Maya-Śūnyatā: Karma-Weighted Synaptic Pruning for Class-Incremental Learning in Affective Spiking Neural Networks

What remains when everything inessential has been released? Śūnyatā does not destroy. It completes. We present Maya-Śūnyatā, the eighth paper in the Maya Research Series, which extends our affective spiking neural network (SNN) architecture for class-incremental learning (CIL) on Split-CIFAR-100 with two new mechanisms. Karma (कर्म) is introduced as the first second-order plasticity history signal in the series: the absolute integral of per-synapse weight trajectory changes accumulated across tasks. A synapse that has been repeatedly pulled in conflicting directions by successive tasks carries that interference in its weight trajectory. Karma makes this interference legible to the architecture. Śūnyatā…