GPU-based Self-Organizing Maps for Post-Labeled Few-Shot Unsupervised\n Learning

Few-shot classification is a challenge in machine learning where the goal is\nto train a classifier using a very limited number of labeled examples. This\nscenario is likely to occur frequently in real life, for example when data\nacquisition or labeling is expensive. In this work, we consider the problem of\npost-labeled few-shot unsupervised learning, a classification task where\nrepresentations are learned in an unsupervised fashion, to be later labeled\nusing very few annotated examples. We argue that this problem is very likely to\noccur on the edge, when the embedded device directly acquires the data, and the\nexpert needed to perform labeling cannot be prompted often. To address this\nproblem, we consider an algorithm consisting of the concatenation of transfer\nlearning with clustering using Self-Organizing Maps (SOMs). We introduce a\nTensorFlow-based implementation to speed-up the process in multi-core CPUs and\nGPUs. Finally, we demonstrate the effectiveness of the method using standard\noff-the-shelf few-shot classification benchmarks.\n