Data processing method, and training method and device of deep learning model

Abstract

The present disclosure provides a data processing method, a training method and device of a deep learning model, which can be used in the field of artificial intelligence technology. The data processing method comprises: performing convolution processing on to-be-processed data corresponding to a processing task to obtain data features; performing fusion importance analysis on the data features to obtain a feature index matrix, wherein each element in the feature index matrix represents the fusion importance of each data feature with other global data features; determining a target feature sequence from a self-attention fusion parameter matrix according to the feature index matrix, wherein the self-attention fusion parameter matrix comprises a query feature parameter matrix and a key feature parameter matrix, and the target feature sequence comprises a plurality of feature parameter pairs composed of a target query feature parameter and a target key feature parameter; performing fusion processing on the data features based on the target feature sequence to obtain fused features; and processing the fused features according to the processing task to obtain a processing result.

Images