Autoregressive Quantile Networks for Generative Modeling

By Georg Ostrovski, Will Dabney, R´emi Munos. 2018

94 citations recorded by the source corpus.

We introduce autoregressive implicit quantile networks (AIQN), a fundamentally different approach to generative modeling than those commonly used, that implicitly captures the distribution using quantile regression. AIQN is able to achieve superior perceptual quality and improvements in evaluation metrics, without incurring a loss of sample diversity. The method can be applied to many existing models and architectures. In this work we extend the PixelCNN model with AIQN and demonstrate results on CIFAR-10 and ImageNet using Inception score, FID, non-cherry-picked samples, and inpainting results. We consistently observe that AIQN yields a highly stable algorithm that improves perceptual quality while maintaining a highly diverse distribution.

Subfields: 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 Computer Vision and Pattern Recognition

In Generative Models

Citations in the corpus

Denoising Diffusion Probabilistic Models (2020, neutral): “80 PixelIQN [43] 5.”
Your GAN is Secretly an Energy-based Model and You Should use Discriminator Driven Latent Sampling (2020, neutral): “93 PixelIQN (Ostrovski et al., 2018) 5.”
Generative Modeling by Estimating Gradients of the Data Distribution (2019, neutral): “93 PixelIQN [42] 5.”
Classification Accuracy Score for Conditional Generative Models (2019, methodology): “Finally, PixelIQN has similar performance to the newer GANs.”
Hierarchical Autoregressive Image Models with Auxiliary Decoders (2019, methodology): “The IS and FID are much worse than those reported for recent adversarial models (Brock et al., 2019), but they are in the same ball-park as those reported for PixelCNN on 32 × 32 ImageNet by Ostrovski et al. (2018) (IS 8.33, FID 33.27).”
Residual Flows for Invertible Generative Modeling (2019, neutral): “∗Results taken from Ostrovski et al. (2018).”
Preventing Posterior Collapse with delta-VAEs (2019, background): “While these models are good at capturing local statistics, they often fail to produce globally coherent structures (Ostrovski et al., 2018).”
Implicit Generation and Generalization in Energy-Based Models (2019, methodology): “On unconditional CIFAR-10, we ﬁnd higher inception scores and lower FID scores than Pixel-CNN and followup work such as PixelIQN (Ostrovski et al., 2018).”
Augmented Normalizing Flows: Bridging the Gap Between Generative Flows and Latent Variable Models (2020, methodology): “Sample quality For quantitative evaluation of sample quality, we report the Inception Score (IS) (Salimans et al., 2016) and the Fréchet Inception Distance (FID) (Heusel et al., 2017), expanding the table of Ostrovski et al. (2018).”
Probabilistic Forecasting with Spline Quantile Function RNNs (2019, methodology): “More recently, Ostrovski et al. (2018) proposed generative modeling via quantile functions and minimizing the CRPS.”
Learning Energy-Based Models in High-Dimensional Spaces with Multiscale Denoising-Score Matching (2019, background): “, 2015; Kingma and Dhariwal, 2018), or auto-regressive models (van den Oord et al., 2016; Ostrovski et al., 2018), energy-based models have significant advantages.”
Annealed Denoising Score Matching: Learning Energy-Based Models in High-Dimensional Spaces (2019, background): “, 2015; Kingma and Dhariwal, 2018), or auto-regressive models (van den Oord et al., 2016; Ostrovski et al., 2018), energy-based models have significant advantages.”
Deep Generative Modelling: A Comparative Review of VAEs, GANs, Normalizing Flows, Energy-Based and Autoregressive Models (2021, methodology): “Alternatively, quantile regression, which minimises Wasserstein distance, can be used to learn an approximation of the inverse cumulative distribution [172].”
Implicit Generation and Modeling with Energy Based Models (2019, unknown): “PixelCNN [Van Oord et al., 2016] 4.60 65.93 PixelIQN [Ostrovski et al., 2018] 5.29 49.46 EBM (single) 6.02 40.58 DCGAN [Radford et al., 2016] 6.40 37.11 WGAN + GP [Gulrajani et al., 2017] 6.50 36.4 EBM (10 historical ensemble) 6.78 38.2 SNGAN [Miyato et al., 2018] 8 We quantitatively evaluate image…”