample count with a number of updates proportional to points; it does not establish linear cost for all graph construction or convergence.
- Perplexity, graph quality, negative distribution, optimizer and chosen low-dimensional probability function affect results. No universal geometric fidelity or benchmark superiority is asserted.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:DimensionalityReduction
Dimensionality Reduction
|
related_match | A specific algorithm for constructing lower-dimensional representations, associated with the pinned dimensionality-reduction task. The target inherits UnsupervisedLearning under MachineLearningTask; related_match avoids treating the algorithm as identical to or a subclass of that task. Shared embedding components do not make this concept identical to PCA, MDS or t-SNE. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
original-paper
|
Jian Tang, Jingzhou Liu, Ming Zhang, Qiaozhu Mei
|
primary paper
full_text
|
ID: doi:10.1145/2872427.2883041
sha256:556c0e20fd7d…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T05:21:17Z
Updated: 2026-10-11T05:21:17Z
Source File:
data/curated/LargeVis.yaml