Dropout

Dropout是Google提出的一種正則化技術^[1]，用以在人工神經網絡中對抗過擬合。Dropout有效的原因，是它能夠避免在訓練數據上產生複雜的相互適應。^[1]Dropout這個術語代指在神經網絡中丟棄部分神經元（包括隱藏神經元和可見神經元）。^[2]^[3]在訓練階段，dropout使得每次只有部分網絡結構得到更新，因而是一種高效的神經網絡模型平均化的方法。^[4]

這是一篇與人工神經網絡相關的小作品。你可以透過編輯或修訂擴充其內容。

參考文獻[編輯]

^ ^1.0 ^1.1 [1],「System and method for addressing overfitting in a neural network」（頁面存檔備份，存於網際網路檔案館）存档副本. [2019-12-11]. 原始內容存檔於2021-07-25.
^ Dropout: A Simple Way to Prevent Neural Networks from Overfitting. Jmlr.org. [July 26, 2015]. （原始內容存檔於2019-12-05）. 參數|newspaper=與模板{{cite web}}不匹配（建議改用{{cite news}}或|website=） (幫助)
^ Warde-Farley, David; Goodfellow, Ian J.; Courville, Aaron; Bengio, Yoshua. An empirical analysis of dropout in piecewise linear networks. 2013-12-20. arXiv:1312.6197  [stat.ML].
^ Hinton, Geoffrey E.; Srivastava, Nitish; Krizhevsky, Alex; Sutskever, Ilya; Salakhutdinov, Ruslan R. Improving neural networks by preventing co-adaptation of feature detectors. 2012. arXiv:1207.0580  [cs.NE].

[pat-1] 1.0 ^1.1 [1],「System and method for addressing overfitting in a neural network」（頁面存檔備份，存於網際網路檔案館）存档副本. [2019-12-11]. 原始內容存檔於2021-07-25.

[MyUser_Jmlr.org_July_26_2015c-2] Dropout: A Simple Way to Prevent Neural Networks from Overfitting. Jmlr.org. [July 26, 2015]. （原始內容存檔於2019-12-05）. 參數|newspaper=與模板{{cite web}}不匹配（建議改用{{cite news}}或|website=） (幫助)

[3] Warde-Farley, David; Goodfellow, Ian J.; Courville, Aaron; Bengio, Yoshua. An empirical analysis of dropout in piecewise linear networks. 2013-12-20. arXiv:1312.6197  [stat.ML].

[MyUser_Arxiv.org_July_26_2015c-4] Hinton, Geoffrey E.; Srivastava, Nitish; Krizhevsky, Alex; Sutskever, Ilya; Salakhutdinov, Ruslan R. Improving neural networks by preventing co-adaptation of feature detectors. 2012. arXiv:1207.0580  [cs.NE].

[1]

[2]

[3]

[4]