Information recognition method and device based on attention module

Abstract

This application discloses an information recognition method, apparatus, storage medium, and electronic device based on attention modules. Embodiments of this application can be applied to various scenarios such as vehicle-mounted systems, cloud technology, artificial intelligence, intelligent transportation, and assisted driving, for example, in voice scenarios based on parallel computing. The method includes: acquiring target media resource features of a target media resource, inputting the target media resource features into a target information recognition model, processing the target media resource features through N layers of attention modules to obtain a target representation vector, wherein the i-th layer attention module is used to determine the i-th layer representation vector output by the i-th layer attention module based on a set of shared parameters and an i-th set of non-shared parameters, and determining the target information recognition result based on the target representation vector, wherein the target information recognition result represents the target information recognized from the target media resource. This application solves the technical problem in related technologies where attention recognition models suffer significant performance loss due to the effort to accelerate the computation process.

Images