McGan: Mean and Covariance Feature Matching GAN

By Youssef Mroueh, Tom Sercu, Vaibhava Goel. 2017

166 citations recorded by the source corpus.

We introduce new families of Integral Probability Metrics (IPM) for training Generative Adversarial Networks (GAN). Our IPMs are based on matching statistics of distributions embedded in a finite dimensional feature space. Mean and covariance feature matching IPMs allow for stable training of GANs, which we will call McGan. McGan minimizes a meaningful loss between distributions.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Generative Models

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Computer Vision and Pattern Recognition

In Generative Models

Citations in the corpus

Coulomb GANs: Provably Optimal Nash Equilibria via Potential Fields (2017, background): “McGANs match mean and covariance of synthetic and target data, therefore are more suited than standard GANs to approximate the target distribution [34].”
Demystifying MMD GANs (2018, background): “Many other GAN variants fall into the framework of IPMs (e.g. Mroueh et al., 2017; Mroueh & Sercu, 2017; Berthelot et al., 2017).”
Geometric GAN (2017, methodology): “In fact, Wasserstein GAN is a special instance of minimizing the integral probability metric (IPM) [8], and Mroueh et al [9] recently generalized the W-GAN for wider function classes and proposed the mean feature matching and/or covariance feature matching GAN (McGAN) using the IPM minimization…”
Moment Matching for Multi-Source Domain Adaptation (2018, methodology): “McGAN [29] utilizes a GAN to match the mean and covariance of feature distributions.”
Boundary-Seeking Generative Adversarial Networks (2017, background): “, 2017) or by constraining its output distribution (Mroueh & Sercu, 2017; Mroueh et al., 2017; Sutherland et al., 2016).”
MMD GAN: Towards Deeper Understanding of Moment Matching Network (2017, background): “[23] also propose McGAN that matches second order moment from the primal-dual norm perspective.”
The Cramer Distance as a Solution to Biased Wasserstein Gradients (2017, neutral): “Recently, Mroueh et al. (2017) reported good results with mean and covariance feature matching GAN (McGan).”
Symmetric Variational Autoencoder and Connections to Adversarial Learning (2017, background): “Different classes of functions fψ(x) yield different learning metrics, including energy-based GAN [28], Maximum Mean Discrepancy [10], Fisher GAN [16], Wasserstein GAN (WGAN) [2], McGAN [17], among many other possible forms.”
Handling Incomplete Heterogeneous Data using VAEs (2018, unknown): “L G et al. (2018), DVAE++ (Vahdat et al., 2018), Shape Variational Autoencoder (Nash and Williams, 2017) and GAN-style objective functions (f-GAN (Nowozin et al., 2016), DR-GAN (Tran et al., 2019), Wasserstein GANs (Gulrajani et al., 2017), MMD-GAN (Li et al., 2015), Gated-GAN (Chen et al., 2019), A”
Feature Quantization Improves GAN Training (2020, methodology): “…are two issues: (i) Principled methods often require to constrain the discriminator capacity (e.g., weight clipping or gradient penalty) to ensure the boundedness: new architectural adjustments such as a higher number of feature maps are needed to compensate for constraints (Mroueh et al., 2017).”
Parametric Adversarial Divergences are Good Task Losses for Generative Modeling (2017, background): “…the discriminator to be a neural network instead of any function are the MMD-GAN papers: Li et al. (2015); Dziugaite et al. (2015); Li et al. (2017); Mroueh et al. (2017) and Bellemare et al. (2017) who give an interpretation of their energy distance framework in terms of moment matching.”
Learning Implicit Generative Models by Matching Perceptual Features (2019, background): “Finally, our loss function is related to the loss function in McGan [31].”
How Generative Adversarial Networks and Their Variants Work (2017, methodology): “Mean and covariance feature matching GAN (McGAN) [85] extended this concept to match not only the l q mean feature but also the second order moment feature by using the singular value decomposition concept; it aims to also maximize an embedding covariance discrepancy between p data ( x ) and p g ( x”
Learning Implicit Generative Models with the Method of Learned Moments (2018, background): “Similarly, one can think of discriminator penalties – such as gradient penalties in Wasserstein GANs (Gulrajani et al., 2017), the DRAGAN penalty (Kodali et al., 2017), and average second moment penalty Fisher GANs (Mroueh and Sercu, 2017) – as terms to control Σ .”
A Review on Generative Adversarial Networks: Algorithms, Theory, and Applications (2020, background): “New Divergences: New families of Integral Probability Metrics (IPMs) for training GANs such as Fisher GAN [455], [456], mean and covariance feature matching GAN (McGan) [457], and Sobolev GAN [458], have been proposed.”
Memorization Precedes Generation: Learning Unsupervised GANs with Memory Networks (2018, background): “07 Wasserstein + gradient penalty Fisher GAN (Mroueh et al., 2017) 7.”
PassGAN: A Deep Learning Approach for Password Guessing (2017, unknown): “[25], there have been several improvements on GANs [3, 4, 6–8, 10, 14, 16, 27, 35, 36, 42, 47, 53, 55, 56, 58–60, 63, 66, 69, 74, 76, 80, 87, 90, 94, 96], where each new paper provides novel improvements in the domain.”
Fisher GAN (2017, methodology): “In this case, an IPM with data-independent constraint will be equivalent to mean matching [8].”
cGANs with Multi-Hinge Loss (2019, methodology): “The well known mean feature matching loss [38, 32] we use to train the generator is: LG,FM = ‖Ez∼pz,y∼pd [Cfeat(G(z, y)]− Ex∼pd [Cfeat(x)]‖1 , (9)”