Читать книгу Computational Statistics in Data Science - Группа авторов - Страница 80

LeNet‐5 is a CNN introduced by LeCun et al. [8]. This is one of the earliest structures of CNNs and was initially introduced to do handwritten digit recognition on the MNIST dataset [9]. The structure is straightforward and simple to understand, and details are shown in Figure 4.

The LeNet‐5 architecture consists of seven layers, where three are convolutional layers, two are pooling layers, and two are fully connected layers. LeNet‐5 takes images of size as input and outputs a 10‐dimensional vector as the predict scores for each class.

Figure 3 Pooling operation with stride size .

Figure 4

LeNet‐5 of LeCun et al. [8].

Source: Modified from LeCun et al. [8].

The first layer (C1) is a convolutional layer, which consists of six kernel matrices of size and stride 1. Each of the kernel matrices will scan over the input image and produce a feature matrix of size . Therefore, six different kernel matrices will produce six different feature matrices. The second layer (S2) is a Max Pooling layer, which takes the matrices as input. The kernel size of this pooling layer is , and the stride size is 2. Therefore, the outputs of this layer are six feature matrices.

Table 1 Connection between input and output matrices in the third layer of LeNet‐5 [8].

Source: LeCun et al. [8].

	Indices of output matrices
1	1	5	6	7	10	11	12	13	15	16
2	1	2	6	7	8	11	12	13	14	16
3	1	2	3	7	8	9	12	14	15	16
4	2	3	4	7	8	9	10	13	15	16
5	3	4	5	8	9	10	11	13	14	16
6	4	5	6	9	10	11	12	14	15	16

The row names are indices of input matrices, and the second column shows indices of output matrices that are connected to the corresponding input matrix. There are connections in total, meaning different kernel matrices.

The third layer (C3) is the second convolutional layer in LeNet‐5. It consists of 60 kernel matrices of size , and the stride size 1. Therefore, the output feature matrices are of size . Note that the relationship between input matrices and output matrices in this layer is not fully connected. Each of the input matrices is connected to a part of the output matrices. Details of the connection can be found in Table 1. Input matrices connected to the same output matrix will be used to produce the output matrix. Take the first output matrix, which is connected to the first three input matrices, as an example. The first three input matrices will be filtered by three different kernel matrices and result in three feature matrices. The three feature matrices will first be added together, and then a bias is added elementwise, resulting in the first output matrix. There are 16 feature matrices of size produced by layer C3.

The fourth layer (S4) is a Max Pooling layer that produces 16 feature matrices with size . The kernel size of this layer is , and the stride is 2. Therefore, each of the input matrices is reduced to . The fifth layer (C5) is the last convolutional layer in LeNet‐5. The 16 input matrices are fully connected to 120 output matrices. Since both the input matrices and kernel matrices are of size , the output matrices are of size . Therefore, the output is actually a 120‐dimensional vector. Each number in the vector is computed by applying 16 different kernel matrices on the 16 different input matrices and then combining the results and bias.

The sixth and seventh layers are fully connected layers, which are introduced in the previous section. In the sixth layer (S6), 120 input neurons are fully connected to 84 output neurons. In the last layer, 84 neurons are fully connected to 10 output neurons, where the 10‐dimensional output vector contains predict scores of each class. For the classification task, cross‐entropy loss between the model output and the label is usually used to train the model.

There are many other architectures of CNNs, such as AlexNet [10], VGG [11], and ResNet [12]. These neural networks demonstrated state‐of‐the‐art performances on many machine learning tasks, such as image classification, object detection, and speech processing.