Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

请问 是怎么防止过拟合的? #1

Open
epoch8260 opened this issue May 13, 2022 · 5 comments
Open

请问 是怎么防止过拟合的? #1

epoch8260 opened this issue May 13, 2022 · 5 comments

Comments

@epoch8260
Copy link

Total params: 102.552034M,难道显著性检测的结果都是过拟合来的嘛

@MathLee
Copy link
Owner

MathLee commented May 13, 2022

Total params: 102.552034M,难道显著性检测的结果都是过拟合来的嘛

参数量大不意味是过拟合。我们有用数据扩充和深层监督来防止过拟合。

@epoch8260
Copy link
Author

epoch8260 commented May 13, 2022 via email

@epoch8260
Copy link
Author

epoch8260 commented May 13, 2022 via email

@MathLee
Copy link
Owner

MathLee commented May 13, 2022

深监督是不能防止过拟合的,深监督是为了更好地梯度回传,加不加深监督,对于256✖️256的输入,意义不大

---- 回复的原邮件 ---- | 发件人 | Gongyang @.> | | 日期 | 2022年05月13日 21:07 | | 收件人 | @.> | | 抄送至 | @.@.> | | 主题 | Re: [MathLee/ACCoNet] 请问 是怎么防止过拟合的? (Issue #1) | Total params: 102.552034M,难道显著性检测的结果都是过拟合来的嘛 参数量大不意味是过拟合。我们有用数据扩充和深层监督来防治过拟合。 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

深监督在最早提出的论文中是有防止过拟合这一说法的,曾经在另外文章审稿意见中遇到过,查阅过相关文献。这也是SOD中常用的trick.

@MathLee
Copy link
Owner

MathLee commented May 13, 2022

能否展示一下你们的训练和验证loss曲线,师弟一直复现不出来,托我咨询一下。

---- 回复的原邮件 ---- | 发件人 | Gongyang @.> | | 日期 | 2022年05月13日 21:09 | | 收件人 | @.> | | 抄送至 | @.@.> | | 主题 | Re: [MathLee/ACCoNet] 请问 是怎么防止过拟合的? (Issue #1) | 参数量大不意味是过拟合。我们有用数据扩充和深层监督来防止过拟合。 在 2022-05-13 20:54:41,"epoch8260" @.> 写道: Total params: 102.552034M,难道显著性检测的结果都是过拟合来的嘛 — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: @.> — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

这个工作完成的比较早,录用的比较晚,我们在服务器上仅保留了一个比较好的结果,并公布了权重。真是抱歉。

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants