MetaCL augments parameter-regularization continual learners with a Reptile-style inner loop and an adaptive gradient balancing rule, improving accuracy on low-shot permuted MNIST, CIFAR-100, and CUB.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Learning Continually from Low-shot Data Stream
MetaCL augments parameter-regularization continual learners with a Reptile-style inner loop and an adaptive gradient balancing rule, improving accuracy on low-shot permuted MNIST, CIFAR-100, and CUB.