Neural symbolic processing refers to approaches that combine the learning of neural networks with symbolic knowledge representation and logical reasoning.
Formally, it integrates differentiable neural models f_θ with symbolic structures S, where S consists of formal expressions, rules, or graphs used for inference.
For example, differentiable logic modules (e.g., Logic Tensor Networks) can be embedded into neural networks to ensure logical consistency during training.
A typical approach formulates learning as the optimization of a loss L, which is based on both neural predictions and symbolic constraints:
\[ L = L_{neural} + \lambda L_{symbolic} \]
where \( L_{symbolic} \) measures the violation of symbolic rules and \( \lambda \) is a weighting parameter.
Neural symbolic processing addresses the challenges of interpretability, generalization, and knowledge integration in AI systems.
Definition:
“Neural symbolic processing is a hybrid AI approach that combines neural learning models with symbolic knowledge and logical reasoning to enable explainable and generalizable intelligence.”
Source:
Garcez, A. d’Avila, Lamb, L. C., & Gabbay, D. M. (2019). Neural-Symbolic Computing: An Effective Methodology for Principled Integration of Machine Learning and Reasoning. Frontiers in Artificial Intelligence and Applications, 320.