Variadic Learning by Bayesian Nonparametric Deep Embedding

Learning at small or large scales of data is addressed by two strong but divided frontiers: few-shot learning and standard supervised learning. Few-shot learning focuses on sample efficiency at small scale, while supervised learning focuses on accuracy at large scale. Ideally they could be reconciled for effective learning at any number of data points (shot) and number of classes (way). To span the full spectrum of shot and way, we frame the variadic learning regime of learning from any number of inputs. We approach variadic learning by meta-learning a novel multi-modal clustering model that connects bayesian nonparametrics and deep metric learning. Our bayesian nonparametric deep embedding (BANDE) method is optimized end-to-end with a single objective, and adaptively adjusts capacity to learn from variable amounts of supervision. We show that multi-modality is critical for learning complex classes such as Omniglot alphabets and carrying out unsupervised clustering. We explore variadic learning by measuring generalization across shot and way between meta-train and meta-test, show the first results for scaling from few-way, few-shot tasks to 1692-way Omniglot classification and 5k-shot CIFAR-10 classification, and find that nonparametric methods generalize better than parametric methods. On the standard few-shot learning benchmarks of Omniglot and mini-ImageNet, BANDE equals or improves on the state-of-the-art for semi-supervised classification.