DATA-DISTRIBUTION-BASED JOINT DEEP LEARNING METHOD

摘要

Disclosed is a data-distribution-based machine deep learning method. According to the amount and type of data owned by each node, the degree of importance of the node during a training process is evaluated to guide the integration of transmission and a model; and according to the data distribution situations of participant nodes, the transmission frequencies of the nodes are caused to be inconsistent during the training process, such that a node with a relatively good data distribution situation spreads a model thereof as much as possible, otherwise more models of other nodes are received. The influence of data distribution imbalance on a training result is reduced, and since a node with a relatively bad data distribution situation spreads a model thereof as little as possible, network transmission can be reduced under the condition of not influencing the training result.