Deep learning test questions (1) answers and analysis

Deep learning test questions (1) answers and analysis

1. The definition of loss function The gap between the predicted value and the true value. Choose A.

2. What is given in the question is that the limit of a sigmoid function is at (0,1), here is the derivative S'(x)=S(x)(1-S(x)), so it should be 0. Choose B.

3. According to the compound function to find the second derivative, it is easy to get the answer 1/4. Choose A.

4. The first to be calculated is the gradient of the activation function, choose C.

5. Let's review the VGG network structure introduced earlier. VGG can be divided into 6 configurations A, A-LRN, B, C, D, and E according to the size and number of convolution kernels. Among them, VGG16 and VGG19 are more commonly used.
   As shown below
   

We conducted a specific analysis of VGG16 and found that VGG16 contains a total of
13 convolutional layers (Convolutional Layer), each with conv3-XXX represents
3 fully connected layers (Fully connected Layer), and FC-XXXX respectively represents
5 pooling layers (Pool layer), respectively represented by maxpool.
Among them, the convolutional layer and the fully connected layer have weight coefficients, so they are also called weight layers. The total number is 13+3=16, which is
the source of 16 in VGG16. (The pooling layer does not involve weights, so it does not belong to the weighting layer and is not counted).
So the 16 layers here refer to the number of layers that need to be trained. Choose C.

6. This question examines everyone's ability to build convolutional neural networks with keras. A should be selected according to the meaning of the question.
For example, keras builds VGG16 network part display

7. This question is similar to the previous question, refer to the picture above, choose A.

8. Choose D. This figure examines the concept of gradient disappearance, and dropout prevents overfitting.

Two, indefinite multiple choice questions

1. Choose ABD.
   Supplement: The Deep Belief Network (DBN) solves the optimization problem of deep neural networks through layer-by-layer training, and gives the entire network better initial weights through layer-by-layer training, so that the network can be achieved as long as it is fine-tuned Optimal solution.

2. Choose ABD.

3. Choose AB. Hyperparameters are parameters whose values ​​are set before starting the learning process, not parameter data obtained through training.
   Generally, hyperparameters can be divided into two categories:

Optimizer hyperparameters: including learning rate, minn_batch size, and number of epochs;

Model hyperparameters: including the number of network layers and hidden layer units.

4. Choose BD.
   The gradient is a vector. The objective function drops the fastest at a specific point along the opposite direction of the gradient. An image metaphor is to imagine that when you go down a mountain, you can only take one step. The fastest direction down the mountain is the opposite direction of the gradient. One step is equivalent to an iterative update of the gradient descent method.

5. Choose AB. Commonly used are the maximum pooling layer and the average pooling layer.

6. Choose A. The method of Dropout is to randomly ignore or shield some neurons in a certain proportion during the training process.

7. Choose C. D is wrong at every node.
   
   
   Wonderful review of past issues