The content of the cyberwal 2022 workshop day on machine learning security in the real world is available below.
Theoretical part: Machine Learning Security in the Real World
The slides of Maxime Cordy’s lecture are available here.
Lab work: Machine Learning Security in the Real World
Tutorial given at 2022 Cyberwal school. During the lab, the students gain practical knowledge on adversarial attacks via an online game and a hands-on exercise.
1. Notebooks
First, we start with two notebooks. Open them in Google Colab, click on the “Copy on Drive” button, and then modify your own copy to be able to save your changes.
- Introduction to adversarial examples for image classification with PyTorch. Notebook
- Attacking NLP machine learning models with the textattack library (text adversarial examples). Notebook
2. Adversarial Escape Game
Once the notebooks are done, we can play with adversarial examples in the online adversarial escape game. Play the game!
Credits:
- Notebook 1: Thibault SIMONETTO & Martin GUBRI, based on the Adversarial Example Generation notebook
- Notebook 2: Salijona DYRMISHI
- Adversarial Escape Game: Salah GHAMIZI