Model training method and device based on sampling time of sample

The invention relates to the technical field of machine learning, and provides a model training method and device based on sampling time of samples. The method comprises the steps that the sampling time of each batch in model training is determined, and the model training comprises training of multiple batches; according to the sampling time of each batch, a corresponding weight table is generated for each batch, the weight table comprises the weight corresponding to each sample in the current batch, and each batch comprises a plurality of samples; calculating a target loss value corresponding to each batch through a gradient descent algorithm based on the plurality of samples of each batch and the weight table corresponding to each batch; and based on the loss value corresponding to each batch, completing training of each batch by using a gradient back propagation algorithm. By adopting the technical means, the problem that the precision of the final model cannot be guaranteed in the model training related t