Multivariate prototype representation for domain-generalized incremental learning

Can Peng, Piotr Koniusz, Kaiyu Guo, Brian C. Lovell, Peyman Moghadam

Abstract: Highlights•We propose an exemplar-free domain-generalized incremental classification method.•Our method is called TRIplet loss with Pseudo old-class feature Sampling (TRIPS).•TRIPS extracts semantic information and maintains old-class knowledge.•An effective feature sampler is based on the multivariate Normal distribution.•A comprehensive task setting for DGCIL is provided.