Using per-dimension log-likelihood contributions to decide which dimensions to factor out early improves bits/dim for RealNVP on CIFAR-10, ImageNet, and CelebA.
Datasets: We perform experiments on four benchmarked image datasets: CIFAR-10 (Krizhevsky, 2009), Imagenet (Russakovsky et al.,
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2019 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Likelihood Contribution based Multi-scale Architecture for Generative Flows
Using per-dimension log-likelihood contributions to decide which dimensions to factor out early improves bits/dim for RealNVP on CIFAR-10, ImageNet, and CelebA.