OR/AND neurons and the development of interpretable logic models

Area of Science:

• Neuro-fuzzy systems
• Computational intelligence
• Logic-based data analysis

Background:

• Fuzzy logic networks offer advanced data analysis capabilities.
• OR/AND neurons provide parametric flexibility and interpretability.
• Existing methods may lack comprehensive optimization and interpretability.

Purpose of the Study:

• To develop and optimize fuzzy logic networks using OR/AND neurons.
• To enable logic-based approximation of multidimensional mappings.
• To enhance network interpretability and reduce complexity.

Main Methods:

• Utilized homogeneous architectures with OR/AND neurons.
• Employed genetic optimization for structural and parametric tuning.
• Implemented a modular network design reflecting logic expression decomposition.
• Incorporated a pruning mechanism for enhanced interpretability.

Main Results:

• Demonstrated network performance on Boolean and synthetic multivalued data.
• Achieved logic approximation of transformations from [0, 1]n to [0, 1]m.
• Generated logic expressions for machine learning datasets.
• Showcased enhanced interpretability and reduced network size through optimization.

Conclusions:

• The developed fuzzy logic network effectively performs logic-based data analysis and mapping approximation.
• Genetic optimization is crucial for structural and parametric refinement, improving interpretability.
• The network architecture and design procedure offer a robust framework for complex logic tasks.