Averaged stochastic gradient descent
Algorithms AIMech:AveragedStochasticGradientDescent DRAFT
Aliases: ASGD
A stochastic gradient optimization method that maintains an average of parameter iterates in addition to the current iterate.
Identity Notes & Concept Scope
ASGD adds parameter-iterate averaging to stochastic gradient descent. It is not the same operation as averaging gradients in a momentum buffer.
Claims & Grounded Evidence
1 Claim
Mechanism
#update-rule
The inspected implementation updates the current parameters by a scheduled gradient step and maintains an auxiliary parameter average, with averaging weights controlled by the iteration count and a configurable starting point.
⚠️ Curated Limitations & Negative Results
- This record covers iterate averaging in the cited ASGD implementation, not every stochastic weight averaging schedule.
- No convergence rate or generalization advantage is asserted.
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
|
sha256:411369236bca…
|
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/AveragedStochasticGradientDescent.yaml