Per-task multiplicative masks on frozen transformer attention layers give strong class-incremental accuracy, but the claimed convergence proof and forgetting-free guarantee are not supported.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Soft-TransFormers for Continual Learning
Per-task multiplicative masks on frozen transformer attention layers give strong class-incremental accuracy, but the claimed convergence proof and forgetting-free guarantee are not supported.