Hyperbolic Geometry Enhances ANFIS Interpretability and Performance
A new variant of the adaptive neuro-fuzzy inference system (ANFIS), named Hyperbolic ANFIS (HyperANFIS), has been developed by researchers. This innovative model utilizes hyperbolic geometry to enhance both the representation of rules and the accuracy of predictions. Unlike conventional ANFIS models that function within Euclidean space, HyperANFIS maintains the fundamental architecture and fuzzy semantics of ANFIS while executing rule-prototype learning, rule activation, and aggregation in hyperbolic space. This results in improved collaboration among rules and greater reliability of the generated IF-THEN statements, thus increasing the system's interpretability and effectiveness. The methodology and potential applications for transparent reasoning tasks are discussed in detail in the paper available on arXiv (2608.11768).
Key facts
- HyperANFIS is a hyperbolic extension of ANFIS
- It performs rule-prototype learning, rule activation, and consequent aggregation in hyperbolic space
- It preserves fuzzy semantics and core architecture of conventional ANFIS
- It retains the ability to generate interpretable IF-THEN rules
- It improves predictive accuracy, inter-rule collaboration, and credibility
- The paper is available on arXiv with identifier 2608.11768
- The research addresses limitations of Euclidean space in ANFIS models
- Hyperbolic geometry strengthens the fuzzy inference process
Entities
Institutions
- arXiv