Learning Neurosymbolic Generative Models via Program Synthesis

TL;DR

This paper introduces a neurosymbolic generative modeling approach that integrates program synthesis to explicitly encode and learn complex global structures in data, significantly improving image generation and completion tasks.

Contribution

It presents a novel framework combining program synthesis with generative models to better capture global structures in data, addressing limitations of current methods.

Findings

Outperforms state-of-the-art in generating images with global structure

Improves image completion quality for structured data

Effective on both synthetic and real-world datasets

Abstract

Significant strides have been made toward designing better generative models in recent years. Despite this progress, however, state-of-the-art approaches are still largely unable to capture complex global structure in data. For example, images of buildings typically contain spatial patterns such as windows repeating at regular intervals; state-of-the-art generative methods can't easily reproduce these structures. We propose to address this problem by incorporating programs representing global structure into the generative model---e.g., a 2D for-loop may represent a configuration of windows. Furthermore, we propose a framework for learning these models by leveraging program synthesis to generate training data. On both synthetic and real-world data, we demonstrate that our approach is substantially better than the state-of-the-art at both generating and completing images that contain global structure.