Improving GAN Training via Binarized Representation Entropy (BRE) Regularization
This addresses training instability in GANs for generative modeling, though it is incremental as it builds on existing regularization methods.
The paper tackles the problem of unstable GAN training by proposing a regularizer that encourages high joint entropy in binary activation patterns of the discriminator, resulting in improved stability, convergence speed, and sample quality, with higher classification accuracies in semi-supervised learning.
We propose a novel regularizer to improve the training of Generative Adversarial Networks (GANs). The motivation is that when the discriminator D spreads out its model capacity in the right way, the learning signals given to the generator G are more informative and diverse. These in turn help G to explore better and discover the real data manifold while avoiding large unstable jumps due to the erroneous extrapolation made by D. Our regularizer guides the rectifier discriminator D to better allocate its model capacity, by encouraging the binary activation patterns on selected internal layers of D to have a high joint entropy. Experimental results on both synthetic data and real datasets demonstrate improvements in stability and convergence speed of the GAN training, as well as higher sample quality. The approach also leads to higher classification accuracies in semi-supervised learning.