The integration pattern “Neural as Predicate” combines symbolic and neural AI in the following way: within the logic program a neural network takes on the role of a learned predicate. Where a classical logic program would define a predicate as a hand-written rule, this pattern defines it through a trained function whose output enters the logical inference process.
Its representatives are in particular DeepProbLog (Manhaeve, Dumančić, Kimmig, Demeester, and De Raedt, KU Leuven, NeurIPS 2018), which integrates neural predicates into the probabilistic logic program ProbLog, and NeurASP (Yang, Ishay, and Lee, Arizona State University, IJCAI 2020), which carries out the same approach for Answer Set Programming. DeepProbLog compiles the program into arithmetic circuits, which enables efficient gradients; the system typically requires only hundreds rather than tens of thousands of training examples.
In personal-ontological terms this pattern is a special case of neuro-symbolic AI and thus shares its determination as a tool. Neither the symbolic nor the neural layer bears original intentionality — the symbols mean what a person has assigned to them; the learned weights are distributions over training data.
Ontological classification: Superordinate concept: Neuro-symbolic AI.
Methodological point
The pattern is methodologically interesting because it dramatically reduces the data intensity of neural systems: the logic program supplies the structure that the neural network does not have to learn from data. Thereby the learning process shifts from a tabula-rasa generalization to a structured extension of given knowledge landscapes.
This is not trivial: many practical applications fail not at model size but at the availability of qualitatively good training data. The shift from “more data” to “more structure” is the central architectural bet of the early 2020s in neuro-symbolic research.
Sources: Generated by querying the Personhood ontology. (Research basis: dossier Neuro-symbolic AI — worldwide research.)
Further sources:
- Manhaeve, Robin, Dumančić, Sebastijan, Kimmig, Angelika, Demeester, Thomas & De Raedt, Luc (2018): DeepProbLog. Neural Probabilistic Logic Programming. Advances in Neural Information Processing Systems 31, pp. 3753—3763.
- Yang, Zhun, Ishay, Adam & Lee, Joohyung (2020): NeurASP. Embracing Neural Networks into Answer Set Programming. Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence (IJCAI), Yokohama, pp. 1755—1762.
- Marra, Giuseppe, Dumančić, Sebastijan, Manhaeve, Robin & De Raedt, Luc (2024): From Statistical Relational to Neurosymbolic Artificial Intelligence. A Survey. Artificial Intelligence Vol. 328, Art. 104062 (arXiv preprint 2108.11451, August 2021).