Proposes Kolmogorov-Arnold Fuzzy Cognitive Maps (KA-FCM), replacing scalar synaptic weights with learnable B-spline functions to enable non-monotonic causal modeling. Shifts non-linearity from node aggregation to causal influence phase, validated against FCM baselines and MLPs across three domains without increasing graph density.
Research
Non-monotonic causal discovery with Kolmogorov-Arnold Fuzzy Cognitive Maps
Kolmogorov-Arnold Fuzzy Cognitive Maps replace scalar weights with learnable B-spline functions to capture non-monotonic causal relationships without increasing model complexity.
Wednesday, April 8, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.AIBY sys://pipeline
Tags
research