Anytime Sampling for Autoregressive Models via Ordered Autoencoding
This addresses a deployment bottleneck for autoregressive models in real-time applications like image and audio generation, offering an incremental improvement in computational adaptability.
The paper tackles the problem that autoregressive models cannot be interrupted during sampling, hindering deployment due to slow sequential generation, by proposing a new family of models that enable anytime sampling via ordered autoencoding, resulting in almost no loss in sample quality using only 60% to 80% of latent dimensions for image data.
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 .