Title:
METHOD AND SYSTEM FOR EMBEDDING WITH SMART MEMORY POOL
Document Type and Number:
WIPO Patent Application WO/2022/082493
Kind Code:
A1
Abstract:
Methods, systems, and apparatus, including computer programs encoded on computer storage media, for performing embedding operations, are described. An exemplary method may be implemented on a computing system comprising a plurality of interconnected electronic cards each storing at least a portion of one or more embedding tables, and the method comprising: obtaining a machine-learning job for a machine-learning model, the machine-learning job comprises a plurality of features, wherein the plurality of features comprise a plurality of sparse features; performing, by the plurality of electronic cards, a plurality of embedding tasks based on the one or more embedding tables to obtain a plurality of embedding results; aggregating the obtained embedding results to obtain input for the machine-learning model to process the machine-learning job. By implementing near-memory computation in a memory pool for large embedding processing, the method provides technical benefits by solving memory capacity and bandwidth challenges and reducing latency.
Inventors:
ZHU DENGCHENG (US)
ZHENG JUN (US)
DU YIN (CN)
SUN HUA (US)
ZHENG JUN (US)
DU YIN (CN)
SUN HUA (US)
Application Number:
PCT/CN2020/122435
Publication Date:
April 28, 2022
Filing Date:
October 21, 2020
Export Citation:
Assignee:
ALIBABA GROUP HOLDING LTD (CN)
International Classes:
G06F17/16
Foreign References:
| US10146738B2 | 2018-12-04 | |||
| US20170316312A1 | 2017-11-02 | |||
| US20180300600A1 | 2018-10-18 | |||
| CN109871485A | 2019-06-11 |
Attorney, Agent or Firm:
BEIJING TSINGYUANHUI INTELLECTUAL PROPERTY LAW FIRM (CN)
Download PDF: