New paper explains imbalanced forgetting in class-incremental learning

Researchers have identified a phenomenon called imbalanced forgetting in class-incremental learning, where some classes are forgotten more than others despite balanced rehearsal strategies. A new paper proposes three last-layer coefficients derived from gradient analysis to predict and explain this imbalanced forgetting. One coefficient, representing self-induced interference, appears to be the strongest predictor and is influenced by new-class interference. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Provides a mechanistic account for imbalanced forgetting in CIL, suggesting new directions for model improvement.

RANK_REASON The cluster contains an academic paper detailing a new analysis of a machine learning phenomenon. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CV →

• class-incremental learning
• rehearsal
• imbalanced forgetting
• last-layer coefficients
• gradient analysis
• self-induced interference
• new-class interference
• Alberto Tamajo

• paper
• other