id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
work_o5yf5ggv6nefzovxdmf5l4n2iu	Maxim Borisyak	Adaptive divergence for rapid adversarial optimization	2020	24	.pdf	application/pdf	10150	1313	58	Thechoice ofthe modelhas itstradeoff: high-capacity models provide good estimations of the divergence, but, generally, proposed divergence family suggests using low-capacity models to compare distant Adversarial Optimization (AO), introduced in Generative Adversarial Networks (Goodfellow et al., 2014), became popular in many areas of machine learning and beyond with The proposed divergence family suggests using low-capacity models to compare models generally require fewer samples for training, AD allows an optimization algorithm same models as JSD with linear and logarithmic capacity functions, dashed lines represent some pseudodivergences from the families producing adaptive divergences. Require: XP, XQ — samples from distributions P and Q, B — base estimator training algorithm, N — maximal size of the ensemble, c :Z+→[0,1]— capacity function; ρ — Algorithm 3 Adaptive divergence estimation by a dropout-regularized neural network Algorithm 4 Adaptive divergence estimation by a regularized neural network regularization methods can be used to regulate model capacity in adaptive divergences.	./cache/work_o5yf5ggv6nefzovxdmf5l4n2iu.pdf	./txt/work_o5yf5ggv6nefzovxdmf5l4n2iu.txt