Same toy benchmark, different budgets: PCA replay 91.1% vs joint 95.2%

BlackHC · x · 2026-09-16

In the author's continual learning experiments, SGD + PCA replay with 4 epochs per task reaches 91.1% ten-class accuracy, while regular joint training with 10 passes over all tasks' data gets 95.2%. Different budgets, one toy benchmark, 3 seeds — a useful reference, not a ceiling.

Related event: BlackHC's toy continual learning experiment: PCA replay hits 91.1%(6 posts)→

Original post →

More from Research

Research channel →