… incremental learning. On the one hand, we propose to memorize one class-representative
prototype for each old class and adopt prototype … -supervised learning (SSL) to learn more gen…

… Secondly, prototypes of various classes are linearly combined to construct the hybrid
prototype with mixed labels. Along with prototype augmentation, they are used for incremental …

… learning to recognize objects via forming prototypes [4], we propose a novel CGL method
called Class Prototype Construction and Augmentation (… perform the incremental learning step …

… cross-entropy loss, distillation loss and prototypes augmentation loss. Let λ, γ be the
distillation loss trade-off and prototypes augmentation loss trade-off respectively. Following the …

… incremental learning into the bi-level optimization, where the high-dimensional embedding
and projection module are proposed to facilitate prototype … the data augmentation strategy (w/ …

… incremental prototype learning scheme to explicitly learn an extensible feature representation,
and thus facilitate subsequent incremental … Augmentation-based methods [14] synthesize …

… While counteracting forgetting, incremental learners also need to adapt to new tasks. We
design an augmented asymmetric knowledge aggregation scheme that endows the model with …

… In this paper, an effective class incremental learning (CIL) … -incremental with self-training
and prototype augmentation (… prototypes of the old classes are constructed and augmented. …

… Then, combined with the remaining h basis vectors randomly selected to generate the
augmentation features in the sub-prototype space. Since the imbalanced distribution within the …

… tasks, and employ self-supervised learning to learn more transferable features for future tasks…
augmented prototypes, we mainly consider the adjustment for joint representation learning …