Hands-On One-shot Learning with Python

Semi-supervised learning

Suppose we have 10,000 data points where only 20,000 are labeled and 80,000 are unlabeled. In such cases, we would employ semi-supervised learning. In semi-supervised learning, we use unlabeled data to gain more of an understanding of the population structure in general. Semi-supervised learning goes through a pseudo-labeling technique to increase the training set; that is, we train a model using 20,000 labeled datasets and use it on equally sized test data points to create pseudo-labels for them. The following diagram illustrates a semi-supervised learning architecture:

After obtaining pseudo-labels, we concatenate ...