A Self-Supervised Deep Learning Framework for Unsupervised Few-Shot Learning and Clustering

The need to learn a good representation is core problem central AI. We present self-supervised learning framework and demonstrate its use for few-shot classification clustering. Our can be interpreted as repeatedly discovering new categories from learned embeddings training embedding function with signals differentiate the discovered categories. In our framework, we first discover unlabeled data. Next post-process previous partition results remove outliers derive prototypes of each category. then construct tasks previously selected data augmented virtual Lastly, iterative train network through steps final representation. considerably outperform baselines in unsupervised on miniImageNet Omniglot sets. also validate clustering that further improves upon recent deep methods.