Experience

Institute of Information Science and Technologies

• Italy
• Research Services
• 100 - 200 Employee

• Research fellow

  • Apr 2021 - Present

  The activity focuses on the study and design of Machine Learning models that can effectively address image-based recognition tasks in critical scenarios, such as those involved in clinical decision support. Few-Shot Learning models through the Meta-Learning paradigm are among the main topics of the activity as they can effectively learn from a few examples, thus overcoming the limitations given by the available datasets. So far, I have studied the operation of Convolutional Neural Networks (CNNs), the state-of-the-art in Computer Vision. Several CNNs have been built from scratch and trained on 2D images. Due to the challenging clinical question and the small and unbalanced datasets available, I examined the attention mechanism to investigate whether it provided any performance advantage. I first explored how to apply it to CNNs by exploiting Attention Gates, obtaining an actual improvement in performance. Next, I moved on to study the Self-Attention mechanism on which Vision Transformers are based, implementing a Vision Transformer working with 3D volumes. The application scenario is the prediction of prostate cancer aggressiveness in MRI scans. Show less



University of Edinburgh School of Engineering

• United Kingdom
• Higher Education
• 1 - 100 Employee

• Visiting Student

  • May 2023 - Jun 2023



Institute of Information Science and Technologies

• Italy
• Research Services
• 100 - 200 Employee

• Thesis Internship

  • Jul 2020 - Feb 2021

  This internship experience allowed me to acquire the knowledge and skills necessary to implement Machine Learning algorithms. The study of the state-of-the-art, the functioning of the Pytorch library and the comparison with my tutors allowed me to learn how to build Machine Learning, in particular Deep Learning models, including single and multimodal Convolutional Neural Networks. I also acquired knowledge in Explainable Artificial Intelligence, investigating in the literature the types and functioning of algorithms for interpreting Machine Learning models and then focusing on the use of the LIME algorithm to explain the decision criteria of the implemented neural networks. The collaboration with the University Hospital of Careggi, Florence, allowed me to test my models on MRI images of prostate cancer and to put into practice the knowledge and skills learned in theory. Finally, the study and comparison with different professionals allowed me to learn how to analyse the results and assess the correct functioning of the models built. Show less