Autoregressive models are often employed to learn distributions of image data by decomposing the -dimensional density function into a product of one-dimensional conditional distributions. Each conditional depends on preceding variables (pixels, in the case of image data), making the order in which variables are processed fundamental to the model performance. In this paper, we study the problem of observing a small subset of image pixels (referred to as a pixel patch) to predict the unobserved parts of the image. As our prediction mechanism, we propose a generalized and computationally efficient version of the convolutional neural autoregressive distribution estimator (ConvNADE) model adapted for real-valued and color images. Moreover, we investigate the quality of image reconstruction when observing both random pixel patches and low-discrepancy pixel patches inspired by quasi-Monte Carlo theory. Experiments on benchmark datasets demonstrate that choosing the pixels akin to a low-discrepancy sequence reduces test loss and produces more realistic reconstructed images.
View on arXiv@article{emmett-iwaniw2025_2506.05391,
title={ Enhancing Neural Autoregressive Distribution Estimators for Image Reconstruction },
author={ Ambrose Emmett-Iwaniw and Nathan Kirk },
journal={arXiv preprint arXiv:2506.05391},
year={ 2025 }
}