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
   ...