如何調整一個不收斂的神經網路

Candy_GL發表於2019-03-03

轉自:https://blog.csdn.net/wangdongwei0/article/details/81256423

沒錯的,又是傳送門

大概分為以下幾點: 
1、You Forgot to Normalize Your Data 
2、You Forgot to Check your Results 
3、You Forgot to Preprocess Your Data 
4、You Forgot to use any Regularization 
5、You Used a too Large Batch Size 
6、You Used an Incorrect Learning Rate 
7、You Used the Wrong Activation Function on the Final Layer 
8、Your Network contains Bad Gradients 
9、You Initialized your Network Weights Incorrectly 
10、You Used a Network that was too Deep 
11、You Used the Wrong Number of Hidden Units

相關文章