Limited-memory BFGS
Algorithms AIMech:LBFGS DRAFT
Aliases: L-BFGS, Limited-memory Broyden-Fletcher-Goldfarb-Shanno
A limited-memory quasi-Newton optimization method using a bounded history of parameter and gradient differences to compute approximate inverse-Hessian search directions.
Identity Notes & Concept Scope
L-BFGS is an optimization algorithm, not a neural architecture; bounded curvature history distinguishes it from full-memory BFGS.
Claims & Grounded Evidence
1 Claim
Mechanism
#update-rule
The implementation retains a bounded set of curvature pairs and applies a two-loop recursion to obtain a direction from the gradient without storing a dense inverse Hessian.
⚠️ Curated Limitations & Negative Results
- The cited implementation imposes parameter-group and device restrictions that are not mathematical requirements of every L-BFGS implementation.
- Line-search choices and termination tolerances affect a particular run; no universal convergence or speed guarantee is curated.
Artificial Intelligence Ontology (AIO) Grounding
Mapped AIO Classes
| AIO Target Class | Mapping Relation | Mapping Rationale |
|---|---|---|
AIO:TrainingStrategy
TrainingStrategy
|
related_match | This algorithm can implement parameter fitting, but the pinned TrainingStrategy definition uses a preprocessing genus. This is a contextual training relationship, not equivalence or a subclass assertion. |
Evidence Sources & Primary Literature
1 Source| Source ID | Publication / Document | Type / Access | Reference & Checksum |
|---|---|---|---|
pytorch-docs
|
PyTorch contributors
|
official documentation
documentation
|
sha256:d15dff277f15…
|
Curation Provenance
Curator: Codex
Method: ai_assisted
Model: GPT-6
Created: 2026-10-11T03:25:18Z
Updated: 2026-10-11T03:25:18Z
Source File:
data/curated/LBFGS.yaml