Tamara HAJJAR

This research project aims to study the impact of privacy attacks, notably model inversion attacks, on the leakage of information when applied to diffusion models

This project introduces an innovative approach to DNN watermarking designed to protect the intellectual property rights of legitimate model owners, while maintaining the model's performance in its primary task. By building upon two previously established watermarking methods, we developed a hybrid technique that leverages the strengths of both approaches. This results in a more resilient and secure watermarking solution, capable of withstanding various attacks and watermark removal attempts, thus ensuring robust protection against unauthorized exploitation.

Internship within the association "Together for Chehim", whose activity is to ensure the development of social, cultural and educational sectors of the city of Chehim. This work was carried out under the supervision of the comittee responsible for applied research projects in the association, in collaboration with the Laboratory for Research in Engineering Systems (LARIS) of the University Of Angers, France. In this project, medical imagery and machine learning techniques are used in order to obtain a robust classifier, capable of distinguishing between several pulmonary diseases, and attributing each image to the right class.