Yang et al.: sequence-parallel training of delta-rule linear models

What changed

Used a memory-efficient representation of Householder-matrix products to parallelize delta-rule linear-transformer training across sequence positions.

Scientific scope

Used a memory-efficient representation of Householder-matrix products to parallelize delta-rule linear-transformer training across sequence positions. DeltaNet and the delta-rule fast-weight architecture precede this paper; the acquired contribution is efficient sequence-parallel training and scaling evidence, not invention of delta learning. Source-reported recall/quality gains are task-specific.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. DeltaNet and the delta-rule fast-weight architecture precede this paper; the acquired contribution is efficient sequence-parallel training and scaling evidence, not invention of delta learning. Source-reported recall/quality gains are task-specific.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Parallelizing Linear Transformers with the Delta Rule over Sequence Length
• Yang et al.: sequence-parallel training of delta-rule linear models — official method source

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction