Continual Self-Supervised Learning

Giacomo Cignoni

Research Fellow at the University of Pisa

Learning continually from non-stationary data streams is a challenging research topic of growing popularity in the last few years. Being able to learn, adapt and generalize continually, in an efficient way appears to be fundamental for a more sustainable development of Artificial Intelligent systems. However, research in Continual Learning has been largely focused on supervised learning, despite the trend in Deep Learning shifting towards Self-Supervised approaches in recent years. Existing Continual strategies for Self-Supervised methods suffer from limitations, such as requiring scenarios with known task boundaries or being limited to Instance Discrimination Self-Supervised methods. The scope of this project is to develop novel strategies for Contnual Self-Supervised Learning that do not suffer from these limitations.

Keywords: Continual Learning, Self-Supervised Learning, Lifelong Learning, Representation Learning, Deep Learning

Scientific area: Artificial Intelligence

Bio: I am a research fellow at the Department of Computer Science of the University of Pisa, where I work on developing novel strategies to counter forgetting for Self-Supervised machine learning methods when trained in a Continual scenario, together with Prof. Antonio Carta and Andrea Cossu. My main research interests are Self-Supervised Learning, Continual Learning Computer Vision and Deep Learning in general. In the past, I have worked as a research fellow for Prof. Alina Sîrbu on analysis of biomedical images, specifically classification of weakly labeled Whole Slide Images of tumors. I obtained my Master’s Degree in Computer Science, with an Artificial Intelligence specialization, from the University of Pisa in February 2024. I obtained my Bachelor’s Degree in Computer Science from University of Pisa in December 2021. My Master’s thesis delved into developing a family of strategies and a framework for Continual Self-Supervised Learning in an Online setting.

Visiting period: May – August 2024 at LAMP laboratory, CVC Barcelona