Towards “Greener” AI on the Edge: Energy-Efficient Neuromorphic Learning and Inference