Binary Clustering for Deep Network Trained by Feature Growth

Clustering, a class of unsupervised learning methods, has been extensively studied and applied in machine learning. By designing the training process, we are able to impose CNN with ability clustering. In this paper, propose binary clustering framework by implementing deep network instead barely acting as feature extractor. The proposed strategy consists five stages: elimination, seeding, germination, growing grafting. Feature elimination works eliminate impact random weights initialization leave unbiased. germination altogether realize similarity generation, while grafting filters strengthens generated similarity. Because uncertainty clustering, take genres into consideration evaluation metrics. Only performance analysis is considered first second samples selected during belong different genres. Compared DEC DEC-DA, our shown achieve state-of-the-art performance, accuracy up 0.996 for dataset fashion MNIST 4/5 -full outperforms other methods terms stability.