The generation of synthetic data has gained a lot of relevance, particularly to provide more data for learning models, for example with GANs (Generative Adversarial Networks), which are effective and have a relatively simple implementation method, making them one of the most widely used methodologies for generating synthetic data. However, (semi-)synthetic data is also relevant to an even more important task, which is to improve our understanding of the behaviour of ML (Machine Learning) models and algorithms.
Carlos Soares, Professor at DEI and researcher in this area, invited by the Università degli Studi di Bari Aldo Moro (Italy) to integrate a doctoral jury, also gave in this institution a seminar on March 26th entitled “Synthetic data for a better understanding of models and algorithms“, addressing limitations of current research practices in Machine Learning /AI, describing some of the work underway at FEUP with the aim of improving these practices, within the scope of projects such as the Center for Responsible AI and AISym4Med.