A generative-model-based test for equality of conditional distributions that uses cross-generation, an RKHS-indexed supremum statistic, and multiplier bootstrap, with claimed double robustness to generator errors.
Training generative neural networks via Maximum Mean Discrepancy optimization
9 Pith papers cite this work, alongside 222 external citations. Polarity classification is still indexing.
9
Pith papers citing it
222
external citations · Pith
abstract
We consider training a deep neural network to generate samples from an unknown distribution given i.i.d. data. We frame learning as an optimization minimizing a two-sample test statistic---informally speaking, a good generator network produces samples that cause a two-sample test to fail to reject the null hypothesis. As our two-sample test statistic, we use an unbiased estimate of the maximum mean discrepancy, which is the centerpiece of the nonparametric kernel two-sample test proposed by Gretton et al. (2012). We compare to the adversarial nets framework introduced by Goodfellow et al. (2014), in which learning is a two-player game between a generator network and an adversarial discriminator network, both trained to outwit the other. From this perspective, the MMD statistic plays the role of the discriminator. In addition to empirical comparisons, we prove bounds on the generalization error incurred by optimizing the empirical MMD.
citation-role summary
background 1
citation-polarity summary
roles
background 1polarities
background 1representative citing papers
A new evaluation framework using MMD on Biber features shows LLMs deviate from human linguistic distributions across registers, with closest models varying by register rather than size.
Derives MSIP algorithm from MMD gradient flows for weighted quantization, extending mean shift and relating to preconditioned gradient descent and Lloyd's clustering.
MMD GANs have unbiased critic gradients but biased generator gradients from sample-based learning, and the Kernel Inception Distance provides a practical new measure for GAN convergence and dynamic learning rate adaptation.
Mixture-of-experts flow matching enables non-autoregressive language models to achieve autoregressive-level quality in three sampling steps, delivering up to 1000x faster inference than diffusion models.
A stochastic MPC controller for HCCI engines using learned uncertainty distributions, polynomial chaos expansion, and an MMD-based cost reduces combustion phasing variation by over 28% and improves load tracking by over 26% in simulations compared to standard methods.
N-body simulations show the Galactic bar's pattern speed strongly shapes open cluster tidal tails, enabling a potential independent measurement of that speed from a small set of nearby clusters.
DriftXpress approximates the attraction field of drifting models with a Nyström landmark projection, reducing training time by 2.6–6.7× at comparable FID.
Introduces three knockoff filters for FDR-controlled variable screening in regularized DNNs and reports satisfactory empirical performance versus existing algorithms.
citing papers explorer
-
Weighted quantization using MMD: From mean field to mean shift via gradient flows
Derives MSIP algorithm from MMD gradient flows for weighted quantization, extending mean shift and relating to preconditioned gradient descent and Lloyd's clustering.
-
Demystifying MMD GANs
MMD GANs have unbiased critic gradients but biased generator gradients from sample-based learning, and the Kernel Inception Distance provides a practical new measure for GAN convergence and dynamic learning rate adaptation.
-
Knockoffs-based False Discovery Rate Control and Simplification for Deep Neural Networks
Introduces three knockoff filters for FDR-controlled variable screening in regularized DNNs and reports satisfactory empirical performance versus existing algorithms.