Common designs: 1) Pipeline: neural front end extracts symb…

Common designs:
1) Pipeline: neural front end extracts symbols (objects/relations) → symbolic back end reasons with rules/constraints.
2) Differentiable logic: embed soft logic/probabilistic reasoning as differentiable layers for end-to-end training.
3) Neuro‑guided program induction/ILP: learn or refine rules/programs from examples with neural guidance.