Title:
DEEP LEARNING TRAINING METHOD AND APPARATUS
Document Type and Number:
WIPO Patent Application WO/2018/153201
Kind Code:
A1
Abstract:
The present disclosure provides a deep learning training method and apparatus.
The deep learning training method comprises: in a forward propagation process of each iterative training, determining a loss value of each training data instance in batch training data instances; determining all difficult instances from the batch training data instances according to the loss value of each training data instance; and giving up learning characteristics of instances that are not difficult, and learning characteristics of all the difficult instances.
Inventors:
GAO YAN (CN)
LV DA (CN)
LUO SHENGMEI (CN)
LI WEIHUA (CN)
LV DA (CN)
LUO SHENGMEI (CN)
LI WEIHUA (CN)
Application Number:
PCT/CN2018/073955
Publication Date:
August 30, 2018
Filing Date:
January 24, 2018
Export Citation:
Assignee:
ZTE CORP (CN)
International Classes:
G06K9/62; G06N99/00
Foreign References:
CN106096538A | 2016-11-09 | |||
CN104992223A | 2015-10-21 | |||
CN103593474A | 2014-02-19 | |||
CN105608450A | 2016-05-25 | |||
US20150248764A1 | 2015-09-03 |
Attorney, Agent or Firm:
TEE&HOWE INTELLECTUAL PROPERTY ATTORNEYS (CN)
Download PDF:
Previous Patent: HLSTM MODEL-BASED ACOUSTIC MODELING METHOD AND DEVICE, AND STORAGE MEDIUM
Next Patent: DATA CACHING METHOD AND APPARATUS
Next Patent: DATA CACHING METHOD AND APPARATUS