| 摘要 |
A system and method for a training parameter set in a neural network. The system comprises a master control set (110) and N training node sets (120). The master control set is used for controlling a training process and storing a data set and a parameter set used in training. The master control set comprises M master control nodes, every two of the M master control nodes are in communication connection, and at least one of the M master control nodes is used for backing up the parameter set. The training node sets comprise multiple training nodes. The training nodes are used for performing training according to the data set and the parameter set delivered by the master control set, and sending training results to the corresponding master control nodes. The system and the method can avoid a case in which the whole training fails due to the failure of a master control node, thereby improving the reliability of the training process; training is performed in parallel by configuring multiple training node sets, thereby improving the training efficiency. |
