logic-tensor-networks

The paper introduces function-symbol grounding in LTN-GANs to handle hard structural constraints in generative models, showing it learns margin distributions more faithfully than existing methods, as validated on high-resolution datasets.