試験
通常
試験時間: 00:00問題時間: 00:00
問題一覧
問題 54 / 65
What are the primary causes of the vanishing gradient problem in deep neural networks and the common remedies?
①The main cause is that the learning rate is too high, and the remedy is to lower the learning rate.
②The primary cause is insufficient data, and the countermeasure is data augmentation.
③The main cause is that the weight initialization is excessively large, and the remedy is to use small initial values.
④To counteract vanishing gradients caused by the choice of activation function and network depth, ReLU, batch normalization, and residual connections are used.
⑤The primary cause is that the loss function is non-convex, and the remedy is to change it to a convex function.