Neuro-Inspired Error Propagation (NIEP): Refractory Period-Based Learning for Stable, Incremental AI Systems

This technical note presents Neuro-Inspired Error Propagation (NIEP), a biologically motivated learning scheme that stabilizes gradient updates through refractory gating and local temporal constraints. It introduces delayed eligibility traces, adaptive error budgeting, and a safe-commit protocol with shadow weights to improve stability, reduce interference, and enhance energy efficiency in continual and distributed learning systems. The approach provides a foundation for robust, low-interference training in next-generation AI architectures.