RECAST decomposes pretrained layer weights into shared templates and per-task scalar coefficients, achieving task-incremental learning with fewer than 50 trainable parameters per task and matching or improving accuracy on six image datasets.
Online learned continual compression with adaptive quantization modules
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CV 1years
2024 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
RECAST: Reparameterized, Compact weight Adaptation for Sequential Tasks
RECAST decomposes pretrained layer weights into shared templates and per-task scalar coefficients, achieving task-incremental learning with fewer than 50 trainable parameters per task and matching or improving accuracy on six image datasets.