Neuro-Symbolic Integration refers to a hybrid AI paradigm that combines neural networks (NN) with symbolic AI methods to enable both subsymbolic pattern recognition and symbolic reasoning.
Formally, this can be understood as a combination of functions \(f_{NN}\) and symbolic operators \(S\), where \(f_{NN}\) operates on raw data and generates latent representations that are subsequently interpreted and manipulated by symbolic logic \(S\):
\[ y = S(f_{NN}(x), K) \]
where \(K\) represents symbolic knowledge.
This integration addresses the trade-offs between generalization and interpretability by coupling neural models with explicit knowledge.
Known approaches include models such as Neural Theorem Provers (Rocktäschel & Riedel, 2017) or Differentiable Inductive Logic Programming (Evans & Grefenstette, 2018).
Definition:
“Neuro-Symbolic Integration is a hybrid AI paradigm that unites neural networks and symbolic AI systems to enable both data-driven pattern recognition and logical reasoning within a single model.”
Source:
Garcez, A. d. A., Lamb, L. C., & Gabbay, D. M. (2019). Neural-Symbolic Cognitive Reasoning. Springer.