Learning to Modulate Random Weights: Neuromodulation-inspired Neural Networks For Efficient Continual Learning