Improved Techniques for training GANs
T.Salimans,I.Goodfellow,W.Zaremba,V.Cheung,A.Radford,X.Chen, arXiv 2016.
code
This paper focuses to improve the effectiveness of GAN for semi-supervised learning to generate images that humans find visually realistic.To do so feature matching and virtual batch normalization technique is applied. Feature matching prevents GANs from being unstable by specifying a new objective for generator and prevents from overtraining on the current discriminator. Feature matching is effective in situations where regular GANs become unstable. Virtual batch normalization helps to optimize neural networks. For a given output, the input x is highly dependent on several other inputs in the same minibatch , VBN helps to avoid the following problem in NN.
Semi-supervised learning improves the quality of generated images and the Performance of different models are evaluated by using Inception score. Experiments were performed on MNIST, CIFAR-10 and SVHN datasets.
This paper focuses to improve the effectiveness of GAN for semi-supervised learning to generate images that humans find visually realistic.To do so feature matching and virtual batch normalization technique is applied. Feature matching prevents GANs from being unstable by specifying a new objective for generator and prevents from overtraining on the current discriminator. Feature matching is effective in situations where regular GANs become unstable. Virtual batch normalization helps to optimize neural networks. For a given output, the input x is highly dependent on several other inputs in the same minibatch , VBN helps to avoid the following problem in NN.
Semi-supervised learning improves the quality of generated images and the Performance of different models are evaluated by using Inception score. Experiments were performed on MNIST, CIFAR-10 and SVHN datasets.