Autoregressive Quantile Networks for Generative Modeling
This addresses the problem of improving perceptual quality and diversity in generative models for machine learning researchers, though it appears incremental as it extends existing models like PixelCNN.
The authors tackled generative modeling by introducing autoregressive implicit quantile networks (AIQN), a fundamentally different approach that implicitly captures distributions using quantile regression, achieving superior perceptual quality and improvements in evaluation metrics without losing sample diversity, as demonstrated on CIFAR-10 and ImageNet with metrics like Inception score and FID.
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.