【ML】Parameters in Neural Network
2022-07-01 本文已影响0人
盐果儿
1. Learning Rate:
Definition
Batch Size:
Training Loss:
Training Time:
iteration:
For instance, if the training set has 1000 samples, batch size = 10, then training the dataset need 100 iterations, training once means 1 epoch.
Learning rate is , step in every epoch
, training time
, problem: miss the lowest point.
Learning rate is , step in every epoch
, training time
, problem: training is too low,
find the local lowest point.
Learning Rate: https://en.wikipedia.org/wiki/Learning_rate
Learning Rate是在哪个图上走的？LR for gradient descent, step, weight updates in order to minimize the network's loss function.
学习率调整：
https://blog.csdn.net/lty_sky/article/details/105223840
局部最小值和鞍点：
https://blog.csdn.net/m0_37957160/article/details/121913311
