For convolutional networks, the following statement is correct ()
A. The role of the fully connected layer is to classify. Each node of the fully connected layer is connected to all the nodes of the previous layer, and the features of the previous layer are further combined.
B. The layers in front of the convolutional neural network map the low-dimensional features of the image to the abstract feature space, and the fully connected layer maps the abstract features to the sample label space.
C. The function of the loss function is a function used to estimate the gap between the predicted value and the actual value of the model
D. The pooling layer in the convolutional neural network is also called the down sampling layer. The purpose of this layer is to reduce the dimension of the feature map, thereby reducing the number of parameters of the convolutional neural network and reducing overfitting.
Deep Convolutional Neural Networks have been widely used in ()
A. target image acquisition
B. target extraction
C. target recognition
D. target classification
The reason why deep learning uses a neural network such as a hierarchical network structure is ()
A. Visual hierarchy, in which the learning of attributes is a combination mapping that makes categories into attributes.
B. Structural (or semantic) high-level features are more meaningful for classification.
C. High-level features or images are often composed of some basic structures (shallow features).
D. Adopt a hierarchical network structure to achieve structural feature representation, and different features are represented by different structures.
The following statements about neural networks and deep learning are correct()
A. Both use a layered structure. The system includes a multi-layer network composed of an input layer, a hidden layer (multi-layer), and an output layer. Only nodes in adjacent layers are connected, and nodes in the same layer and cross-layers are not connected to each other. , Each layer can be regarded as a logistic regression model.
B. Uses the BP algorithm to adjust the parameters, that is, iterative algorithm to train the entire network. Randomly set the initial value, calculate the output of the current network, and then change the parameters of the previous layers according to the difference between the current output and the true label of the sample until convergence
C. Adopts layer-by-layer training mechanism The reason for adopting this mechanism is that if the BP mechanism is adopted, for a deep network (more than 7 layers), the residual propagation to the front most layer will become very small, and the so-called gradient diffusion occurs.
D. Its advantage lies in that it can realize complex function approximation and characterize the distributed representation of input data by learning a deep nonlinear network structure.
The reason why deep neural networks in deep learning generally do not use the BP algorithm is ()
A. During feedback adjustment, the gradient becomes more and more sparse, and the error correction signal becomes smaller and smaller from the top layer
B. Convergence is easy to the local minimum, because it is initialized with random values, it is easy to cause this situation when the initial value is far from the optimal region
C. BP algorithm requires labeled data to train, but most of the data is unlabeled
Deep learning clearly highlights the importance of feature learning. By layer-by-layer feature transformation, the feature representation of the sample in the original space is transformed into a new feature space, thereby making classification or prediction easier. Compared with the method of artificially constructing features, using big data to learn features is abler to characterize the rich internal information of the data
微信支付 
支付宝支付