An Adversarial Regularisation for Semi-Supervised Training of Structured Output Neural Networks

TL;DR

This paper introduces an adversarial regularization approach for semi-supervised training of structured-output neural networks, leveraging a discriminator to improve performance with fewer labeled data in image segmentation tasks.

Contribution

It presents a novel adversarial regularization framework inspired by GANs for semi-supervised learning of structured-output neural networks.

Findings

Achieves comparable performance to fully supervised models with half the annotations.

Demonstrates effectiveness in image segmentation tasks.

Boosts network performance using discriminator-based error signals.

Abstract

We propose a method for semi-supervised training of structured-output neural networks. Inspired by the framework of Generative Adversarial Networks (GAN), we train a discriminator network to capture the notion of a quality of network output. To this end, we leverage the qualitative difference between outputs obtained on the labelled training data and unannotated data. We then use the discriminator as a source of error signal for unlabelled data. This effectively boosts the performance of a network on a held out test set. Initial experiments in image segmentation demonstrate that the proposed framework enables achieving the same network performance as in a fully supervised scenario, while using two times less annotations.