Unsupervised Embedding Adaptation via Early-Stage Feature Reconstruction
for Few-Shot Classifcation
Dong Hoon Lee 1 Sae-Young Chung 1
Abstract setting. The co-existence of labeled- and unlabeled-data in
We propose unsupervised embedding adaptation this setting motivates the use of transductive inference or
for the downstream few-shot classifcation task. semi-supervised learning. A popular transductive approach
...


雷达卡




京公网安备 11010802022788号







