The need for labeled data is among the most common and well-known practical obstacles to deploying deeplearning algorithms to solve real-world problems. The current generation of learning algorithms requires a largevolume of data labeled according to a static and pre-defined schema. Conversely, humans can quickly learngeneralizations based on large quantities of unlabeled data, and turn these generalizations into classificationsusing spontaneous labels, often including labels not seen before. We apply a state-of-the-art unsupervisedlearning algorithm to the noisy and extremely imbalanced xView data set to train a feature extractor thatadapts to several tasks: visual similarity search that performs well on both common and rare classes; identifyingoutliers within a labeled data set; and learning a natural class hierarchy automatically.
展开▼
