Abstract:Recent advances in semi-supervised learning methods rely on estimating the categories of unlabeled data using a model trained on the labeled data (pseudo-labeling) and using the unlabeled data for various consistency-based regularization. In this work, we propose to explicitly leverage the structure of the data manifold based on a Manifold Graph constructed over the image instances within the feature space. Specifically, we propose an architecture based on graph networks that jointly optimizes feature extraction, graph connectivity, and feature propagation and aggregation to unlabeled data in an end-to-end manner. Further, we present a novel Prototype Generator for producing a diverse set of prototypes that compactly represent each category, which supports feature propagation. To evaluate our method, we first contribute a strong baseline that combines two consistency-based regularizers that already achieves state-of-the-art results especially with fewer labels. We then show that when combined with these regularizers, the proposed method facilitates the propagation of information from generated prototypes to image data to further improve results. We provide extensive qualitative and quantitative experimental results on semi-supervised benchmarks demonstrating the improvements arising from our design and show that our method achieves state-of-the-art performance when compared with existing methods using a single model and comparable with ensemble methods. Specifically, we achieve error rates of 3.35% on SVHN, 8.27% on CIFAR-10, and 33.83% on CIFAR-100. With much fewer labels, we surpass the state of the arts by significant margins of 41% relative error decrease on average.

Manifold regularization with GANs for semi-supervised learning

Semi-Supervised Learning with GANs: Revisiting Manifold Regularization

GENERATING MANIFOLD-ALIGNED SEMANTIC FEATURE FOR ZERO-SHOT LEARNING

MR-GAN: Manifold Regularized Generative Adversarial Networks

Manifold Constraint Regularization for Remote Sensing Image Generation

Dual Distribution Matching GAN

Improving FM-GAN through Mixup Manifold Regularization

Manifold adversarial training for supervised and semi-supervised learning

A Regularized Approach for Geodesic-Based Semisupervised Multimanifold Learning

MMCGAN: Generative Adversarial Network with Explicit Manifold Prior

Manifold-preserved GANs

Regularizing Discriminative Capability Of Cgans For Semi-Supervised Generative Learning

Optimizing Generative Adversarial Networks for Image Super Resolution via Latent Space Regularization

Enhanced Manifold Regularization for Semi-Supervised Classification.

Semi-supervised Logistic Regression Via Manifold Regularization

SMaRt: Improving GANs with Score Matching Regularity

Consistency Regularization with Generative Adversarial Networks for Semi-Supervised Learning

Manifold Fitting with CycleGAN.

Semi-supervised Manifold Regularization with Adaptive Graph Construction

Manifold Graph with Learned Prototypes for Semi-Supervised Image Classification

Branched Generative Adversarial Networks for Multi-Scale Image Manifold Learning.