Uniform Herding Reduces Forgetting in Class-Incremental Learning
August 14, 2026
Uniform Herding mitigates forgetting in incremental learning by allocating active exemplar sets across observed classes and refreshing them using a bounded candidate pool. On CIFAR-100 with a ResNet-18 backbone, it achieved 44.00% accuracy and 17.22% forgetting, outperforming iCaRL's 42.33% accuracy and 24.87% forgetting.
HOW THIS AFFECTS YOU
●
researcherYou can improve class-incremental learning performance by refreshing exemplars in current representations rather than relying on static sets.