Adaptive error-bounded compression for recommendation all-to-all traffic

What changed

Combined GPU vector-aware LZ/entropy compression with table-wise and iteration-wise error-bound adjustment for embedding lookups and gradients communicated during distributed recommendation training.

Scientific scope

Combined GPU vector-aware LZ/entropy compression with table-wise and iteration-wise error-bound adjustment for embedding lookups and gradients communicated during distributed recommendation training. Pointwise reconstruction error bounds do not establish general training convergence or accuracy preservation. Compression overhead and distribution changes matter. Communication acceleration is distinct from end-to-end training speedup; reported small quality loss is dataset/model-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. Pointwise reconstruction error bounds do not establish general training convergence or accuracy preservation. Compression overhead and distribution changes matter. Communication acceleration is distinct from end-to-end training speedup; reported small quality loss is dataset/model-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.

• Accelerating Communication in Deep Learning Recommendation Model Training with Dual-Level Adaptive Lossy Compression

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