Anytime Sampling for Autoregressive Models via Ordered Autoencoding

TL;DR

This paper introduces a novel autoregressive model framework that allows for anytime sampling by ordering latent dimensions based on importance, enabling flexible trade-offs between sample quality and computational resources.

Contribution

It proposes a structured latent space inspired by PCA for autoregressive models, enabling truncation during sampling to adapt to computational constraints.

Findings

Sample quality degrades gracefully with reduced computation.

Using 60-80% of latent dimensions maintains near-original quality.

The method is effective for image and audio generation tasks.

Abstract

Autoregressive models are widely used for tasks such as image and audio generation. The sampling process of these models, however, does not allow interruptions and cannot adapt to real-time computational resources. This challenge impedes the deployment of powerful autoregressive models, which involve a slow sampling process that is sequential in nature and typically scales linearly with respect to the data dimension. To address this difficulty, we propose a new family of autoregressive models that enables anytime sampling. Inspired by Principal Component Analysis, we learn a structured representation space where dimensions are ordered based on their importance with respect to reconstruction. Using an autoregressive model in this latent space, we trade off sample quality for computational efficiency by truncating the generation process before decoding into the original data space. Experimentally, we demonstrate in several image and audio generation tasks that sample quality degrades gracefully as we reduce the computational budget for sampling. The approach suffers almost no loss in sample quality (measured by FID) using only 60\% to 80\% of all latent dimensions for image data. Code is available at https://github.com/Newbeeer/Anytime-Auto-Regressive-Model .