Hi Aayush, thanks for reaching out. I'm afraid I don't understand your question, do you mean code for generation of mixture of Gaussians as they do in [1]? If that's the question, I unfortunately don't have anything nice and clean to commit for this toy problem. I did do the experiment back in the day (along with a swiss roll) with good results but didn't publish my findings on this toy dataset anywhere. I recall it was easy enough to get running with a 2 layer network with ReLU activation in both the generator and discriminator, no batch norm, no momentum or other fancy tricks and more or less the same parameters as in the CelebA experiments.
[1] Srivastava, A., Valkov, L., Russell, C., Gutmann, M., and Sutton, C. VEEGAN: Reducing mode collapse in GANs using implicit variational learning. In Advances in Neural Information Processing Systems 30 (NIPS), 2017.
Hi Aayush, thanks for reaching out. I'm afraid I don't understand your question, do you mean code for generation of mixture of Gaussians as they do in [1]? If that's the question, I unfortunately don't have anything nice and clean to commit for this toy problem. I did do the experiment back in the day (along with a swiss roll) with good results but didn't publish my findings on this toy dataset anywhere. I recall it was easy enough to get running with a 2 layer network with ReLU activation in both the generator and discriminator, no batch norm, no momentum or other fancy tricks and more or less the same parameters as in the CelebA experiments.
[1] Srivastava, A., Valkov, L., Russell, C., Gutmann, M., and Sutton, C. VEEGAN: Reducing mode collapse in GANs using implicit variational learning. In Advances in Neural Information Processing Systems 30 (NIPS), 2017.