Vulnerability of Natural Language Classifiers to Evolutionary Generated Adversarial Text

Computing Science honours dissertation project aimed at generating adversarial examples against a sentiment analysis model, using genetic algorithms.

To run the code, run the Jupyter notebooks in the following order

(You will need a machine with a decent NVIDIA GPU to build and test the model)

• 01-Build_model_and_Distance_Matrix.ipynb
• 02-Generate_Attacks.ipynb
• 03-Stats.ipynb