Model training method, device control method, apparatus, and electronic device

Abstract

The application provides a model training method, a device control method, a device and an electronic device, and belongs to the technical field of data processing. By adopting the technical scheme provided in the embodiments of the application, an initial model to be trained is used to process sample problems in acquired training samples, at least one group of prediction results including a prediction type and a prediction answer are determined, then a first confidence is determined through a sample decision type and the prediction type, a second confidence is determined through a sample answer and the prediction answer, and then the initial model is updated through the first confidence and the second confidence to obtain a target model used for device control according to an input target problem. Through double confidence joint optimization, all-round supervision from a macro intention to a micro semantic can be realized, the thinking path of the model is effectively constrained, a reliable semantic foundation is laid for subsequent device control tasks, and the accuracy of the intelligent home system in controlling devices in a complex home environment is improved.

Images