Fine-tuning with local learning rules helps to compress and accelerate spiking neural networks without accuracy loss - Neural Computing and Applications