AdvSecureNet: A Python Toolkit for Adversarial Machine Learning

Introduction

The AdvSecureNet Master Project aims to enhance the existing AdvSecureNet  toolkit. This toolkit is designed for adversarial machine learning and includes support for multi-GPU setups and various attacks, defenses, and evaluation metrics. The project's expansion focuses on integrating capabilities for Natural Language Processing (NLP) Attacks and Defenses, Audio Recognition Attacks and Defenses, Large Language Models (LLMs) Vulnerabilities, and Fairness and Bias Evaluation Metrics. 

This project is a collaborative effort between the AIML  and S.E.A.L. research groups. Prof. Dr. Manuel Günther from the AIML research group serves as the responsible professor, while Melih Catal from the S.E.A.L. research group acts as the main supervisor, managing the project’s progress and execution.

Project Requirements

Technical Skills:

• Software Engineering: Experience in software development, unit testing, documentation, version control, and CI/CD pipelines.
• Machine Learning: Good understanding of machine learning concepts with hands-on experience in Python and PyTorch.
• Interest in Responsible and Trustworthy AI: Familiarity with adversarial attacks, defenses, and the principles of Trustworthy AI, or the willingness to learn.

Responsibilities:

• Implementation of new features while ensuring high code quality and maintainability.
• Support for both CLI and API interfaces, and external YAML configuration files.
• Engagement in research activities, including reading relevant research papers.
• Active participation in regular project meetings with the main supervisor and collaboration with team members.

For more detailed information on the project scope and specific tasks, please refer to the project description.

Start: ASAP

Group Size: 2-5

Contact: Melih Catal catal@ifi.uzh.ch