Maya-Smriti: Episodic Memory as a Biological Prior for Class-Incremental Learning in Affective Spiking Neural Networks

We present Maya-Smriti, extending the Maya affective SNN architecture to Class-Incremental Learning (CIL) on Split-CIFAR-10 through a minimal class-wise ring buffer with interleaved replay. We introduce Buddhi — discriminative intellect — as a fifth affective dimension governing Vairagya consolidation rate, and identify Ahamkara — ego-driven task attachment — as the failure mode responsible for affective mechanism collapse under CIL without replay. A five-condition ablation study establishes that Maya mechanisms alone at CIL (AA=17.77%) produce performance indistinguishable from the SGD baseline (AA=17.98%), while full Maya-Smriti with replay achieves AA=31.84%, BWT=−68.36%, outperforming replay-only by +0.77% AA and +1.02% BWT. We further identify affective quiescence — the suppression of Bhaya throughout replay-stabilised training — as a novel emergent property of affective SNN architectures. Part of the Maya Research Series. Extends Swaminathan (2026a, 2026b, 2026c).