Welcome to the AI-PULSE Project Website: an ANR JCJC project led by Héber H. Arcolezi.

AI-PULSE aims to advance the integration of Differential Privacy (DP) and fairness principles into Machine Learning (ML) models, ensuring the development of efficient, scalable, and responsible AI systems.

AI-PULSE will officially start on Mar 1st, 2025.

AI-PULSE will be recruiting a PhD student and a postdoctoral researcher for Fall 2025. Join us in advancing the intersection of Differential Privacy and fairness in Machine Learning!

Motivation

With the AI-PULSE project, we are driven by the growing demand for responsible AI in a world where the vast potential of data is coupled with critical concerns about privacy, fairness, and transparency. Recent incidents, such as biased AI systems and high-profile data breaches, have highlighted the need for robust solutions that ensure AI technologies are both ethical and reliable. By exploring and advancing privacy-preserving techniques like Differential Privacy and integrating fairness into AI development, we aim to create systems that are privacy-preserving, mitigate bias, and align with evolving regulatory standards.

Goal

AI-PULSE aims to pioneer the integration of Differential Privacy and fairness principles into Machine Learning models, creating responsible AI systems that balance privacy, fairness, and utility. We seek to challenge the current assumption that enhancing privacy compromises fairness (or vice-versa), by proving that these goals can not only coexist but complement each other. Through cutting-edge methodologies, theoretical frameworks, and practical tools, AI-PULSE will offer scalable solutions that support both privacy and fairness, setting new standards for responsible AI development and contributing valuable resources to both the scientific and industrial communities.

Principal Activities

AI-PULSE project contains three major tasks:

• Task 1. Developing and evaluating methodologies that effectively combine Differential Privacy (DP) and fairness mechanisms in Machine Learning. The work will explore different strategies, including pre-processing, in-processing, and post-processing fairness mechanisms applied to differentially private data, and extend this research into decentralized settings like Federated Learning.
• Task 2. Develop robust theoretical foundations to balance privacy, fairness, and utility in ML models under (Local) Differential Privacy. The research will include a comprehensive analysis of (L)DP’s impact on fairness and the development of novel (L)DP mechanisms that satisfy fairness constraints without compromising utility. 
• Task 3. Translate the project’s theoretical and methodological advancements into practical tools. It aims to develop an open-source DP-Fairness toolkit that integrates privacy-preserving and fairness-enhancing mechanisms into Machine Learning models.

ANR JCJC AI-PULSE project is funded by ANR, the French National Research Agency.