[1] multiple choice can extract the image edge feature is a network (A).
A, convolution layer
B, pooled layer
C, the whole connection layer
D, output layer
Multiple choice vector [2] [0.1,0.1,0.2,0.3,0.6] is the number of dimensions (B).
A、10
B、5
C、3
D、1
[3] multiple choice (A) was used to assess the magnitude of the error between the predicted value and the true value of the neural network computational model of the sample.
A, loss of function
B, optimization function
C, back-propagation
D, gradient descent
[4] In the case of multiple choice Chapter handwritten digit recognition, the input images are the length and width of 28 pixel image, the output probability determination 0-9. To build a feed-forward neural networks to solve this problem, the input layer dimension, a dimension of the output layer. (A)
A、784;10
B、28;10
C、784;1
D、28;1
Multiple choice between [5] feedforward neural network of each layer is (C), the feedback-type neural network is (C) between the various layers.
A, there ring; the ring
B, has a ring; acyclic
C, acyclic; annulate
D, acyclic; acyclic
[6] multiple choice on MNIST, the following statement is false (C).
A, is known handwritten numeral recognition data set
B, there are training and test sets of two parts
C, with a human-like learning a variety of training set exam papers
D, test set contains about 10,000 samples and labels
7 pooling effect layer] [Multiple choice is hidden layer (A) training parameters, wherein the original signal is sampled.
A, reduce
B, increase
C, divided
D, combined
[8] multiple choice if there is a hidden layer of four layers, then the layer which is closest to the output (D).
A, convolution layer
B, pooled layer
C, the whole connection layer
D, normalized index layer
[9] a multiple-choice full of artificial neural network comprises (AC).
A, one input layer
B, Analysis of a multilayer layer
C, the multilayer hidden layer
D, two output layer
[10] multiple choice feedforward neural networks used in (AD).
A, image recognition
B, text processing
C, question answering system
D, image detection
[11] Analyzing the title of each hidden layer neural network can extract and as seen features human. (×)
[12] True or False Artificial Neural Network Training The purpose is to make the loss function is minimized. (√)
[13] True or False error back propagation, i.e. from the first hidden layer to the output layer, layer by layer to modify the parameter values of connection weights of neurons, so that the minimum loss function value. (×).
True or False [14] a major role in the hidden layer fully connected layers are fused together all features. (√)
Optimization approach [15] to determine questions gradient descent algorithm is the most common and most effective neural network, fully meet the needs of different types. (×)