📝 SummarXiv

Abstract

A central feature of biological intelligence is the ability to learn to learn, enabling rapid adaptation to novel tasks and environments. Yet its neural basis remains elusive, particularly regarding intrinsic properties, as conventional models rely on simplified point-neuron approximations that neglect their dynamics. Inspired by astrocyte-mediated neuromodulation, we propose a hierarchically modulated recurrent spiking neural network (HM-RSNN) that models learning-to-learn with regulation of intrinsic neuronal properties at two spatiotemporal scales. Global modulation captures task-dependent gating of plasticity driven by wide-field calcium waves, whereas local adaptation simulates microdomain calcium-mediated fine-tuning of intrinsic properties within task-relevant subspaces. We evaluate HM-RSNN on four cognitive tasks, demonstrating its computational advantages over standard RSNNs and artificial neural networks, and revealing task-dependent adaptations across multiple scales, including intrinsic properties, neuronal specialization, membrane potential dynamics, and network modularity. Converging evidence and biological consistency position HM-RSNN as a biologically grounded framework, providing testable insights into how astrocyte-mediated hierarchical modulation of intrinsic properties shapes multi-scale neural dynamics that support learning-to-learn.