Data-driven statistical sharding for industry-scale neural recommendation

Geet Sethi

Data-driven statistical sharding for industry-scale neural recommendation

Stanford University, 2023

Abstract

This dissertation uses workload statistics to partition and place the large embedding tables used by industry-scale neural recommendation models across tiered memory systems, improving training efficiency and scalability.

Materials

Stanford Library

Bibtex

@phdthesis{2023datadrivenstatisticalshardingforindustryscaleneuralrecommendation,
  title = {Data-driven statistical sharding for industry-scale neural recommendation},
  author = {Geet Sethi},
  school = {Stanford University},
  year = {2023},
}

Hi, we are the Stanford MAST team

Data-driven statistical sharding for industry-scale neural recommendation

Abstract

Materials

Bibtex