Deep Learning Optimization for Edge Devices: Analysis of Training Quantization Parameters