Maya-CL: Nociceptive Metaplasticity and Vairagya-Governed Heterosynaptic Decay for Continual Learning in Spiking Neural Networks

We present Maya-CL, scaling the Maya affective SNN architecture to the Split-CIFAR-10 Task-Incremental Learning benchmark. Maya-CL combines nociceptive metaplasticity, Vairagya-governed heterosynaptic gradient masking, and BCM boundary decay on a shared convolutional backbone without replay or architectural expansion. A three-condition ablation study isolates the Vairagya contribution: lability elevation alone degrades AA by 5.67% while Vairagya masking recovers +3.48% AA and +3.76% BWT. Full Maya-CL achieves AA 62.38%, BWT −30.55%, FWT +40.00% under TIL evaluation. To our knowledge, no prior SNN architecture unifies these three mechanisms on a standard visual continual learning benchmark. Codebase: https://github.com/venky2099/Maya-CL