Sparse-firing regularization methods for spiking neural networks with time-to-first spike coding