Dr Joost van de Weijer pracuje jako Senior Scientist w Computer Vision Center w Barcelonie. Kieruje grupą badawczą Learning and Machine Perception (LAMP). W swoim wystąpieniu omówi przejście od uczenia ciągłego nadzorowanego do metod nienadzorowanych (i samo-nadzorowanych) w uczeniu ciągłym oraz systemy wieloagentowe w kontekście uczenia ciągłego.
W wydarzeniu można uczestniczyć online.
Meeting ID: 997 9623 8035
Login: Imię i nazwisko
Tytuł: „Towards Label-Efficient and Multi-Agent Continual Learning”
Abstract: Continual learning aims to accumulate knowledge from a stream of incoming data. The first part of this talk will focus on how we can move from supervised continual learning towards unsupervised (and self-supervised) methods for continual learning. The majority of continual learning literature focuses on supervised continual learning, where a learner adapts to a stream of fully labeled data while consolidating previously learned knowledge. Results for domain incremental and class incremental learning will be discussed. In the second part of the talk, the focus will be on multi-agent systems for continual learning. Instead of a single agent which learns on a stream of incoming data, systems of two agents will be considered. This allows agents to specialize in certain tasks without requiring them to consolidate knowledge from previous tasks. This technique is found to be beneficial in supervised, self-supervised, and other applications like incremental semantic segmentation. The results confirm that multi-agent continual learning allows for a good trade-off between plasticity and stability.
Bio: Joost van de Weijer is a Senior Scientist at the Computer Vision Center and leader of the Learning and Machine Perception (LAMP) group. He received his Ph.D. degree in 2005 from the University of Amsterdam. From 2005 to 2007, he was a Marie Curie Intra-European Fellow in the LEAR Team, INRIA Rhone-Alpes, France. From 2008 to 2012, he was a Ramon y Cajal Fellow at the Universidad Autonoma de Barcelona. He has served as an area chair for the main computer vision and machine learning conferences CVPR; ICCV; ECCV, ICML, NeurIPS. His main research interests include active learning, continual learning, transfer learning, domain adaptation, and generative models.