Data augmentation, transfer learning, federated learning,exlainable AI, as well as innovative model-based and data driven methodologies for closed loop systems

Available data for the application of AI methodologies are never enough. In this respect, the University of Rome offers well established expertise related to data pre-processing, augmentation and analytics, aiming at identifying the most effective methods based on convolutional neural architectures and on the transfer learning paradigm. In relation to data augmentation and data enhancing, we can design systems for the removal of optical artefacts/defects through Deep Image Inpainting, Diffusion Models and Generative Adversarial Networks. In addition, we can design and implement convolutional neural architectures, operating on multimedia data. We can guarantee compliance with the most stringent privacy regulations (GDPR) by the use of the federated learning paradigm, while we can improve the performance of the machine learning models through the use of the continuous learning paradigm. Furthermore, we can develop Explainable AI (XAI) solutions in relation to the type of data considered, to provide clinical partners with information on the decision-making processes of the learning architectures.

Last, but not least, we can close the feedback, by using advanced control and optimization methodoogies, both model-based and data-driven (e.g. neural network based model predicitive control, deep reinforcement learning,...), we can develop Decision Support Systems or real-time controllers which, on the grounds of suitably pre-analyzed data (derived from the monitoring of the process) suggest the most appropriate control actions to be actuated on the process.

We have applied the above-mentioned methodologies to a plentu o f applications in several fields (cellular networks, satellite networks, energy networks, health systems, critical infrastructures...) even in teh framework of many EU financed projects.