Visual Learning in an Imperfect World
Martedì 9 Febbraio 2021 - ore 14.30
Elisa Ricci
Dipartimento di Ingegneria e Scienza dell'Informazione
Università di Trento
Over the last few years deep learning has enabled unprecedented successes in computer vision. Despite their effectiveness, deep networks rely on large scale, carefully annotated training data.Therefore, when considering complex problems in real world applications, their performances drop since both data and labels can be noisy or difficult and costly to obtain. In this seminar, I will discuss recent approaches and trends in computer vision for addressing the challenge of visual learning in realistic scenarios. I will also focus on recent activities from my research group on this topic, discussing methods forunsupervised and self-supervised learning, domain adaptation and continual learning.