EricGJB / UN-Based-RISBF

7 stars 8 forks source link

请教个问题哈 #3

Open MayeZhang opened 3 years ago

MayeZhang commented 3 years ago

大佬请教个问题哈!就是按您论文里面的方法去做一些类似的工作,但是验证集上的loss一直下不去啊? 按理说验证集上的loss不应该和训练集上的差不太多吗,都是一个分布产生的。

image

EricGJB commented 3 years ago

你好,不同问题情况不一样,可能需要再调参,包括网络规模,学习率等。数据量可以适当大一些,防止过拟合,以及用上BN层等

---原始邮件--- 发件人: "Ch @.> 发送时间: 2021年10月9日(周六) 上午9:32 收件人: @.>; 抄送: @.***>; 主题: [EricGJB/UN-Based-RISBF] 请教个问题哈 (#3)

大佬请教个问题哈!就是按您论文里面的方法去做一些类似的工作,但是验证集上的loss一直下不去啊? 按理说验证集上的loss不应该和训练集上的差不太多吗,都是一个分布产生的。

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

MayeZhang commented 3 years ago

好的好的,谢谢大佬哈

在2021年10月9日 @.***> 写道:

你好,不同问题情况不一样,可能需要再调参,包括网络规模,学习率等。数据量可以适当大一些,防止过拟合,以及用上BN层等

---原始邮件--- 发件人: "Ch @.> 发送时间: 2021年10月9日(周六) 上午9:32 收件人: @.>; 抄送: @.***>; 主题: [EricGJB/UN-Based-RISBF] 请教个问题哈 (#3)

大佬请教个问题哈!就是按您论文里面的方法去做一些类似的工作,但是验证集上的loss一直下不去啊? 按理说验证集上的loss不应该和训练集上的差不太多吗,都是一个分布产生的。

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.

— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.