Classification method based on residual fuzzy polynomial neural network
By introducing residual fuzzy structures and quick connections in polynomial neural networks, the problems of poor generalization ability and gradient disappearance in complex nonlinear classification are solved, and higher classification accuracy and efficiency are achieved, suitable for multiple types of classification type of data.
Patent Information
- Application Number
- CN202410814438.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-24
- Publication Date
- 2024-12-03
AI Technical Summary
Traditional polynomial neural networks have poor generalization ability and strong data uninterpretation when dealing with complex nonlinear classification problems, making it difficult to effectively deal with gradient vanishing and model degradation problems in deep neural networks.
Using the classification method based on residual fuzzy polynomial neural network, the residual network and polynomial neural network with fast connections are constructed, combined with iterative training and regularization technology, the network parameters and structure are optimized to improve the stability and generalization ability of the model. .
It significantly improves classification accuracy and efficiency, enhances the universality and flexibility of the model, can better adapt to complex classification tasks, and improves the performance of deep learning technology.
Smart Images

Figure CN119066479A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of sample classification, and in particular to a classification method based on residual fuzzy polynomial neural network. Background Art
[0002] Neural network is a computational model that simulates the human brain nervous system. It recognizes patterns and regularities in data by learning a large amount of sample data. The study of neural network as a classifier began in the 1940s, and has gone through several stages, including the proposal of formal neurons, the establishment of neuron learning criteria, the proposal of perceptual models, and the development of multi-layer perceptron back propagation algorithms. However, traditional neural network classifiers are difficult to handle complex nonlinear problems. Polynomial neural networks, because of their multi-layer neural network structure composed of multiple polynomial neurons, enhance the nonlinear expression ability of the network, enabling them to better handle complex and nonlinear classification problems. With the deepening of research, some disadvantages of polynomial neural networks have gradually emerged, such as poor generalization ability and data uninterpretability, which have become issues that cannot be ignored in improving the performance of polynomial neural network classifiers. Summary of the invention
[0003] The purpose of the present invention is to overcome the above technical shortcomings and deficiencies and provide a classification method based on residual fuzzy polynomial neural network.
[0004] The technical solution adopted by the present invention is as follows:
[0005] The classification method based on residual fuzzy polynomial neural network includes the following steps:
[0006] Step 1: Prepare the data set required to implement the classification algorithm. This data set needs to be strictly screened and preprocessed to ensure its data quality and applicability. Then, divide the data set into a training set and a test set to ensure that the training set can fully train the model, while the test set is used to verify the generalization ability of the model;
[0007] Step 2: Construct a residual network that implements quick connections. We conduct in-depth research on the structural characteristics of residual networks and design a residual block structure with efficient information transmission capabilities. This structure introduces quick connections, which enables the network to maintain good performance while increasing its depth, effectively alleviating the common problems of gradient vanishing and model degradation in deep neural networks.
[0008] Step 3: Further construct the residual polynomial neural network and initialize the network parameters. Polynomial neural networks have strong approximation capabilities and can fit complex nonlinear relationships. By combining residual networks and polynomial neural networks, a model with both depth and breadth can be constructed to better adapt to complex classification tasks. At the same time, in order to ensure the stability and convergence of the network, an appropriate initialization strategy is used to initialize the network parameters;
[0009] Step 4: Use the training data set to train the residual fuzzy polynomial neural network. Through iteration, the parameters of the network are continuously adjusted to minimize the loss function on the training set. At the same time, appropriate optimization algorithms and regularization techniques are used to prevent overfitting and improve the generalization ability of the model.
[0010] Step 5: Evaluate the performance of the trained polynomial neural network classifier on the test set. By calculating the accuracy, recall, F1 value and other indicators on the test set, the classification performance of the model can be comprehensively evaluated.
[0011] Step 1: Prepare the data set required to implement the classification algorithm and divide the data set into a training set and a test set. The specific steps are as follows:
[0012] Step 1.1: Select 6 international standard classification datasets from the UCI machine learning database for model training and testing;
[0013] Step 1.2: Use PCA algorithm to reduce the dimension of the selected data set;
[0014] Step 1.2.1: Standardize the original data set and scale each feature to a range of mean 0 and standard deviation 1.
[0015] Step 1.2.2: Calculate the covariance matrix of all features in the dataset
[0016] Step 1.2.3: Find the eigenvalues and eigenvectors of the covariance matrix C;
[0017] Step 1.2.4: According to the size of the eigenvalue, select the eigenvectors corresponding to the first k largest eigenvalues;
[0018] Step 1.2.5: Use the selected k feature vectors to transform each sample in the original dataset into the new feature space;
[0019] Step 1.3: Divide the preprocessed data into training set and test set in a ratio of 6 to 4.
[0020] Step 2: Construct a residual network that implements quick connections and design a residual block structure. The specific steps are as follows:
[0021] Step 2.1: Design the residual block structure. The residual block is the core component of the residual network. The specific steps are as follows:
[0022] Step 2.1.1: Determine the number of network layers in each residual block and design a shortcut connection;
[0023] Step 2.1.2: Design the role and function implementation process of the residual block. The network layer is used to extract features, while the shortcut connection allows information to be directly transmitted in the network.
[0024] Step 2.1.3: Design the specific implementation of the residual connection. Assume that the input of the residual block is x, and the output after a series of network mapping operations is F(x). Add x and F(x) as the input of the next layer, and get H(x) = F(x) + x.
[0025] Step 2.1.4: Select the activation function (such as ReLU) that needs to be used after the network layer in the residual block;
[0026] Step 2.1.5: Deal with the dimension mismatch problem. If the input and output dimensions of the residual block do not match, a linear transformation needs to be added to the cross-layer connection.
[0027] Step 2.2: stack the designed residual blocks in a specific order to form a complete residual network;
[0028] Step 2.3: Initialize weights and train the network. The parameters are initialized using random values or pre-trained weights.
[0029] Step 3: Construct a residual polynomial neural network and initialize the network parameters. The specific steps are as follows:
[0030] Step 3.1: Construct a polynomial neural network. The specific steps are as follows:
[0031] Step 3.1.1: Determine the number of basic layers of the network and the number of neurons in each layer;
[0032] Step 3.1.2, determine the number of input variables and the order of the polynomial;
[0033] Step 3.1.3: Select the type of polynomial, which generally includes the following four types:
[0034] Type 1: Constant
[0035]
[0036] Type 2: Linear Polynomial
[0037] f j =a j0 +aj1 x1+...+a jk x n
[0038] Type 3: Quadratic Polynomial
[0039] f j =a j0 +a j1 x1+...+a jn x n +a j(n+1) x1 2 +...+a j(2n) x n 2
[0040] +a j(2n+1) x1x2+...+a j((n+2)(n+1) / 2) x n-1 x n
[0041] Type 4: Optimized quadratic polynomial
[0042] f j =a j0 +a j1 x1+...+a jn x n +a j(n+1) x1x2+a j(n(n+1) / 2) x n-1 x n
[0043] Step 3.1.3: Evaluate the correlation coefficient C of the polynomial neurons using the least squares method i , the calculation formula is as follows:
[0044] C i =(X i T X i ) -1 X i T Y
[0045] Step 3.1.4: Select the neuron with the best prediction ability;
[0046] Step 3.1.5: Check whether the termination conditions are met;
[0047] Step 3.1.6: Determine the new input variables for the next layer (if the termination condition is not met, select the outputs of several best neurons in the current layer as the inputs of the neurons in the next layer);
[0048] Step 3.1.7: Repeat steps 3.1.2-3.1.6 until the termination condition is met to construct the polynomial neural network;
[0049] Step 3.2: Change the network structure based on the polynomial network and build residual connections. The specific steps are as follows:
[0050] Step 3.2.1: Each two polynomial network layers and a cross-layer connection are used as a basic residual block. Nonlinear transformation is performed on the input data in the network layer, and the cross-layer connection realizes the cross-layer flow of data.
[0051] Step 3.2.2: Establish a feedback mechanism in the network and introduce identity mapping;
[0052] Step 3.3: Calculate the category membership of the data output by the network through the fuzzy C-means clustering algorithm. The specific steps are as follows:
[0053] Step 3.3.1, select the number of categories k, the appropriate m, and initialize the matrix U0 determined by the membership function;
[0054] Step 3.3.2: Calculate the cluster center value C i
[0055]
[0056] Step 3.3.3: Use u according to the obtained cluster center ij Formula Update Membership Value
[0057]
[0058] Step 3.3.4: Calculate the objective function J m The value of (u, c) is denoted as J (t) (t means this cycle)
[0059]
[0060] Step 3.3.5: If the value of this loop minus the value of the previous loop is less than the given value, then ‖J (t) -J (t-1) ‖<∈ then the loop ends and the final value is obtained, otherwise repeat steps 3.3.2-3.3.5;
[0061] Step 3.4: Predict the category to which the input data belongs based on the final membership value.
[0062] Step 4: Use the training data set to train the residual fuzzy polynomial neural network. The specific steps are as follows:
[0063] Step 4.1: Iteratively train the model using the training dataset. In each iteration, input the training data into the model and calculate the loss between the output and the true label;
[0064] Step 4.2: Calculate the gradient through the back-propagation algorithm and use the optimization algorithm to update the parameters of the model.
[0065] Step 5: Evaluate the performance of the trained polynomial neural network classifier on the test set. The specific steps are as follows:
[0066] Step 5.1: According to the determined polynomial structure, the test sample set is used as input, and the label value of each test sample belonging to each category is obtained through the model;
[0067] Step 5.2: According to the output label values, determine the category to which the corresponding test sample belongs;
[0068] Step 5.3: Output the classification results and calculate the model's accuracy, recall and other indicators to comprehensively evaluate the model's performance.
[0069] Advantages and beneficial effects of the present invention
[0070] The present invention proposes a classification method based on residual fuzzy polynomial neural network, which is an innovative improvement on polynomial neural network. This neural network combines the advantages of the "cross-layer connection" structure in the residual neural network with the powerful high-order nonlinear expression ability of the polynomial neural network, so that the designed residual polynomial neural network can fit the classification function more accurately, thereby achieving higher classification accuracy. This method not only has a fast convergence speed and can significantly improve the efficiency of classification tasks, but also has excellent versatility and flexibility, and can easily cope with various classification scenarios. Whether it is images, texts or other types of data, this method can show excellent classification performance, injecting new vitality into the development of deep learning technology. BRIEF DESCRIPTION OF THE DRAWINGS
[0071] Figure 1 It is a workflow diagram of solving classification problems of the present invention;
[0072] Figure 2 It is a network structure diagram of a polynomial neural network;
[0073] Figure 3 It is a residual unit structure diagram;
[0074] Figure 4 It is a network structure diagram of a residual polynomial neural network. DETAILED DESCRIPTION
[0075] The present invention is described in detail below with reference to the accompanying drawings and examples.
[0076] A classification method based on residual fuzzy polynomial neural network is proposed. The six data sets of Zoo, Wine, Liver, Iris, Balance and Vehicle in the international standard data set are trained and tested. The Iris data set is used as an example in the following description. The operations of the other five data sets are similar to those of the Iris data set. The specific implementation plan includes the following contents and steps:
[0077] Step 1: Prepare the data set required to implement the classification algorithm and divide the data set into training set and test set. The specific steps are as follows:
[0078] Step 1.1: Obtain the Iris dataset from the UCI machine learning database, and divide the model into training and test sets based on the Iris dataset with 150 records and 4 features;
[0079] Step 1.2: Use the PCA algorithm to reduce the dimension of the selected Iris data set to k=2 (2-dimensional);
[0080] Step 1.2.1: Standardize the original data set and scale each feature to a range of mean 0 and standard deviation 1.
[0081] Step 1.2.2: Calculate the covariance matrix of all features in the dataset
[0082] Step 1.2.3: Find the eigenvalues and eigenvectors of the covariance matrix C. First, use the eigenvalue decomposition method to solve the eigenvalues of the covariance matrix C, and then find the eigenvalues and eigenvectors respectively;
[0083] Step 1.2.4: According to λ1-λ n Arrange according to the size of the eigenvalues, select the eigenvectors corresponding to the first k largest eigenvalues, and form the eigenvector matrix P′;
[0084] Step 1.2.5: Use the selected k feature vectors to transform each sample in the original data set into a new feature space, i.e., Y = P′X;
[0085] Step 1.3: Divide the preprocessed data into a training set and a test set in a ratio of 6 to 4, that is, 90 data are used as the training data set and 60 data are used as the test data set.
[0086] Step 2: Construct a residual network that implements quick connections and design a residual block structure. The specific steps are as follows:
[0087] Step 2.1: Design the residual block structure. The residual block is the core component of the residual network. The specific steps are as follows:
[0088] Step 2.1.1: Determine that each residual block contains two network layers and a cross-layer connection spanning two layers;
[0089] Step 2.1.2: Design the role and function implementation process of the residual block. The network layer is used to extract features, while the shortcut connection allows information to be directly transmitted in the network.
[0090] Step 2.1.3: Design the specific implementation of the residual connection. Assume that the input of the residual block is x, and the output after a series of network mapping operations is F(x). Add x and F(x) as the input of the next layer, and get H(x) = F(x) + x.
[0091] Step 2.1.4: Select the activation function ReLU that needs to be used after the network layer in the residual block;
[0092] Step 2.1.5: Deal with the dimension mismatch problem. If the input and output dimensions of the residual block do not match, a linear transformation needs to be added to the cross-layer connection.
[0093] Step 2.2: stack the designed residual blocks in a specific order to form a complete residual network;
[0094] Step 2.3: Initialize weights and train the network. The parameters are initialized using random values or pre-trained weights.
[0095] Step 3: Construct a residual polynomial neural network and initialize the network parameters. The specific steps are as follows:
[0096] Step 3.1: Construct a polynomial neural network. The specific steps are as follows:
[0097] Step 3.1.1: Determine the basic number of layers of the network (greater than 1 layer) and the number of neurons in each layer (the maximum number of neuron nodes in each layer is set to 20);
[0098] Step 3.1.2, determine the number of neuron input variables (2-4) and the order of the polynomial (0, 1, 2);
[0099] Step 3.1.3: Select the type of polynomial, which generally includes the following four types:
[0100] Type 1: Constant
[0101] f j =a j0
[0102] Type 2: Linear Polynomial
[0103] f j =a j0 +a j1x1+...+a jk x n
[0104] Type 3: Quadratic Polynomial
[0105] f j =a j0 +a j1 x1+...+a jn x n +a j(n+1) x1 2 +...+a j(2n) x n 2
[0106] +a j(2n+1) x1x2+...+a j((n+2)(n+1) / 2) x n-1 x n
[0107] Type 4: Optimized quadratic polynomial
[0108] f j =a j0 +a j1 x1+...+a jn x n +a j(n+1) x1x2+a j(n(n+1) / 2) x n-1 x n
[0109] Step 3.1.3: Evaluate the correlation coefficient C of the polynomial neurons using the least squares method i , the calculation formula is as follows:
[0110] C i =(X i T X i ) -1 X i T Y
[0111] Step 3.1.4: Select the neuron with the best prediction ability;
[0112] Step 3.1.5: Check whether the termination condition is met (based on performance indicators or number of iterations);
[0113] Step 3.1.6: Determine the new input variables for the next layer (if the termination condition is not met, select the outputs of several best neurons in the current layer as the inputs of the neurons in the next layer);
[0114] Step 3.1.7: Repeat steps 3.1.2-3.1.6 until the termination condition is met to construct the polynomial neural network;
[0115] Step 3.2: Change the network structure based on the polynomial network and build residual connections. The specific steps are as follows:
[0116] Step 3.2.1: Each two polynomial network layers and a cross-layer connection are used as a basic residual block. Nonlinear transformation is performed on the input data in the network layer, and the cross-layer connection realizes the cross-layer flow of data.
[0117] Step 3.2.2: Establish a feedback mechanism in the network and introduce identity mapping;
[0118] Step 3.3: Calculate the category membership of the data output by the network through the fuzzy C-means clustering algorithm. The specific steps are as follows:
[0119] Step 3.3.1, select the number of categories k = 3, the appropriate m = 2, and initialize the matrix U0 determined by the membership function;
[0120] Step 3.3.2: Calculate the cluster center value C i
[0121]
[0122] Step 3.3.3: Use u according to the obtained cluster center ij Formula Update Membership Value
[0123]
[0124] Step 3.3.4: Calculate the objective function J m The value of (u, c) is denoted as J (t) (t means this cycle)
[0125]
[0126] Step 3.3.5: If the value of this loop minus the value of the previous loop is less than the given value, then ‖J (t) -J (t-1) ‖<∈ then the loop ends and the final value is obtained, otherwise repeat steps 3.3.2-3.3.5;
[0127] Step 3.4: Predict the category to which the input data belongs based on the final membership matrix.
[0128] Step 4: Use the training data set to train the residual fuzzy polynomial neural network. The specific steps are as follows:
[0129] Step 4.1: Iteratively train the model using the training dataset. In each iteration, input the training data into the model and calculate the loss between the output and the true label;
[0130] Step 4.2: Calculate the gradient through the back-propagation algorithm and use the optimization algorithm to update the parameters of the model.
[0131] Step 5: Evaluate the performance of the trained polynomial neural network classifier on the test set. The specific steps are as follows:
[0132] Step 5.1: According to the determined polynomial structure, the test sample set is used as input, and the label value of each test sample belonging to each category is obtained through the model;
[0133] Step 5.2: According to the output label values, determine the category to which the corresponding test sample belongs;
[0134] Step 5.3: Output the classification results and calculate the model's accuracy, recall and other indicators to comprehensively evaluate the model's performance.
[0135] The above description is only a preferred example of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of the claims of the present invention.
Claims
1. A classification method based on residual fuzzy polynomial neural network, Its characteristics are distinct, It is specifically reflected in the following steps: Step 1: Prepare the data set required to implement the classification algorithm. This data set needs to be strictly screened and preprocessed to ensure its data quality and applicability. Then, divide the data set into a training set and a test set to ensure that the training set can fully train the model, while the test set is used to verify the generalization ability of the model; Step 2: Construct a residual network that implements quick connections. We conduct in-depth research on the structural characteristics of residual networks and design a residual block structure with efficient information transmission capabilities. This structure introduces quick connections, which enables the network to maintain good performance while increasing its depth, effectively alleviating the common problems of gradient vanishing and model degradation in deep neural networks. Step 3: Further construct the residual polynomial neural network and initialize the network parameters. Polynomial neural networks have powerful approximation capabilities and can fit complex nonlinear relationships. By combining residual networks and polynomial neural networks, a model with both depth and breadth can be constructed to better adapt to complex classification tasks. At the same time, in order to ensure the stability and convergence of the network, an appropriate initialization strategy is used to initialize the network parameters; Step 4: Use the training data set to train the residual fuzzy polynomial neural network. Through iteration, the parameters of the network are continuously adjusted to minimize the loss function on the training set. At the same time, appropriate optimization algorithms and regularization techniques are used to prevent overfitting and improve the generalization ability of the model. Step 5: Evaluate the performance of the trained polynomial neural network classifier on the test set. By calculating the accuracy, recall, F1 value and other indicators on the test set, the classification performance of the model can be comprehensively evaluated.
2. The classification method based on residual fuzzy polynomial neural network according to claim 1, Features: Step 1 describes the preparation of the data set required for the classification algorithm and the division of the data set into a training set and a test set. The specific steps are as follows: Step 1.1: Select 6 international standard classification datasets from the UCI machine learning database for model training and testing; Step 1.2: Use PCA algorithm to reduce the dimension of the selected data set; Step 1.2.1: Standardize the original data set and scale each feature to a range of mean 0 and standard deviation 1. Step 1.2.2: Calculate the covariance matrix of all features in the dataset Step 1.2.3: Find the eigenvalues and eigenvectors of the covariance matrix C; Step 1.2.4: According to the size of the eigenvalue, select the eigenvectors corresponding to the first k largest eigenvalues; Step 1.2.5: Use the selected k feature vectors to transform each sample in the original dataset into the new feature space; Step 1.3: Divide the preprocessed data into training set and test set in a ratio of 6 to 4.
3. The classification method based on residual fuzzy polynomial neural network according to claim 1, Features: The construction described in step 2 realizes the residual network of quick connection and designs the residual block structure. The specific steps are as follows: Step 2.1: Design the residual block structure. The residual block is the core component of the residual network. The specific steps are as follows: Step 2.1.1: Determine the number of network layers in each residual block and design a shortcut connection; Step 2.1.2: Design the role and function implementation process of the residual block. The network layer is used to extract features, while the shortcut connection allows information to be directly transmitted in the network. Step 2.1.3: Design the specific implementation of the residual connection. Assume that the input of the residual block is x, and the output after a series of network mapping operations is F(x). Add x and F(x) as the input of the next layer, and get H(x) = F(x) + x. Step 2.1.4: Select the activation function (such as ReLU) that needs to be used after the network layer in the residual block; Step 2.1.5: Deal with the dimension mismatch problem. If the input and output dimensions of the residual block do not match, a linear transformation needs to be added to the cross-layer connection. Step 2.2: stack the designed residual blocks in a specific order to form a complete residual network; Step 2.3: Initialize weights and train the network. The parameters are initialized using random values or pre-trained weights.
4. The classification method based on residual fuzzy polynomial neural network according to claim 1, Features: The residual polynomial neural network is constructed as described in step 3, and the parameters of the network are initialized. The specific steps are as follows: Step 3.1: Construct a polynomial neural network. The specific steps are as follows: Step 3.1.1: Determine the number of basic layers of the network and the number of neurons in each layer; Step 3.1.2, determine the number of input variables and the order of the polynomial; Step 3.1.3: Select the type of polynomial, which generally includes the following four types: Type 1: Constant f j =a j0 Type 2: Linear Polynomial f j =a j0 +a j1 x 1 +...+a jk x n Type 3: Quadratic Polynomial f j =a j0 +a j1 x 1 +...+a jn x n +a j(n+1) x 1 2 +...+a j(2n) x n 2 +a j(2n+1) x 1 x 2 +...+a j((n+2)(n+1) / 2) x n-1 x n Type 4: Optimized quadratic polynomial f j =a j0 +a j1 x 1 +...+a jn x n +a j(n+1) x 1 x 2 +a j(n(n+1) / 2) x n-1 x n Step 3.1.3: Evaluate the correlation coefficient C of the polynomial neurons using the least squares method i , the calculation formula is as follows: C i =(X i T X i ) -1 X i T Y Step 3.1.4: Select the neuron with the best prediction ability; Step 3.1.5: Check whether the termination conditions are met; Step 3.1.6: Determine the new input variables for the next layer (if the termination condition is not met, select the outputs of several best neurons in the current layer as the inputs of the neurons in the next layer); Step 3.1.7: Repeat steps 3.1.2-3.1.6 until the termination condition is met to construct the polynomial neural network; Step 3.2: Change the network structure based on the polynomial network and build residual connections. The specific steps are as follows: Step 3.2.1: Each two polynomial network layers and a cross-layer connection are used as a basic residual block. Nonlinear transformation is performed on the input data in the network layer, and the cross-layer connection realizes the cross-layer flow of data. Step 3.2.2: Establish a feedback mechanism in the network and introduce identity mapping; Step 3.3: Calculate the category membership of the data output by the network through the fuzzy C-means clustering algorithm. The specific steps are as follows: Step 3.3.1: Select the number of categories k, the appropriate m, and initialize the matrix U determined by the membership function 0 ; Step 3.3.2: Calculate the cluster center value C i Step 3.3.3: Use u according to the obtained cluster center ij Formula Update Membership Value Step 3.3.4: Calculate the objective function Jm ( u, c) is recorded as J (t) (t means this cycle) Step 3.3.5: If the value of this loop minus the value of the previous loop is less than the given value, that is, ||J (t) -J (t-1 )||<∈Then, the loop ends and the final value is obtained, otherwise repeat steps 3.3.2-3.3.5; Step 3.4: Predict the category to which the input data belongs based on the final membership value.
5. The classification method based on residual fuzzy polynomial neural network according to claim 1, Features: The specific steps of using the training data set to train the residual fuzzy polynomial neural network in step 4 are as follows: Step 4.1: Iteratively train the model using the training dataset. In each iteration, input the training data into the model and calculate the loss between the output and the true label; Step 4.2: Calculate the gradient through the back-propagation algorithm and use the optimization algorithm to update the parameters of the model.
6. The classification method based on residual fuzzy polynomial neural network according to claim 1, Features: Evaluate the performance of the trained polynomial neural network classifier on the test set as described in step 5. The specific steps are as follows: Step 5.1: According to the determined polynomial structure, the test sample set is used as input, and the label value of each test sample belonging to each category is obtained through the model; Step 5.2: According to the output label values, determine the category to which the corresponding test sample belongs; Step 5.3: Output the classification results and calculate the model's accuracy, recall and other indicators to comprehensively evaluate the model's performance.
Citation Information
Patent Citations
DTCO modeling method based on residual network
CN115481570A
Classification method based on random resampling polynomial neural network
CN117520944A
Transformer state evaluation method based on echo state network and deep residual neural network
US20230112749A1
