PARAMETER UPDATE METHOD DEVICE AND ELECTRONIC EQUIPMENT FOR MULTITASK MODEL

摘要

The present application discloses a parameter update method, apparatus and electronic device of a multi-task model, and relates to the field of computer technology, in particular, to the field of deep learning and artificial intelligence. A specific implementation method is to obtain a training sample set; sequentially putting each sample into a corresponding sample queue according to the task to which each sample belongs; Under the condition that the number of samples in the random sample sequence reaches the training data demand, using the data in the random sample sequence, the shared network layer in the multitask model and the target subnetwork layer corresponding to the task related to the random sample sequence are to train for generating a model parameter update gradient corresponding to a task related to an arbitrary sample sequence; and updating the parameters of the shared network layer and the target subnetwork layer in the parameter server according to the update hardness. Thus, through this parameter update method of the multitask model, when distributed training on the multitask model, local parameter update is implemented to improve the precision of the multitask model.