With that in mind, semi-supervised learning is a technique in which both labeled and unlabeled data is used to train a classifier.

This type of classifier takes a tiny portion of labeled data and a much larger amount of unlabeled data (from the same domain). The goal is to combine these sources of data to train a Deep Convolution Neural Network (DCNN) to learn an inferred function capable of mapping a new datapoint to its desirable outcome.

In this frontier, we present a GAN model to classify street view house numbers using a very small labeled training set. In fact, the model uses roughly 1.3% of the original SVHN training labels i.e. 1000 (one thousand) labeled examples. We use some of the techniques described in the paper Improved Techniques for Training GANs from OpenAI (https://arxiv.org/abs/1606.03498).