https://blog.csdn.net/u012582383/article/details/123292530
OptimizerSGD参考视频：https://www.bilibili.com/video/BV1bP4y1p7Gq?p=1
小批量随机梯度下降
Momentum平滑
Adam非常平滑，鲁棒性强，学习率不敏感，预处理影响小
PyTorch使用Tensorboard: https://blog.csdn.net/u012582383/article/details/123052577?spm=1001.2014.3001.5501
