Mixed batches and symmetric discriminators for GAN training

Explore this paper's citation graph

Summary

This work proposes a generic permutation-invariant discriminator architecture, which is provably a universal approximator of all symmetric functions and reduces mode collapse in GANs on two synthetic datasets, and obtains good results on the CIFAR10 and CelebA datasets.

Type
preprint
Published
2018-06-19
Cited by
39
References
26
Access
Open access

Keywords

Discriminator, Permutation (music), Computer science, Generator (circuit theory), Invariant (physics)

References

Cited by

Related papers