Adversarial sample generation method, medium, device and computing device

Abstract

The present invention provides a method for generating an adversarial example, comprising: obtaining an original sample from a data set; determining a replaceable word / word in the original sample; based on the context of the replaceable word / word in the original sample Determine candidate replacement words / characters; use the candidate replacement words / characters to replace corresponding replaceable words / characters in the original sample to generate adversarial samples. This method can generate an adversarial example that replaces a specific word in the original sample based on the text data in the data set. After the adversarial example is input into the target model, it can reduce the influence of the target model from the interference of "difficult common words" and reduce the impact on other Partially judged bias in order to improve the performance (robustness) of the target model. In addition, there is lower bag-of-words similarity between two sentences in the adversarial examples generated according to the present invention. In addition, the present invention also provides an adversarial sample generation device, medium and computing equipment.

Description

technical field [0001] Embodiments of the present invention relate to the field of natural language processing, and more specifically, embodiments of the present invention relate to a method, medium, device, and computing device for generating an adversarial example. Background technique [0002] Similarity calculation is used to measure the degree of similarity between objects. In the field of natural language processing technology, similarity calculation is a basic operation, which is widely used in technical scenarios such as data mining, data classification, information retrieval and information collection. Text similarity calculation is a type of similarity calculation often involved in the field of natural language processing technology. By calculating the similarity between different texts, it is possible to perform cluster analysis, text matching or deduplication processing on large-scale text corpora. [0003] In recent years, machine learning has achieved breakthr...

AI Technical Summary

Problems solved by technology

[0004] The adversarial examples generated by the existing technology are basically similar to the semantics of the original samples, and it is difficult to measure the generalization ability of the target model

Images