Flow-GAN: Combining Maximum Likelihood and Adversarial Learning in Generative Models

By Aditya Grover, Manik Dhar, Stefano Ermon. 2017

226 citations recorded by the source corpus.

Adversarial learning of probabilistic models has recently emerged as a promising alternative to maximum likelihood. Implicit models such as generative adversarial networks (GAN) often generate better samples compared to explicit models trained by maximum likelihood. Yet, GANs sidestep the characterization of an explicit density which makes quantitative evaluations challenging. To bridge this gap, we propose Flow-GANs, a generative adversarial network for which we can perform exact likelihood evaluation, thus supporting both adversarial and maximum likelihood training. When trained adversarially, Flow-GANs generate high-quality samples but attain extremely poor log-likelihood scores, inferior even to a mixture model memorizing the training data; the opposite is true when trained by maximum likelihood. Results on MNIST and CIFAR-10 demonstrate that hybrid training can attain high held-out likelihoods while retaining visual fidelity in the generated samples.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Statistical and Nonlinear Physics, 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 Statistical and Nonlinear Physics

In Generative Models

Citations in the corpus

Evaluating Lossy Compression Rates of Deep Generative Models (2020, background): “Lossless compression metrics for GANs often give absurdly large bits-per-dimension (e.g. 10 14 ) which fails to reﬂect the true performance of the model (Grover et al., 2018; Danihelka et al., 2017).”
Glow: Generative Flow with Invertible 1x1 Convolutions (2018, background): “Downsides of GANs are their general lack of latent-space encoders, their general lack of full support over the data (Grover et al., 2018), their difficulty of optimization, and their difficulty of assessing overfitting and generalization.”
Improved Precision and Recall Metric for Assessing Generative Models (2019, background): “While the quality of results from generative adversarial networks (GAN) [8], variational autoencoders (VAE) [15], autoregressive models [28, 29], and likelihood-based models [6, 14] have seen rapid improvement recently [12, 9, 27, 20, 4, 13], the automatic evaluation of these results continues to be”
Normalizing Flows for Probabilistic Modeling and Inference (2019, methodology): “If we choose the generator of a generative adversarial network as a normalizing flow, we can train the flow parameters using adversarial training (Grover et al., 2018; Danihelka et al., 2017), with Wasserstein losses (Arjovsky et al.”
Constructing Unrestricted Adversarial Examples with Generative Models (2018, background): “Generative adversarial networks (GANs) A GAN [19, 20, 34, 35, 36] is composed of a generator gθ(z) and a discriminator dφ(x).”
Bias and Generalization in Deep Generative Models: An Empirical Study (2018, background): “In fact, even evaluating how “close” the learned distribution q is to p is an open question, and there is no commonly accepted evaluation metric [18–20].”
Analyzing Inverse Problems with Invertible Neural Networks (2018, background): “Invertible networks offer the opportunity to simultaneously optimize for losses on both the inand output domains (Grover et al., 2017), which allows for more effective training.”
Continuous-Time Flows for Efficient Inference and Density Estimation (2017, methodology): “GAN has recently been augmented by Flow-GAN (Grover et al., 2017) to incorporate a likelihood term for explicit density estimation.”
Guided Image Generation with Conditional Invertible Neural Networks (2019, methodology): “…likelihood training in RealNVPs, [36] trained i-2 RevNets in the same manner as adversarial auto-encoders, i.e. with a discriminator acting in latent rather than data space, and Flow-GANs [11] performed best using bidirectional training, a combination of maximum likelihood and adversarial loss.”
Conditional Generative Models are not Robust (2019, methodology): “Since flow algorithms are bijective, we could combine the two objective as was done in the flow-GAN paper [Grover et al., 2018].”
Disentangled Inference for GANs With Latently Invertible Autoencoder (2019, background): “There are alternative attempts of specifying the generator of GAN with normalizing ﬂow (Grover et al., 2017) or mapping images into feature spaces with partially invertible network (Lucas et al., 2019b).”
An Introduction to Variational Autoencoders (2019, background): “VAEs and GANs seem to have complementary properties: while GANs can generate images of high subjective perceptual quality, they tend to lack full support over the data (Grover et al., 2018), as opposed to likelihood-based generative models.”
PointFlow: 3D Point Cloud Generation With Continuous Normalizing Flows (2019, background): “It has also been explored as a metric for evaluating GANs [50].”
Fair Generative Modeling via Weak Supervision (2019, unknown): “, 2014) or hybrids (Grover et al., 2018).”
BayesFlow: Learning Complex Stochastic Models With Invertible Neural Networks (2020, methodology): “In the context of Bayesian inference, the target distribution is the posterior p ( θ | x 1: N ) of model parameters given observed data.”
Adversarial Transfer Learning (2018, background): “[46] similarly found a disconnect even in a GAN with tractable log-likelihood.”
On GANs and GMMs (2018, methodology): “Recent advances in deep learning, have also enabled unsupervised learning of full sized images using various models: Variational Auto Encoders [21, 17], PixelCNN [40, 39, 23, 38], Normalizing Flow [9, 8] and Flow GAN [14].”
On the Discrimination-Generalization Tradeoff in GANs (2017, methodology): “The FlowGAN (Grover et al., 2017) proposed to combine the WGAN loss and the log likelihood to solve the inconsistency problem.”
A Lagrangian Perspective on Latent Variable Generative Models (2018, background): “All these models attempt to fit an empirical data distribution, but differ in multiple ways: how they measure the similarity between distributions; whether or not they allow for efficient (amortized) inference; whether the latent variables should retain or discard information about the data; and how”