隨堂測驗：GAN 解剖
透過集合功能整理內容
你可以依據偏好儲存及分類內容。
是非題：辨別器網路和產生器網路只會透過產生器產生的資料，以及辨別器產生的標籤相互影響。在反向傳播方面，它們是獨立的網路。
是
錯誤：在產生器訓練期間，梯度會透過鑑別器網路傳播至產生器網路 (雖然鑑別器不會在產生器訓練期間更新其權重)。
否
正確：在產生器訓練期間，梯度會透過鑑別器網路傳播至產生器網路 (雖然鑑別器不會在產生器訓練期間更新其權重)。因此，判別器網路中的權重會影響生成器網路的更新作業。
是非題：一般 GAN 會同時訓練產生器和鑑別器。
是
答錯了。一般來說，GAN 會交替訓練鑑別器和產生器。我們有幾項 [研究](https://arxiv.org/abs/1706.04156)，可同時訓練產生器和鑑別器。
否
沒錯。一般來說，GAN 會交替訓練鑑別器和產生器。
是非題：GAN 在判別器和產生器訓練時，一律會使用相同的損失函式。
是
答錯了。雖然 GAN 可以使用相同的損失值來訓練產生器和判別器 (或相同的損失值，但符號不同)，但這並非必要。事實上，針對判別器和產生器使用不同的損失值更為常見。
否
沒錯。雖然 GAN 可以使用相同的損失值來訓練產生器和判別器 (或相同的損失值，但符號不同)，但這並非必要。事實上，針對判別器和產生器使用不同的損失值更為常見。
除非另有註明，否則本頁面中的內容是採用創用 CC 姓名標示 4.0 授權，程式碼範例則為阿帕契 2.0 授權。詳情請參閱《Google Developers 網站政策》。Java 是 Oracle 和/或其關聯企業的註冊商標。
上次更新時間：2025-02-26 (世界標準時間)。
[null,null,["上次更新時間：2025-02-26 (世界標準時間)。"],[[["\u003cp\u003eDuring generator training, gradients propagate through the discriminator to the generator, influencing its updates.\u003c/p\u003e\n"],["\u003cp\u003eA typical GAN alternates between training the discriminator and training the generator, rather than simultaneous training.\u003c/p\u003e\n"],["\u003cp\u003eGANs often employ different loss functions for the discriminator and generator, optimizing each network separately.\u003c/p\u003e\n"]]],[],null,["# Check Your Understanding: GAN Anatomy\n\n\u003cbr /\u003e\n\nTrue or false: the discriminator network and generator network influence each other solely through the data produced by the generator and the labels produced by the discriminator. When it comes to backpropagation, they are separate networks. \nTrue \nIncorrect: during generator training, gradients propagate through the discriminator network to the generator network (although the discriminator does not update its weights during generator training). \nFalse \nCorrect: during generator training, gradients propagate through the discriminator network to the generator network (although the discriminator does not update its weights during generator training). So the weights in the discriminator network influence the updates to the generator network. \nTrue or false: a typical GAN trains the generator and the discriminator simultaneously. \nTrue \nIncorrect. A typical GAN alternates between training the discriminator and training the generator. There is some \\[research \\](https://arxiv.org/abs/1706.04156) on training the generator and discriminator simultaneously. \nFalse \nCorrect. A typical GAN alternates between training the discriminator and training the generator. \nTrue or false: a GAN always uses the same loss function for both discriminator and generator training. \nTrue \nIncorrect. While it's possible for a GAN to use the same loss for both generator and discriminator training (or the same loss differing only in sign), it's not required. In fact it's more common to use different losses for the discriminator and the generator. \nFalse \nCorrect. While it's possible for a GAN to use the same loss for both generator and discriminator training (or the same loss differing only in sign), it's not required. In fact it's more common to use different losses for the discriminator and the generator."]]
