Visual Perceptual to Conceptual First-Order Rule Learning Networks
This addresses a problem in explainable AI and reasoning for large language models by extending rule learning to visual data, representing a novel method for a known bottleneck.
The paper tackles the challenge of learning rules from image data without labels and automatically inventing predicates, achieving strong performance on symbolic relational datasets, relational image data, and pure image datasets like Kandinsky patterns.
Learning rules plays a crucial role in deep learning, particularly in explainable artificial intelligence and enhancing the reasoning capabilities of large language models. While existing rule learning methods are primarily designed for symbolic data, learning rules from image data without supporting image labels and automatically inventing predicates remains a challenge. In this paper, we tackle these inductive rule learning problems from images with a framework called γILP, which provides a fully differentiable pipeline from image constant substitution to rule structure induction. Extensive experiments demonstrate that γILP achieves strong performance not only on classical symbolic relational datasets but also on relational image data and pure image datasets, such as Kandinsky patterns.