We have been awarded a research grant PRIN 2022 by MUR (Ministry of University and Research)

AdVVent: Adversarial Venture, the Mixed Blessing of Adversarial Attacks

Two-fold objective of AdVVent: (left) Building defensive mechanisms for AI system prediction y; (right) The paradoxical aspect of our project: to secure personal data x from a Deepfake or generative AI tools, we use adversarial perturbations from the scenario on the left.

Project description

Image synthesis and generative modeling had unprecedented growth in recent years, unleashing the possibility of realistically manipulating faces in videos or correctly matching a face among millions. At the same time, very little effort has been devoted to cease the dualism between perturbations as evil entities versus benign defense mechanisms. This project challenges this view and advances the hypothesis that adversarial machine learning may improve the state-of-the-art in privacy protection and in the preservation of human dignity. To this end, the project will work in synergy along two opposing axes: on the one hand, it will aim at creating proactive defenses for multimedia data to protect them from malicious AI tools (deepfake or face recognition); another thread of research will study the robustness of AI to various types of perturbations.

Goal

Perform fundamental and applied research with the objective of disseminating research findings and publishing at top-international venues in computer vision, machine learning, deep learning, and related fields.

Keywords

adversarial attacks, robust model, proactive image defense, face analysis, energy-based models, deepfakes, generative AI, diffusion models.

The research will be conducted under the supervision of Prof. Iacopo Masi, University of Rome (Sapienza, University of Rome) and Prof. Giuseppe Lisanti, University of Bologna.