The authors of the study cited as [75] propose a distribution-based method that embeds symbolic logic, such as propositional formulas and first-order logic, into neural network loss functions. These constraints are encoded as a distribution and incorporated into the optimization procedure using measures like the Fisher-Rao distance or Kullback-Leibler divergence to guide the neural network to adhere to symbolic constraints.

Authors

• Person: Not available Organization: arXiv
  Unlocking the Potential of Generative AI through Neuro-Symbolic ...

Sources

• Unlocking the Potential of Generative AI through Neuro-Symbolic ... arxiv.org arXiv via serper

Referenced by nodes (1)

• symbolic logic concept