Bert attack

​

Goal

BERT Attack is an evasion attack that replaces words in a text with the aim of causing the target model to misclassify the sentiment of the text, while still maintaining the same semantic meaning to a human reader.

​

Example

​

Original

fire victims waive deadline for governor's OK of reorganization plan

​

Evasion example:

loss victims waive contract for governor's OK of reorganization plan

​

Remediation