Network risk control neural rule updating method and device based on characteristic mode consistency constraint

Abstract

The invention discloses a network risk control neural rule updating method and device based on feature mode consistency constraint, and the method comprises the steps: S1, determining a neural rule which needs to be updated, and mainly comprises a neural rule in a depth image classifier based on a convolutional neural network; s2, performing random reinforcement on the input data of the neural rule within a constrained range; s3, running a neural rule, and calculating loss based on a feature mode consistency constraint algorithm and a classification result; and S4, optimizing loss through back propagation, and realizing updating of the neural rule. According to the invention, on the premise of introducing small extra calculation overhead, the neural rule updating method with a better network risk control effect is provided for completing neural rule updating with network risk control as a target, and the security and robustness of a target network can be more effectively improved; the invention provides a neural rule updating method which is flexible and can realize network risk control in a plug-and-play mode at any use stage of the network, including an initial stage and a post-event stage.

Description

Technical Field

[0001] The invention relates to the fields of deep neural networks, network risk control and image classification, and in particular to a method and device for updating network risk control neural rules based on feature pattern consistency constraints. Background Art

[0002] Deep neural network technology: A model in computer science that consists of multiple layers of neurons, simulating the connection mode of biological nervous systems, and learning and information processing through a series of neural rules. The existing technology has constructed a variety of deep neural networks and widely used them in image recognition, speech processing, natural language processing and other fields. The image classifier based on deep neural network is the object of this invention.

[0003] Neural rules and their updates: The prior art abstracts the process of signal transmission, updating, learning, and creating input and output mappings between neurons in deep neural networks into neural rules, and regards deep neural networks as a combination of several neural rules. Some neural rules in deep neural networks are parameterized. In order to adapt the network to specific scenarios and meet requirements, the parameters in the neural rules need to be updated. The prior art calculates the loss function based on the results and goals of the network operation, and adjusts the network parameters by backpropagating the loss, thereby optimizing and updating along the gradient direction. The present invention takes network risk control as the goal and introduces new calculation terms in the loss function. The neural rules are updated by optimizing on the new loss function, and ultimately the goal of network risk control is achieved.

[0004] Features and feature patterns: The prior art believes that input samples will be abstracted into features by neural rules in deep neural networks, usually presented in the form of vectors with a certain dimension. Based on this, this patent proposes a new feature pattern concept as a supplement to the original concept, and further proposes a feature subset selection operator and a feature pattern constraint operator for updating network risk control neural rules.

[0005] Image classification and network risk control: The most widely used task for deep neural networks in the prior art is image classification, which is also the task faced by the present invention. The goal is to classify the input image. The prior art usually hopes to improve the accuracy of image classification by updating neural rules. Network risk control focuses on network security and robustness, aiming to reduce application risks. Deep neural networks are widely used in medicine, autonomous driving, security, finance and other fields with extremely high security requirements, so network risk control is crucial in these fields. However, the existing technology is difficult to balance the accuracy, security and robustness of the network. One of the goals of updating the neural rules of the present invention is to improve the network risk control capabilities and enhance the security and robustness of the network while ensuring the accuracy of the network.

[0006] The prior art has the following deficiencies:

[0007] (1) When implementing network risk control, the existing technology usually needs to introduce attack samples during the network training process so that the network can learn how to deal with attacks. However, this method is prone to overfitting, which significantly reduces the accuracy of the network on normal samples. In contrast, the method of this patent updates the neural rules by learning robust and stable feature patterns on normal samples, effectively alleviating the overfitting phenomenon.

[0008] (2) Another existing method for implementing network risk control is to filter network input to eliminate harmful information. However, this type of method has the problem of unstable results and cannot effectively deal with diverse harmful samples. The method of this patent avoids these limitations by improving the security and robustness of the network itself without relying on filtering operations.

[0009] (3) When implementing network risk control with existing technologies, it usually consumes a lot of additional computing resources. However, the method of this patent only introduces the loss calculated based on the characteristic pattern consistency algorithm when updating the neural rules, and the additional computing overhead is relatively small; and in actual use, no additional computing overhead is required.

[0010] (4) The network risk control methods in the prior art often lack flexibility. The method of this patent is highly flexible and can be introduced in a plug-and-play manner at any stage of use of the target neural network, thereby updating the neural rules in the target network and achieving the purpose of network risk control. Summary of the invention

[0011] The present invention aims to solve the problems that the existing technologies for implementing network risk control generate a large amount of additional computing overhead, the effects are not ideal, and the network risk control is not flexible enough. A method and device for updating neural rules for network risk control based on characteristic pattern consistency constraints is provided. By introducing this method at any stage of network use, the neural rules in the network are updated, thereby achieving the effect of improving network security and robustness in network risk control.

[0012] The present invention is achieved through the following technical solutions:

[0013] The first aspect of the present invention relates to a method for updating network risk control neural rules based on feature pattern consistency constraints, specifically comprising:

[0014] S1. determining the neural rules that need to be updated, including the neural rules in the deep image classifier based on the convolutional neural network;

[0015] S2, randomly strengthen the input data of the neural rule within a constrained range;

[0016] S3. Run neural rules and calculate losses based on feature pattern consistency constraint algorithm and classification results ;

[0017] S4. Optimize the loss through back-propagation to update the neural rules.

[0018] The neural rules that need to be updated in step S1 include:

[0019] S11. The neural rule is represented as a deep image classifier based on a convolutional neural network, wherein the neural rule is composed of a rule set including an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer. The input layer is used to represent the input image, and the input image is a matrix ,in , Indicates the height and width of the image, Indicates the number of channels of the image.

[0020] S12, the convolution layer extracts the features of the input image through convolution operations; the pooling layer performs dimensionality reduction on the extracted features; the fully connected layer maps the high-dimensional features after dimensionality reduction into vectors of fixed dimensions; the output layer generates the probability distribution of the classification results through the activation function.

[0021] S13. The neural rule implements the image classification task in the following manner: ensuring the correct connection of the input layer, convolution layer, pooling layer, fully connected layer and output layer in the rule set, and optimizing and updating the parameters involved in the rule through a parameter update mechanism, thereby achieving the correct execution of the image classification task.

[0022] The random reinforcement described in step S2 includes: for each sample in the input layer data , add a random noise , and obtain the enhanced sample ,use As a new training sample replacement .

[0023] The loss in step S3 By classification loss and the loss calculated based on the characteristic pattern consistency algorithm The composition is specifically expressed as

[0024] , (1)

[0025] Represents the balance factor.

[0026] Classification Loss The cross entropy between the predicted and true results obtained by running the neural rule:

[0027] , (2)

[0028] is the number of all input samples, is a logarithmic function, is the total number of categories, The natural logarithm The exponential function with base , is the probability distribution vector of the classification results output by the neural rule, and are the first and Number of digits.

[0029] Loss calculated based on the characteristic pattern consistency algorithm Depend on It is obtained by jointly calculating the feature subset selection operator and the feature pattern constraint operator.

[0030] The feature subset selection operator includes the following method: selecting a subset of features from the sample features output by some neural rules, including: for the sample In the network Features of the neural rule output of a layer before the layer In the example, a subset of features is randomly selected based on the Bernoulli distribution. ,in , is a Bernoulli distribution The binary vector extracted from . Use Alternative As input to subsequent neural rules.

[0031] like Figure 4 As shown, the characteristic mode constraint operator includes the following method: calculating the loss , which is of the form:

[0032] , (3)

[0033] , (4)

[0034] in Indicates Samples in layer network characteristic patterns; Representation sample True category The characteristic mode is The parameters learned in the neural rules of the layer network; is the characteristic mode constraint operator, and of Distance, which quantifies the difference between the characteristic pattern of the sample and the characteristic pattern of its true category; Is a layer set, indicating that only Network layer loss.

[0035] For the input sample , No. Samples in layer network The characteristic pattern The calculation is performed as follows:

[0036] , (5)

[0037] in, Representation sample After The feature of the layer network output is a vector, is the length of the vector, and Calculate using the following formula.

[0038] , (6)

[0039] , (7)

[0040] Represents the balance factor.

[0041] The specific method of optimizing the loss by back propagation and updating the neural rules as described in step S4 is as follows: by back propagating the loss value generated after running the neural rules, the parameters involved in the neural rules are updated based on the gradient descent algorithm, and the neural rules are optimized and updated in a training manner.

[0042] The second aspect of the present invention relates to a network risk control neural rule updating device based on characteristic pattern consistency constraints, comprising a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the network risk control neural rule updating method based on characteristic pattern consistency constraints of the present invention.

[0043] The working principle of the present invention is:

[0044] (1) A new loss function is constructed with the goal of network risk control to complete the neural rule update. Therefore, compared with the normal training process, the additional computing overhead introduced is relatively small;

[0045] (2) The present invention achieves the purpose of network risk control by learning robust feature patterns on normal samples, so the accuracy on normal samples is less affected. Compared with the existing technology, it can more effectively improve the security and robustness of the network.

[0046] (3) The present invention constructs a neural rule update process by inputting random reinforcement, feature subset selection operators and feature pattern constraint operators, and has a low degree of coupling with the deep neural network constructed by the prior art. Therefore, it can be introduced into various usage stages of the network in a plug-and-play manner and has good flexibility.

[0047] The innovative features of the present invention are:

[0048] (1) Based on the concept of network features, the concept of feature pattern was proposed, revealing the essence of deep neural network learning.

[0049] (2) Based on the concept of feature pattern proposed in the present invention, novel feature subset selection operators and feature pattern constraint operators are designed and used to update network risk control neural rules, which is closer to strengthening the essence of the network itself than existing methods.

[0050] (3) Compared with the existing methods, the method of the present invention has the advantages of small additional computational overhead, good effect and flexible use.

[0051] The advantages of the present invention are:

[0052] (1) The present invention can complete the neural rule update for the purpose of network risk control with less additional computing overhead;

[0053] (2) The present invention has advantages over existing methods in terms of network risk control, mainly in that it can more effectively improve the security and robustness of the network while having less impact on the accuracy of the network on normal samples;

[0054] (3) Compared with the existing methods, the present invention has better flexibility and can be used in various use stages of the network, including the initial stage before training and the post-training stage, and can be introduced in a plug-and-play manner. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a flow chart of the network risk control neural rule updating method based on feature pattern consistency constraints of the present invention.

[0056] Figure 2 This is a schematic diagram of randomly enhancing input samples in the present invention, including images of 10 categories of the CIFAR-10 image dataset.

[0057] Figure 3This is a schematic diagram of the feature subset selection operator of the present invention, where each circle represents a feature. After the feature subset operator, the feature subset selected based on the Bernoulli distribution is retained, and the white dotted circle represents the discarded feature, and the corresponding vector value is set to 0.

[0058] Figure 4 This is a schematic diagram of the feature pattern constraint operator of the present invention. The feature pattern of the sample and the feature pattern of the corresponding correct category are used to calculate the feature pattern consistency constraint loss and are finally used for neural rule updating.

[0059] Figure 5a and Figure 5b Schematic diagram of the neural rule structure of the present invention in two deep neural networks for image classification, with a dotted box used to simplify the repeated residual block neural rule structure, where Figure 5a is the WRN-28-10 deep neural network, Figure 5b It is the ResNet-50 deep neural network.

[0060] Figure 6 This is a neural rule structure diagram of VGG-16 of the present invention.

[0061] Figure 7 It is a schematic diagram of the device of the present invention. DETAILED DESCRIPTION

[0062] The technical solution of the present invention is further described below in conjunction with the accompanying drawings.

[0063] Example 1

[0064] like Figure 1 As shown, this embodiment relates to a network risk control neural rule updating method based on feature pattern consistency constraints, and the specific method is as follows:

[0065] S1. determining the neural rules that need to be updated, including the neural rules in the deep image classifier based on the convolutional neural network;

[0066] S2, randomly strengthen the input data of the neural rule within a constrained range;

[0067] S3, run the neural rule, and calculate the loss L based on the feature pattern consistency constraint algorithm and the classification result;

[0068] S4. Optimize the loss through back-propagation to update the neural rules.

[0069] The neural rules that need to be updated include:

[0070] S11. The neural rule is represented as a deep image classifier based on a convolutional neural network, wherein the neural rule is composed of a rule set including an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer. The input layer is used to represent the input image, and the input image is a matrix ,in , Indicates the height and width of the image, Indicates the number of channels of the image.

[0071] S12, the convolution layer extracts the features of the input image through convolution operations; the pooling layer performs dimensionality reduction on the extracted features; the fully connected layer maps the high-dimensional features after dimensionality reduction into vectors of fixed dimensions; the output layer generates the probability distribution of the classification results through the activation function.

[0072] S13. The neural rule implements the image classification task in the following manner: ensuring the correct connection of the input layer, convolution layer, pooling layer, fully connected layer and output layer in the rule set, and optimizing and updating the parameters involved in the rule through a parameter update mechanism, thereby achieving the correct execution of the image classification task.

[0073] like Figure 2 As shown, the random reinforcement described in step S2 includes: for each sample in the input layer data , add a random noise , and obtain the enhanced sample ,use As a new training sample replacement .

[0074] The loss in step S3 By classification loss and the loss calculated based on the characteristic pattern consistency algorithm The composition is specifically expressed as

[0075] , (1)

[0076] Represents the balance factor.

[0077] Classification Loss The cross entropy between the predicted and true results obtained by running the neural rule:

[0078] , (2)

[0079] is the number of all input samples, is a logarithmic function, is the total number of categories, The natural logarithm The exponential function with base , is the probability distribution vector of the classification results output by the neural rule, and are the first and Number of digits.

[0080] Loss calculated based on the characteristic pattern consistency algorithm Depend on It is obtained by jointly calculating the feature subset selection operator and the feature pattern constraint operator.

[0081] like Figure 3 As shown, the feature subset selection operator includes the following method: selecting a subset of features from the sample features output by some neural rules, including: for the sample In the network Features of the neural rule output of a layer before the layer In the example, a subset of features is randomly selected based on the Bernoulli distribution. ,in , is a Bernoulli distribution The binary vector extracted from . Use Alternative As input to subsequent neural rules.

[0082] like Figure 4 As shown, the characteristic mode constraint operator includes the following method: calculating the loss , which is of the form:

[0083] , (3)

[0084] , (4)

[0085] in Indicates Samples in layer network characteristic patterns; Representation sample True category The characteristic mode is The parameters learned in the neural rules of the layer network; is the characteristic mode constraint operator, and of Distance, which quantifies the difference between the characteristic pattern of the sample and the characteristic pattern of its true category; Is a layer set, indicating that only Network layer loss.

[0086] For the input sample , No. Samples in layer network The characteristic pattern The calculation is performed as follows:

[0087] , (5)

[0088] in, Representation sample After The feature of the layer network output is a vector, is the length of the vector, and Calculate using the following formula.

[0089] , (6)

[0090] , (7)

[0091] Represents the balance factor.

[0092] The specific method of optimizing the loss by back propagation and updating the neural rules as described in step S4 is as follows: by back propagating the loss value generated after running the neural rules, the parameters involved in the neural rules are updated based on the gradient descent algorithm, and the neural rules are optimized and updated in a training manner.

[0093] The overall network risk control neural rule updating method based on feature pattern consistency constraints is as follows Figure 1 shown.

[0094] The positive effects of the present invention are:

[0095] (1) Complete the neural rule update for network risk control with minimal additional computational overhead;

[0096] (2) A neural rule updating method with better network risk control effect can more effectively improve the security and robustness of the target network;

[0097] (3) Provide a flexible neural rule updating method for network risk control in a plug-and-play manner at any stage of network use, including the initial stage and the post-stage.

[0098] In order to verify the effectiveness of the present invention and the comparison between the present technology and the existing methods, the present invention uses CIFAR-10 as an image dataset for verification, which contains 10 image categories, such as Figure 2 As shown; the deep image classification networks WRN-28-10 and ResNet-50 are used as verification objects, as shown Figure 5a and Figure 5b As shown, the method of the present invention is introduced in a plug-and-play manner, PGDAT, TRADES and MARGIN are used as comparison methods, PGD, AutoPGD and EOT-PGD are used as deep attack methods, and OURS represents the patented method. The comparison results are the accuracy on normal samples and the accuracy under deep attack. The latter reflects the security and robustness of the network. The higher the accuracy, the better the network risk control effect of the corresponding method. The method of the present invention is optimal in both the accuracy of normal samples and the network risk control effect. Table 1 is a comparison of the accuracy and robustness of the method of the present invention and other methods under the CIFAR-10 dataset and deep attack.

[0099] Table 1

[0100]

[0101] Example 2

[0102] This embodiment relates to a method for enhancing the security of a medical image recognition network using the network risk control neural rule updating method based on feature pattern consistency constraints of the present invention, and the steps include:

[0103] S1. Determine the neural rules that need to be updated. In this case, the neural rules in the medical image recognition network VGG-16 are determined;

[0104] S2, randomly strengthen the input data of the neural rule within a constrained range;

[0105] S3, run the neural rule, and calculate the loss L based on the feature pattern consistency constraint algorithm and the classification result;

[0106] S4. Optimize the loss through back-propagation to update the neural rules.

[0107] S5. By updating the neural rules, the security of the medical image recognition network VGG-16 is enhanced.

[0108] The neural rules that need to be updated include: in this example, the neural rules in the medical image recognition network VGG-16 based on the convolutional neural network are determined, and the neural rule structure is as follows: Figure 6 As shown;

[0109] like Figure 2 As shown, the random reinforcement described in step S2 includes: for each sample in the input layer data , add a random noise , , and obtain the enhanced sample ,use As a new training sample replacement .

[0110] The loss in step S3 By classification loss and the loss calculated based on the characteristic pattern consistency algorithm The composition is specifically expressed as

[0111] , (1)

[0112] Represents the balance factor.

[0113] Classification Loss The cross entropy between the predicted and true results obtained by running the neural rule:

[0114] , (2)

[0115] is the number of all input samples, is a logarithmic function, is the total number of categories, The natural logarithm The exponential function with base , is the probability distribution vector of the classification results output by the neural rule, and are the first and Number of digits.

[0116] Loss calculated based on the characteristic pattern consistency algorithm Depend on It is obtained by jointly calculating the feature subset selection operator and the feature pattern constraint operator.

[0117] The loss in step S3 By classification loss and the loss calculated based on the characteristic pattern consistency algorithm The composition is specifically expressed as

[0118] , (1)

[0119] Represents the balance factor.

[0120] Classification Loss The cross entropy between the predicted and true results obtained by running the neural rule:

[0121] , (2)

[0122] is the number of all input samples, is a logarithmic function, is the total number of categories, The natural logarithm The exponential function with base , is the probability distribution vector of the classification results output by the neural rule, and are the first and Number of digits.

[0123] Loss calculated based on the characteristic pattern consistency algorithm Depend on It is obtained by jointly calculating the feature subset selection operator and the feature pattern constraint operator.

[0124] like Figure 3 As shown, the feature subset selection operator includes the following method: selecting a subset of features from the sample features output by some neural rules, including: for the sample In the network Features of the neural rule output of a layer before the layer In the example, a subset of features is randomly selected based on the Bernoulli distribution. ,in , is a Bernoulli distribution The binary vector extracted from . Use Alternative As input to subsequent neural rules.

[0125] like Figure 4 As shown, the characteristic mode constraint operator includes the following method: calculating the loss , which is of the form:

[0126] , (3)

[0127] , (4)

[0128] in Indicates Samples in layer network characteristic patterns; Representation sample True category The characteristic mode is The parameters learned in the neural rules of the layer network; is the characteristic mode constraint operator, and of Distance, which quantifies the difference between the characteristic pattern of the sample and the characteristic pattern of its true category; Is a layer set, indicating that only Network layer loss.

[0129] For the input sample , No. Samples in layer network The characteristic pattern The calculation is performed as follows:

[0130] , (5)

[0131] in, Representation sample After The feature of the layer network output is a vector, is the length of the vector, and Calculate using the following formula.

[0132] , (6)

[0133] , (7)

[0134] Represents the balance factor.

[0135] The specific method of optimizing the loss by back propagation and updating the neural rules as described in step S4 is as follows: by back propagating the loss value generated after running the neural rules, the parameters involved in the neural rules are updated based on the gradient descent algorithm, and the neural rules are optimized and updated in a training manner.

[0136] The specific method described in step S5 is as follows: After the neural rule update is completed, a more secure medical image recognition network can be obtained, which can better defend against attacks against the network.

[0137] In order to verify the effectiveness of the method for enhancing the security of a medical image recognition network using the network risk control neural rule updating method based on feature pattern consistency constraints of the present invention, PGD-White and PGD-Black were used as methods to attack the medical image recognition network, and the medical image recognition network VGG-16 was attacked before and after the security was enhanced using this method, respectively. The success rate of the attack is shown in Table 2. The lower the attack success rate, the more secure the network is. It can be seen that the method of the present invention greatly improves the security of the network. Table 2 is a security comparison of the VGG-16 network before and after the medical image recognition network security enhancement method using the network risk control neural rule updating method based on feature pattern consistency constraints of the present invention.

[0138] Table 2

[0139]

[0140] Example 3

[0141] like Figure 7 This embodiment relates to a network risk control neural rule updating device based on characteristic pattern consistency constraints, including a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, they are used to implement the network risk control neural rule updating method based on characteristic pattern consistency constraints of this embodiment 1.

[0142] The contents described in the embodiments of this specification are merely an enumeration of the implementation forms of the inventive concept. The protection scope of the present invention should not be regarded as limited to the specific forms described in the embodiments. The protection scope of the present invention also extends to equivalent technical means that can be conceived by those skilled in the art based on the inventive concept.

Claims

1. A network risk control neural rule updating method based on feature pattern consistency constraints, characterized in that: Specifically include: S1. Determine the neural rules that need to be updated, including a deep image classifier based on a convolutional neural network; S2, randomly strengthen the input data of the neural rule within a constrained range; S3. Run neural rules and calculate losses based on feature pattern consistency constraint algorithm and classification results ; S4. Optimize the loss through back-propagation to update the neural rules.

2. According to claim 1, a network risk control neural rule updating method based on characteristic pattern consistency constraints is characterized in that: The neural rules that need to be updated in step S1 include: S11. The neural rule is represented as a deep image classifier based on a convolutional neural network, wherein the neural rule is composed of a rule set including an input layer, a convolution layer, a pooling layer, a fully connected layer, and an output layer; wherein the input layer is used to represent an input image, and the input image is a matrix ,in , Indicates the height and width of the image, Indicates the number of channels of the image; S12, the convolution layer extracts the features of the input image through convolution operations; the pooling layer performs dimensionality reduction on the extracted features; the fully connected layer maps the reduced high-dimensional features into vectors of fixed dimensions; the output layer generates the probability distribution of the classification results through the activation function; S13. The neural rule implements the image classification task in the following manner: ensuring the correct connection of the input layer, convolution layer, pooling layer, fully connected layer and output layer in the rule set, and optimizing and updating the parameters involved in the rule through a parameter update mechanism, thereby achieving the correct execution of the image classification task.

3. According to claim 1, a network risk control neural rule updating method based on characteristic pattern consistency constraints is characterized in that: The random reinforcement described in step S2 includes: for each sample in the input layer data , add a random noise , get the modified feature sample ,use As a new training sample replacement .

4. According to claim 1, a network risk control neural rule updating method based on characteristic pattern consistency constraints is characterized in that: The loss in step S3 By classification loss and the loss calculated based on the characteristic pattern consistency algorithm The composition is specifically expressed as ， （1） Represents the balance factor.

5. According to claim 4, a network risk control neural rule updating method based on characteristic pattern consistency constraints is characterized in that: Classification Loss The cross entropy between the predicted and true results obtained by running the neural rule: ， （2） is the number of all input samples, is a logarithmic function, is the total number of categories, The natural logarithm The exponential function with base , is the probability distribution vector of the classification results output by the neural rule, and are the first and Number of digits.

6. According to claim 4, a network risk control neural rule updating method based on characteristic pattern consistency constraints is characterized in that: Loss calculated based on the characteristic pattern consistency algorithm Depend on It is obtained by jointly calculating the feature subset selection operator and the feature pattern constraint operator.

7. A network risk control neural rule updating method based on characteristic pattern consistency constraints according to claim 6, characterized in that: The feature subset selection operator includes the following method: selecting a subset of features from the sample features output by some neural rules, including: for the sample In the network Features of the neural rule output of a layer before the layer In the example, a subset of features is randomly selected based on the Bernoulli distribution. ,in , is a Bernoulli distribution The binary vector extracted from ; use Alternative As input to subsequent neural rules; The characteristic mode constraint operator includes the following method: calculating the loss , which is of the form: ， （3） ， （4） in Indicates Samples in layer network characteristic patterns; Representation sample True category The characteristic mode is The parameters learned in the neural rules of the layer network; is the characteristic mode constraint operator, and of Distance, which quantifies the difference between the characteristic pattern of the sample and the characteristic pattern of its true category; Is a layer set, indicating that only Network layer loss.

8. According to claim 7, a network risk control neural rule updating method based on characteristic pattern consistency constraints is characterized by: For the input sample , No. Samples in layer network The characteristic pattern The calculation is performed as follows: ， （5） in, Representation sample After The feature of the layer network output is a vector, is the length of the vector, and Calculate using the following formula; ， （6） ， （7） Represents the balance factor.

9. According to claim 1, a network risk control neural rule updating method based on characteristic pattern consistency constraints is characterized in that: The specific method of optimizing the loss by back propagation and updating the neural rules as described in step S4 is as follows: by back propagating the loss value generated after running the neural rules, the parameters involved in the neural rules are updated based on the gradient descent algorithm, and the neural rules are optimized and updated in a training manner.

10. A network risk control neural rule updating device based on characteristic pattern consistency constraints, characterized in that: It includes a memory and one or more processors, wherein the memory stores executable code, and when the one or more processors execute the executable code, it is used to implement the network risk control neural rule updating method based on characteristic pattern consistency constraints as described in any one of claims 1-9.

Images