Discriminator Rejection Sampling

By Samaneh Azadi, Catherine Olsson, Trevor Darrell, Ian J. Goodfellow, Augustus Odena. 2018

144 citations recorded by the source corpus.

We propose a rejection sampling scheme using the discriminator of a GAN to approximately correct errors in the GAN generator distribution. We show that under quite strict assumptions, this will allow us to recover the data distribution exactly. We then examine where those strict assumptions break down and design a practical algorithm - called Discriminator Rejection Sampling (DRS) - that can be used on real data-sets. Finally, we demonstrate the efficacy of DRS on a mixture of Gaussians and on the SAGAN model, state-of-the-art in the image generation task at the time of developing this work. On ImageNet, we train an improved baseline that increases the Inception Score from 52.52 to 62.36 and reduces the Frechet Inception Distance from 18.65 to 14.79. We then use DRS to further improve on this baseline, improving the Inception Score to 76.08 and the FID to 13.75.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Radiology, Nuclear Medicine and Imaging, Generative Models, Probabilistic & Bayesian Methods

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 Radiology, Nuclear Medicine and Imaging

In Generative Models

In Probabilistic & Bayesian Methods

Citations in the corpus

Residual Energy-Based Models for Text Generation (2020, background): “Azadi et al. (2018) also share our same goal but their generator is not locally normalized and they propose to improve the sampling from the generator by using the discriminator for rejection sampling.”
Your GAN is Secretly an Energy-based Model and You Should use Discriminator Driven Latent Sampling (2020, methodology): “In Table 3, we show the Inception score improvements from DDLS on CIFAR-10 and CelebA, compared to MH-GAN [11] and DRS [10], following the same evaluation protocol and using the same baseline models (DCGAN and WGAN) in [11].”
Generating Diverse High-Fidelity Images with VQ-VAE-2 (2019, background): “Improving sample quality by rejection sampling has been previously explored for GANs [1] as well as for VAEs [3] which combines a learned rejecting sampling proposal with the prior in order to reduce its gap with the aggregate posterior.”
Closed-Loop Memory GAN for Continual Learning (2018, methodology): “In addition to CFM, we implemented a more complex procedure, "Discriminator Rejection Sampling" (DRS) proposed in [Azadi et al., 2018].”
Freeze the Discriminator: a Simple Baseline for Fine-Tuning GANs (2020, methodology): “…a target domain [24, 47, 42], using an auxiliary task to facilitate training [7, 26, 48, 49], improving an inference procedure of suboptimal models [2, 39, 29, 38], using an expressive prior distribution [13], actively choosing samples to give supervision for conditional generation [29], or…”
KALE: When Energy-Based Learning Meets Adversarial Training (2020, methodology): “While there has been a recent interest in using the discriminator/energy to improve the quality of the generator/base (Azadi et al., 2019; Grover et al., 2019; Turner et al., 2019; Tanaka, 2019) (Section 5), our approach emerges naturally from the energy model we consider.”
Exponential Tilting of Generative Models: Improving Sample Quality by Training and Sampling from Latent Energy (2020, methodology): “Discriminator rejection sampling [2] and Metropolis-Hastings GANs [24] use the discriminator as a criterion of accepting or rejecting samples from the generator.”
Semantic Bottleneck Scene Generation (2019, background): “Training GANs is notoriously hard and recent efforts focused on improving neural architectures [21, 44, 9], loss functions [1], regularization [15, 30], large-scale training [7], self-supervision [10], and sampling [7, 4].”
Refining Deep Generative Models via Discriminator Gradient Flow (2020, methodology): “Thus far, prior work (Azadi et al., 2019; Turner et al., 2019; Tanaka, 2019; Che et al., 2020) has focused on improving samples for GANs with scalar-valued discriminators, which comprises the canonical GAN as well as recent variants, e.g., WGAN (Gulrajani et al., 2017), and SNGAN (Miyato et al.,…”
GAN Compression: Efficient Architectures for Interactive Conditional GANs (2020, background): “Although we aim to compress the generator, a discriminator D stores useful knowledge of a learned GAN as D learns to spot the weakness of the current generator [3].”
Training Deep Energy-Based Models with f-Divergence Minimization (2020, methodology): “With the goal of sampling from real data distribution p , we employ rejection sampling (Halton, 1970; Azadi et al., 2018), where we accept samples from the EBM with a probability proportional to the density ratio.”
Instance Selection for GANs (2020, background): “…of current techniques attempt to eliminate low quality samples after the model is trained, either by changing the model distribution by truncating the latent space [2, 10] or by performing some form of rejection sampling using a trained discriminator to inform the rejection process [1, 4, 26].”
Residual Energy-Based Models for Text (2021, background): “Azadi et al. (2019) also share our goal but their generator is not locally normalized and they propose to improve the sampling from the generator by using the discriminator for rejection sampling.”
Likelihood-free MCMC with Amortized Approximate Ratio Estimators (2019, background): “…2015; Mohamed & Lakshmi-narayanan, 2016; Gutmann et al., 2017; Dutta et al., 2016; Tran et al., 2017; Brehmer et al., 2020) and is especially prominent in the area of Generative Adversarial Networks ( GAN s) (Goodfellow et al., 2014; Uehara et al., 2016; Turner et al., 2018; Azadi et al., 2018).”
Prescribed Generative Adversarial Networks (2019, background): “Azadi et al. (2018) and Turner et al. (2018) use sampling mechanisms to correct errors of the generative distribution.”