Fine-Tuning Method for Deep Learning