Title:
NEURAL NETWORK MODEL TRAINING METHOD AND SYSTEM, AND PREDICTION METHOD AND SYSTEM
Document Type and Number:
WIPO Patent Application WO/2020/020088
Kind Code:
A1
Abstract:
A training method and system for a neural network model comprising three levels of models, and a prediction method and system. The training method comprises: acquiring a training data record; generating a feature of a training sample according to attribute information of the training data record, and using a mark of the training data record as a mark of the training sample; and training a neural network model by means of a set of training samples, wherein in the process of training the neural network model, a feature information representation of each feature itself is respectively learned by means of a plurality of bottom-layer neural network models (112) included in a first-level model (110) of the neural network model, an interaction representation between corresponding input items is respectively learned by means of a plurality of middle models included in a second-level model (120) of the neural network model, a prediction result is learned at least based on the interaction representation output by the second-level model (120) by means of a third-level model (130) of the neural network model, and the neural network model is adjusted at least based on the difference between the prediction result and the mark.
Inventors:
LUO YUANFEI (CN)
TU WEIWEI (CN)
CAO RUI (CN)
CHEN YUQIANG (CN)
TU WEIWEI (CN)
CAO RUI (CN)
CHEN YUQIANG (CN)
Application Number:
PCT/CN2019/096971
Publication Date:
January 30, 2020
Filing Date:
July 22, 2019
Export Citation:
Assignee:
FOURTH PARADIGM BEIJING TECH CO LTD (CN)
International Classes:
G06N3/04
Foreign References:
| CN105787500A | 2016-07-20 | |||
| CN107220707A | 2017-09-29 | |||
| CN107480774A | 2017-12-15 | |||
| US20180068216A1 | 2018-03-08 |
Attorney, Agent or Firm:
ZYX INTELLECTUAL PROPERTY LAW FIRM (CN)
Download PDF: