Semi-supervised learning with Bidirectional GANs
This work addresses semi-supervised learning for image data, but it appears incremental as it builds on existing BiGAN methods with a triplet loss addition.
The paper tackles semi-supervised learning by training Bidirectional GANs with a triplet loss to improve discriminative data representation in latent space, achieving results evaluated on CIFAR10 and SVHN datasets for classification and image retrieval tasks.
In this work we introduce a novel approach to train Bidirectional Generative Adversarial Model (BiGAN) in a semi-supervised manner. The presented method utilizes triplet loss function as an additional component of the objective function used to train discriminative data representation in the latent space of the BiGAN model. This representation can be further used as a seed for generating artificial images, but also as a good feature embedding for classification and image retrieval tasks. We evaluate the quality of the proposed method in the two mentioned challenging tasks using two benchmark datasets: CIFAR10 and SVHN.