Driftuard -AEDL: Concept-Drift-Aware Continual Neuro-Symbolic Causal Inference for Evolving Time Series
The results show that pairing lightweight symbolic constraints with drift-triggered continual learning yields causal inference that stays valid as the world changes.