发帖
雷达卡
A Representation Learning Perspective on the Importance of
Train-Validation Splitting in Meta-Learning
Nikunj Saunshi 1 Arushi Gupta 1 Wei Hu 1
Abstract from many “train” tasks to learn a useful prior that can
An effective approach in meta-learning is to uti- help solve new “test” tasks. The goal is to do well on test
lize multiple “train tasks” to learn a good initial- tasks with way fewer samples per task than would ...
京公网安备 11010802022788号
