
Gradient Correlation Subspace Learning (GCSL) reduces catastrophic forgetting in incremental class learning by projecting weights into a subspace least affected by previous tasks.


coremsg

gradient-correlation-subspace-learning-to-reduce-catastrophic-forgetting


Gradient Correlation Subspace Learning to Reduce Catastrophic Forgetting


title_rewrite


The authors propose an effective approach, called EDBL, to tackle the decision boundary overfitting problem in class incremental learning (CIL) by addressing the issues of insufficient knowledge distillation and imbalanced data learning.


effective-decision-boundary-learning-for-incremental-class-learning


Effective Decision Boundary Learning for Incremental Class Learning
