Exploring Security Vulnerabilities of Deep Learning Models by Adversarial Attacks

Algorithm 1

The visual similar word replacement algorithm.

Let English text data be presented as X=

A neural network model M

A dataset list consist of all the word in this dataset

for each do:

sort by to get keywords list

for each in keywords list and in do:

 calculating

 for each find out by

use replace

is the adversarial example of