Patent · US Active

Gap-aware mitigation of gradient staleness

US11631035B2 · kind B2 · utility

1Cited by

0References

11Claims

0Family size

Assignee

Technion Research & Development Foundation Limited · IL

Inventors

Assaf Schuster · Haifa, IL
Saar Barkai · Haifa, IL
Ido Hakimi · Haifa, IL

Key dates

Filing date	Nov 5, 2020
Grant date	Apr 18, 2023
Priority date	—
Expiry date	Nov 5, 2040

Classification

Technology area (CPC G)Physics
CPC primaryG06N3/098
WIPO fieldComputer technology
WIPO sectorElectrical engineering

Abstract

Disclosed embodiments are a computing system and a computer-implemented method for distributed training of a machine learning model over a plurality of computing nodes, in a plurality of iterations, characterized by gradient gap based mitigation of the gradient staleness problem. The disclosed method evaluates the staleness of the gradient based on the difference in gradients between a central point, for example an iteration's common starting point, and the points reached by the respective computing node during one or more iterations, and aggregates the update steps from the plurality of computing nodes, while giving more weight to computing nodes having a lesser change in the gradient.

Source: USPTO / EPO open patent data. Objective bibliographic and citation counts.