Semi-supervised Learning with GANs: Manifold Invariance with Improved Inference

Abstract

Semi-supervised learning methods using Generative Adversarial Networks (GANs) have shown promising empirical success recently. Most of these methods use a shared discriminator/classifier which discriminates real examples from fake while also predicting the class label. Motivated by the ability of the GANs generator to capture the data manifold well, we propose to estimate the tangent space to the data manifold using GANs and employ it to inject invariances into the classifier. In the process, we propose enhancements over existing methods for learning the inverse mapping (i.e., the encoder) which greatly improves in terms of semantic similarity of the reconstructed sample with the input sample. We observe considerable empirical gains in semi-supervised learning over baselines, particularly in the cases when the number of labeled examples is low. We also provide insights into how fake examples influence the semi-supervised learning procedure.

NIPS 2017 accepted version, including appendix

References in corpus (5)

• Temporal Ensembling for Semi-Supervised Learning
• Improved Training of Wasserstein GANs
• BEGAN: Boundary Equilibrium Generative Adversarial Networks
• Learning in Implicit Generative Models
• Learning Loss Functions for Semi-supervised Learning via Discriminative Adversarial Networks

Cited by in corpus (22)

• Uncertainty-aware multi-view co-training for semi-supervised medical image segmentation and domain adaptation
• Generative Adversarial Networks (GANs) in Networking: A Comprehensive Survey & Evaluation
• Improving the Improved Training of Wasserstein GANs: A Consistency Term and Its Dual Effect
• Augmented Cyclic Adversarial Learning for Low Resource Domain Adaptation
• Learning Privacy-Preserving Student Networks via Discriminative-Generative Distillation
• Robust Conditional Generative Adversarial Networks
• Smooth Neighbors on Teacher Graphs for Semi-supervised Learning
• Featurized Bidirectional GAN: Adversarial Defense via Adversarially Learned Semantic Inference
• Learning from a Complementary-label Source Domain: Theory and Algorithms
• A Bayesian Perspective of Convolutional Neural Networks through a Deconvolutional Generative Model
• Adversarial Partial Multi-Label Learning
• Clarinet: A One-step Approach Towards Budget-friendly Unsupervised Domain Adaptation
• Modeling EEG data distribution with a Wasserstein Generative Adversarial Network to predict RSVP Events
• Adversarial Transformations for Semi-Supervised Learning
• Isometric Gaussian Process Latent Variable Model for Dissimilarity Data
• Subquadratic Overparameterization for Shallow Neural Networks
• Generative-Discriminative Complementary Learning
• Understanding and Improving Virtual Adversarial Training
• Fast Approximate Geodesics for Deep Generative Models
• Learning Condensed and Aligned Features for Unsupervised Domain Adaptation Using Label Propagation
• Learning Identity-Preserving Transformations on Data Manifolds
• End-To-End Graph-based Deep Semi-Supervised Learning