Tag
This paper introduces gammaILP, a fully differentiable framework for learning first-order rules directly from image data without label leakage, addressing challenges in symbol grounding and predicate invention.