DOTTORATO NAZIONALE IN INTELLIGENZA ARTIFICIALE

The topic of PhD research lies at the intersection of machine learning, computer vision and computer security broadly in the area of adversarial machine learning. The focus of work is: (i) to understand the impact of spurious correlations in the training data when building machine-learning models, and (ii) to identify spurious features and correlations after training, using explainability methods to finally improve the quality of data fed into the model. This line of research will help identify the quality data and give better understanding of the what-goes-behind the “black-box” nature of the learning process of most state-of-the-art computer vision applications and therefore making them secure and robust against plausible attacks.