Semi-Supervised Learning via New Deep Network Inversion

We exploit a recently derived inversion scheme for arbitrary deep neural networks to develop a new semi-supervised learning framework that applies to a wide range of systems and problems. The approach outperforms current state-of-the-art methods on MNIST reaching 99.14% of test set accuracy while using 5 labeled examples per class. Experiments with one-dimensional signals highlight the generality of the method. Importantly, our approach is simple, efficient, and requires no change in the deep network architecture. 1 ar X iv :1 71 1. 04 31 3v 1 [ st at .M L ] 1 2 N ov 2 01 7