Neuromorphic adaptation of the Tri-Memory continual learning framework

Investigate adaptations of the Tri-Memory Continual Learning architecture to spiking neural networks and neuromorphic hardware, and determine the feasibility and energy savings of implementing its Hebbian-style updates, pruning, replay, and consolidation on event-driven neuromorphic platforms.

Background

The paper discusses neuromorphic computing as a promising route for energy-efficient, on-device continual learning, noting spiking neural networks and local plasticity rules. However, a concrete mapping of the proposed Tri-Memory design to such hardware is not established.

The authors explicitly pose the question of how to adapt the framework to neuromorphic contexts, suggesting research into algorithm-hardware co-design, compatibility of learning rules, and empirical assessment of energy and performance.