Semi-Supervised Learning via New Deep Network Inversion

Abstract

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.

arXiv admin note: substantial text overlap with arXiv:1710.09302

References in corpus (7)

• TensorFlow: Large-Scale Machine Learning on Heterogeneous Distributed Systems
• Improved Techniques for Training GANs
• Energy-based Generative Adversarial Network
• Unsupervised and Semi-supervised Learning with Categorical Generative Adversarial Networks
• Auxiliary Deep Generative Models
• A Probabilistic Theory of Deep Learning
• Semi-Supervised Learning with the Deep Rendering Mixture Model

Cited by in corpus (2)

• Workshop Report: Detection and Classification in Marine Bioacoustics with Deep Learning
• Adversarial Code Learning for Image Generation