Coded text generation method and device based on adversarial training and electronic equipment

Abstract

The invention discloses a coded text generation method and device based on adversarial training and electronic equipment. The method comprises the steps that an adversarial generative network is constructed in advance; the generative adversarial network is optimized, an optimized generative adversarial network is generated, and discarded masks in each training in the optimized generative adversarial network are not fixed; performing adversarial training on the pre-training model according to the generative adversarial network to generate a target pre-training model; and inputting bond information to be processed into the target pre-training model to generate a coded text. According to the embodiment of the invention, a fixed dropout mask is not adopted any more, but a probability distribution measurement index or a mean square error and the like are adopted to constrain the output difference of the two different dropout models, so that the dropout can play the role of the dropout, the output of the two different dropout models can be kept consistent as much as possible, the understanding ability of the pre-training model to a text is improved, and the method and the device have the advantages that the method and the device are easy to implement. And the text coding efficiency is improved.

Description

technical field [0001] The present invention relates to the technical field of data processing, and in particular, to a method, device and electronic device for generating encoded text based on adversarial training. Background technique [0002] In the field of deep learning, adversarial training is an effective method to improve model robustness. Adversarial training methods were first used in the field of computer vision. GAN is a relatively successful example in adversarial training. It generates samples that can deceive the discriminant network through the generation network, and the discriminant network determines the authenticity of the samples generated by the generation network, so that the samples generated by the generation network cannot deceive the discriminant network. Under the interaction of the discriminative network and the generative network, the recognition ability of the discriminant network is continuously improved, which is the idea of ​​adversarial tr...

AI Technical Summary

Problems solved by technology

[0007] In view of the above deficiencies in the prior art, the present invention provides a method, device, and electronic device for generating coded text based on confrontational training, aiming to solve the problem of reducing the robustness of the model when the confrontational training method is applied to the pre-trained model in the prior art. sticky question

Images