A spectral regularization penalty on the discriminator's singular value distribution prevents mode collapse in GANs and improves Inception Score and FID over spectral normalization in the tested settings.
Jamming and percolation properties of random sequential adsorption with relaxation
1 Pith paper cite this work. Polarity classification is still indexing.
abstract
The random sequential adsorption (RSA) model is a classical model in Statistical Physics for adsorption on two-dimensional surfaces. Objects are deposited sequentially at random and adsorb irreversibly on the landing site, provided that they do not overlap any previously adsorbed object. The kinetics of adsorption ceases when no more objects can be adsorbed (jamming state). Here, we investigate the role of post-relaxation on the jamming state and percolation properties of RSA of dimers on a two-dimensional lattice. We consider that, if the deposited dimer partially overlaps with a previously adsorbed one, a sequence of dimer displacements may occur to accommodate the new dimer. The introduction of this simple relaxation dynamics leads to a more dense jamming state than the one obtained with RSA without relaxation. We also consider the anisotropic case, where one dimer orientation is favored over the other, finding a non-monotonic dependence of the jamming coverage on the strength of anisotropy. We find that the density of adsorbed dimers at which percolation occurs is reduced with relaxation, but the value depends on the strength of anisotropy.
citation-role summary
citation-polarity summary
fields
cs.LG 1years
2019 1verdicts
CONDITIONAL 1roles
background 1polarities
unclear 1representative citing papers
citing papers explorer
-
Spectral Regularization for Combating Mode Collapse in GANs
A spectral regularization penalty on the discriminator's singular value distribution prevents mode collapse in GANs and improves Inception Score and FID over spectral normalization in the tested settings.