AWhat changed
Used scaled stochastic ternary gradients with shared scaling/parameter localization, plus layerwise quantization and clipping to reduce synchronization traffic in distributed training.
Scientific scope
Used scaled stochastic ternary gradients with shared scaling/parameter localization, plus layerwise quantization and clipping to reduce synchronization traffic in distributed training. Unbiased rounding assumes the specified unmodified gradient/scaler distribution; clipping changes the target estimator. Original convergence needs a single attracting minimum, summable-squared but non-summable steps and a stronger mixed-norm gradient bound, not arbitrary nonconvex neural loss. Aggregated multiworker gradients are not necessarily ternary; measured accuracy and modeled communication ratios are task/hardware-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. Unbiased rounding assumes the specified unmodified gradient/scaler distribution; clipping changes the target estimator. Original convergence needs a single attracting minimum, summable-squared but non-summable steps and a stronger mixed-norm gradient bound, not arbitrary nonconvex neural loss. Aggregated multiworker gradients are not necessarily ternary; measured accuracy and modeled communication ratios are task/hardware-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.
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