🛡️ Trustworthy AI

Lab Sessions & Projects for the Trustworthy AI Course
University of Genoa (UniGe)

---

📖 About The Course

This repository contains the laboratory exercises, notebooks, and projects completed for the Trustworthy AI course at the University of Genoa.

The course focuses on the vulnerabilities of machine learning algorithms and the techniques used to evaluate, defend, and explain AI models. We explore various adversarial attacks (like Evasion and Poisoning) and countermeasures (like Adversarial Training) primarily using the SecML library.

---

🧪 Labs Overview

Here is a breakdown of the interactive Jupyter notebooks included in this repository:

Directory/Notebook	Description	Concepts Covered
labs/
00-SecML	Introduction to SecML	Basic operations with CArray, data loading, model training, and performance evaluation.
01-Evasion_complete	Adversarial Evasion Attacks	Implementing PGD (Projected Gradient Descent) attacks, generating adversarial examples to fool classifiers.
02-AdversarialEXEmples	Malware Evasion (Adversarial EXEmples)	Crafting adversarial Windows PE malware files to bypass machine learning-based malware detectors.
03-AdversarialTraining	Adversarial Training & Defenses	Hardening machine learning models against adversarial attacks by incorporating adversarial examples into the training phase.
05-Poisoning_complete	Data Poisoning Attacks	Corrupting the training dataset to compromise the learning process and degrade model performance.
06-Explainability	Explainable AI (XAI)	Interpreting model decisions, understanding feature importance, and visualising how models make predictions.
final_project/
FINAL-PART2	Final Project	Comprehensive application of the concepts learned throughout the course.

---

🚀 Getting Started

To run these notebooks locally, you will need a Python environment with the required dependencies installed.

Prerequisites

We recommend using venv or conda to manage your environment. The primary library used across these labs is SecML.

# Clone the repository
git clone https://github.com/yourusername/trustworthy-ai.git
cd trustworthy-ai

# Create a virtual environment
python3 -m venv .venv
source .venv/bin/activate

# Install dependencies
pip install -r requirements.txt

Running the Labs

Once the environment is set up, you can start Jupyter Lab to explore the notebooks:

jupyter lab

---

📚 Acknowledgments

• Course Instructors and TAs from the University of Genoa.
• The SecML project developers for providing an excellent framework for secure and trustworthy machine learning.

_{Built with 💙 for the Trustworthy AI course.}