CN116167833B - Internet financial risk control system and method based on federal learning - Google Patents

Internet financial risk control system and method based on federal learning Download PDF

Info

Publication number
CN116167833B
CN116167833B CN202310419480.4A CN202310419480A CN116167833B CN 116167833 B CN116167833 B CN 116167833B CN 202310419480 A CN202310419480 A CN 202310419480A CN 116167833 B CN116167833 B CN 116167833B
Authority
CN
China
Prior art keywords
weight
feature
training
matrix
financial risk
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202310419480.4A
Other languages
Chinese (zh)
Other versions
CN116167833A (en
Inventor
胡正
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Wuxi Several Letter Mutual Melt Science And Technology Development Co ltd
Original Assignee
Wuxi Several Letter Mutual Melt Science And Technology Development Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Wuxi Several Letter Mutual Melt Science And Technology Development Co ltd filed Critical Wuxi Several Letter Mutual Melt Science And Technology Development Co ltd
Priority to CN202310419480.4A priority Critical patent/CN116167833B/en
Publication of CN116167833A publication Critical patent/CN116167833A/en
Application granted granted Critical
Publication of CN116167833B publication Critical patent/CN116167833B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Abstract

The application relates to the field of intelligent control, and particularly discloses an internet financial risk control system and method based on federal learning, which are used for mining implicit association characteristic information among weight matrixes of all internet financial risk assessment models by adopting a neural network model based on deep learning so as to fuse all internet financial risk assessment models and improve the assessment accuracy of the fused internet financial risk assessment models.

Description

Internet financial risk control system and method based on federal learning
Technical Field
The present application relates to the field of intelligent control, and more particularly, to an internet financial risk control system based on federal learning and a method thereof.
Background
At present, internet finance develops rapidly and is in a straight line rising trend. However, there are a number of risks associated with internet finance, including but not limited to: credit risk, unknown legal positioning, mobility risk, higher supervision difficulty, information leakage, technical risk and the like. In recent years, the development of technologies such as big data, artificial intelligence and the like provides a new solution and thinking for financial risk control.
In particular, an internet financial risk assessment model based on a deep neural network can be constructed. It should be appreciated that the performance (performance) of the internet financial risk assessment model is largely dependent on training data and training strategies. However, in the real world, data of each financial service node cannot be exported, which causes problems such as information leakage and information security. That is, when training the internet financial risk assessment model based on the deep neural network, there are problems that training data is insufficient and diversity of the training data is insufficient.
Disclosure of Invention
The present application has been made in order to solve the above technical problems. The embodiment of the application provides an internet financial risk control system and method based on federal learning, which are used for mining implicit association characteristic information among weight matrixes of all internet financial risk assessment models by adopting a neural network model based on deep learning so as to fuse all internet financial risk assessment models and improve the assessment accuracy of the fused internet financial risk assessment models.
According to one aspect of the present application, there is provided an internet financial risk control system based on federal learning, comprising: the data acquisition module is used for acquiring intra-domain training data provided by each financial service node, wherein the intra-domain training data comprises pedestrian credit data, three-party credit data, judicial information, industrial and commercial information, tax information, invoice information, public accumulation information, public transaction information and transaction information; the internet financial risk assessment model training module is used for respectively training the internet financial risk assessment model with a plurality of weight matrixes by taking the intra-domain training data provided by each financial service node as training data so as to obtain a plurality of trained internet financial risk assessment models; the channel weight calculation module is used for calculating global average values of a plurality of weight matrixes of the trained internet financial risk assessment models respectively to obtain channel weight feature vectors composed of a plurality of global average values so as to obtain a plurality of channel weight feature vectors; the weight global association module is used for arranging the channel weight feature vectors into a two-dimensional feature matrix and then obtaining a global weight feature matrix through a convolutional neural network model serving as a feature extractor; the query module is used for taking the weight feature vectors of the channels as query feature vectors, and calculating the matrix product between the query feature vectors and the global weight feature matrix to obtain a plurality of classification feature vectors; the weight probability value calculation module is used for enabling the classification feature vectors to pass through a classifier to obtain a plurality of probability values, and normalizing the probability values so that the sum of the probability values is 1; and the model integration module is used for fusing the plurality of trained internet financial risk assessment models by taking the plurality of probability values as weights.
In the internet financial risk control system based on federal learning, the weight global association module is configured to: each layer of the convolutional neural network model using the feature extractor performs, in forward transfer of the layer, input data: carrying out convolution processing on input data to obtain a convolution characteristic diagram; pooling the convolution feature map along a channel dimension to obtain a pooled feature map; performing nonlinear activation on the pooled feature map to obtain an activated feature map; the output of the last layer of the convolutional neural network serving as the feature extractor is the global weight feature matrix, and the input of the first layer of the convolutional neural network serving as the feature extractor is the two-dimensional feature matrix of the arrangement of the plurality of channel weight feature vectors.
In the above internet financial risk control system based on federal learning, the query module is configured to: calculating a matrix product between each channel weight feature vector and the global weight feature matrix to obtain a plurality of classification feature vectors according to the following formula; wherein, the formula is:
Figure SMS_1
wherein->
Figure SMS_2
Representing the weight feature vector of each channel, +. >
Figure SMS_3
Representing the global weight feature matrix, +.>
Figure SMS_4
Representing the plurality of classification feature vectors, +.>
Figure SMS_5
Representing a matrix multiplication.
In the above internet financial risk control system based on federal learning, the weight probability value calculation module is configured to: processing the plurality of classification feature vectors using the classifier to obtain a plurality of probability values in a classification formula, wherein the formula is:
Figure SMS_6
wherein->
Figure SMS_7
To the point of
Figure SMS_8
Is a weight matrix>
Figure SMS_9
To->
Figure SMS_10
For the bias vector +.>
Figure SMS_11
For a plurality of classification feature vectors, < >>
Figure SMS_12
Representing the plurality of probability values.
The internet financial risk control system based on federal learning further comprises an optimization training module for training the convolutional neural network model serving as the feature extractor and the classifier.
In the above internet financial risk control system based on federal learning, the optimization training module includes: the training data acquisition module is used for acquiring optimized training data, wherein the optimized training data comprises a plurality of training weight matrixes of the internet financial risk assessment models with which the training is completed and the true values of the probability values; the training channel weight calculation module is used for calculating training global average values of a plurality of training weight matrixes of the internet financial risk assessment models after the training is completed respectively to obtain training channel weight feature vectors composed of the training global average values so as to obtain a plurality of training channel weight feature vectors; the training weight global association module is used for arranging the training channel weight feature vectors into a training two-dimensional feature matrix and obtaining a training global weight feature matrix through the convolutional neural network model serving as the feature extractor; the training query module is used for taking the weight feature vectors of the training channels as query feature vectors, and calculating the matrix product between the weight feature vectors and the training global weight feature matrix to obtain a plurality of training classification feature vectors; the classification loss module is used for enabling the training classification feature vectors to pass through the classifier to obtain a plurality of classification loss function values; and the optimization training module is used for training the convolutional neural network model serving as the feature extractor and the classifier based on the plurality of classification loss function values and through gradient descent direction propagation, wherein in each round of iteration of training, the spatial regularization constraint iteration of the weight matrix of the classifier is carried out.
In the above internet financial risk control system based on federal learning, the classification loss module includes: performing full-connection coding on the training classification feature vectors by using a plurality of full-connection layers of the classifier to obtain coded classification feature vectors; and passing the encoded classification feature vector through the classifier
Figure SMS_13
Classifying the function to obtain the plurality of classification loss function values.
In the internet financial risk control system based on federal learning, in each iteration of the training, performing spatial regularization constraint iteration of a weight matrix of the classifier according to the following optimization formula; wherein, the optimization formula is:
Figure SMS_14
wherein->
Figure SMS_19
Is a weight matrix of the classifier, +.>
Figure SMS_21
Frobenius norms of the matrix are represented, < >>
Figure SMS_15
Transpose of the weight matrix representing the classifier, < >>
Figure SMS_17
Is a bias matrix, +.>
Figure SMS_22
Representing matrix multiplication +.>
Figure SMS_23
Representing matrix addition, ++>
Figure SMS_16
Representing multiplication by location +.>
Figure SMS_18
An exponential operation representing a matrix representing the calculation of a natural exponential function value raised to a power by a characteristic value at each position in the matrix,/v>
Figure SMS_20
And representing the weight matrix of the classifier after iteration.
According to another aspect of the present application, there is provided an internet financial risk control method based on federal learning, including: acquiring intra-domain training data provided by each financial service node, wherein the intra-domain training data comprises pedestrian credit data, three-party credit data, judicial information, industrial and commercial information, tax information, invoice information, public deposit information, public transaction information and transaction information; training the internet financial risk assessment model with a plurality of weight matrixes by taking intra-domain training data provided by each financial service node as training data to obtain a plurality of trained internet financial risk assessment models; respectively calculating global average values of a plurality of weight matrixes of the trained internet financial risk assessment models to obtain channel weight feature vectors composed of the global average values so as to obtain a plurality of channel weight feature vectors; the channel weight feature vectors are arranged into a two-dimensional feature matrix and then a global weight feature matrix is obtained through a convolutional neural network model serving as a feature extractor; taking each channel weight feature vector as a query feature vector, and calculating a matrix product between the query feature vector and the global weight feature matrix to obtain a plurality of classification feature vectors; passing the plurality of classification feature vectors through a classifier to obtain a plurality of probability values, and normalizing the plurality of probability values so that the sum of the plurality of probability values is 1; and fusing the plurality of trained internet financial risk assessment models with the plurality of probability values as weights.
According to still another aspect of the present application, there is provided an electronic apparatus including: a processor; and a memory having stored therein computer program instructions that, when executed by the processor, cause the processor to perform the federal learning-based internet financial risk control method as described above.
According to yet another aspect of the present application, there is provided a computer readable medium having stored thereon computer program instructions which, when executed by a processor, cause the processor to perform the federally learning based internet financial risk control method as described above.
Compared with the prior art, the internet financial risk control system and the internet financial risk control method based on federal learning provided by the application have the advantages that the implicit association characteristic information among the weight matrixes of all internet financial risk assessment models is mined by adopting the neural network model based on deep learning, so that all internet financial risk assessment models are fused, and the assessment accuracy of the fused internet financial risk assessment models is improved.
Drawings
The foregoing and other objects, features and advantages of the present application will become more apparent from the following more particular description of embodiments of the present application, as illustrated in the accompanying drawings. The accompanying drawings are included to provide a further understanding of embodiments of the application and are incorporated in and constitute a part of this specification, illustrate the application and not constitute a limitation to the application. In the drawings, like reference numerals generally refer to like parts or steps.
FIG. 1 is a block diagram of an inference module in an Internet financial risk control system based on federal learning according to an embodiment of the present application.
Fig. 2 is a block diagram of training modules in an internet financial risk control system based on federal learning according to an embodiment of the present application.
Fig. 3 is a system architecture diagram of an inference module in an internet financial risk control system based on federal learning according to an embodiment of the present application.
Fig. 4 is a system architecture diagram of a training module in an internet financial risk control system based on federal learning according to an embodiment of the present application.
Fig. 5 is a flowchart of convolutional neural network coding in an internet financial risk control system based on federal learning according to an embodiment of the present application.
Fig. 6 is a flowchart of an internet financial risk control method based on federal learning according to an embodiment of the present application.
Fig. 7 is a block diagram of an electronic device according to an embodiment of the present application.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only some of the embodiments of the present application and not all of the embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein.
Scene overview: according to the technical scheme, the internet financial risk control model is trained based on the data of each node through federal learning, and then parameters of each model are integrated, and in such a way, a global model based on virtual fusion data is constructed, so that balance of data privacy protection and data sharing calculation is achieved.
Accordingly, considering that the common fusion strategy is to perform weighted fusion on the weight matrix of each trained internet financial risk assessment model in the process of actually performing the integration of each model parameter, however, since the trained internet financial risk assessment model is obtained through intra-domain data training of each financial service node, performance of the data in other domains may be relatively poor, that is, contribution degree of each internet financial risk assessment model degree to the final internet financial risk assessment model based on the global data is different. Accordingly, in the present embodiment, the following is expected
And the correlation among the weight matrixes of the trained internet financial risk assessment models is used for integrating model parameters, so that the generalization capability of the final fusion model is improved. Therefore, in this process, it is difficult to mine the implicit association feature information between the weight matrices of the internet financial risk assessment models, so as to fuse the internet financial risk assessment models, so as to improve the assessment accuracy of the fused internet financial risk assessment models.
Specifically, in the technical scheme of the application, first, intra-domain training data provided by each financial service node is obtained, wherein the intra-domain training data comprises pedestrian credit data, three-party credit data, judicial information, industrial and commercial information, tax information, invoice information, public deposit information, public transaction information and transaction information. Then, training of the internet financial risk control model is performed based on the data of each node through federal learning, that is, specifically, the internet financial risk assessment model with a plurality of weight matrices is trained by taking the intra-domain training data provided by each financial service node as training data, so as to obtain a plurality of trained internet financial risk assessment models.
Then, for each trained internet financial risk assessment model, the weight matrixes of the trained internet financial risk assessment models have a correlative relation, that is, each trained internet financial risk assessment model is interrelated, and the trained internet financial risk assessment models form a risk assessment model whole. Therefore, if the trained internet financial risk assessment models are to be better integrated, so as to optimize the risk assessment accuracy of the models, the correlation characteristics between the weight matrixes of the trained internet financial risk assessment models need to be fully expressed. Specifically, in the technical scheme of the application, first, global average values of a plurality of weight matrixes of each trained internet financial risk assessment model are calculated respectively, so that the weight matrixes are converted into global average values, and a channel weight feature vector formed by the global average values is obtained. And further, obtaining a plurality of channel weight feature vectors for each trained internet financial risk assessment model.
And then, after the channel weight feature vectors are arranged into a two-dimensional feature matrix, performing feature mining on the two-dimensional feature matrix by using a convolutional neural network model which is used as a feature extractor and has excellent performance in terms of implicit association feature extraction so as to extract the implicit association feature distribution information among the weights of the trained internet financial risk assessment models, thereby obtaining a global weight feature matrix.
Further, after obtaining implicit association features among weights of the internet financial risk assessment models after the training is completed, taking the weight feature vectors of the channels as query feature vectors, and calculating matrix products between the query feature vectors and the global weight feature matrix to obtain a plurality of classification feature vectors. That is, the weight characteristics of the respective trained internet financial risk assessment models are extracted with implicit correlation characteristics between the respective weights based on the respective trained internet financial risk assessment models as a base background.
And then, the classification feature vectors are further classified by a classifier, and the weight features of the trained internet financial risk assessment models under the implicit association features among the weights are classified to obtain a plurality of probability values. And, in order to facilitate the subsequent weight weighting process, the plurality of probability values are further normalized so that the sum of the plurality of probability values is 1. Accordingly, in a specific example of the present application, the normalization processing of the plurality of probability values may be performed by means of average normalization. And then, fusing the trained internet financial risk assessment models by taking the probability values as weights. Therefore, the evaluation accuracy of the fused Internet financial risk evaluation model can be improved, and a global model based on virtual fusion data can be constructed, so that the balance of data privacy protection and data sharing calculation is realized.
In particular, in the technical solution of the present application, when the channel weight feature vector is used as a query feature vector and a matrix product between the query feature vector and the global weight feature matrix is calculated to obtain the classification feature vector, matrix-sample two-dimensional cross correlation features of channel weights represented by the weight matrix expressed by the global weight feature matrix are mapped into feature distributions of the weight matrix expressed by the channel weight feature vector, so that the weight matrix feature vector expresses not only the feature distribution of the weight matrix of a single model, but also feature expressions of the weight matrix feature vector under the feature distribution of the global weight matrix of each model, thereby enhancing feature expression capability of the weight matrix feature vector. On the other hand, the feature distribution of the classification feature vector is higher than the discretization degree of the feature distribution of the channel weight feature vector due to the feature representation under the feature distribution of the global weight matrix fused with each model, so that the convergence speed of the weight matrix of the classifier is slow in the training process, and the overall training speed of the model is influenced when the classification of the classification feature vectors through the classifier is respectively carried out.
Therefore, in the technical solution of the present application, the applicant of the present application performs spatial regularization constraint of the weight matrix at each iteration of the weight matrix of each classifier, expressed as:
Figure SMS_24
，/>
Figure SMS_25
is the weight matrix of the classifier, < >>
Figure SMS_26
Frobenius norms of the matrix are represented, < >>
Figure SMS_27
Is a bias matrix and may be initially set as an identity matrix, for example.
The spatial regularization constraint of the weight matrix is based on an endophytic correlation matrix obtained by spatial embedding the weight matrix with the transpose of the weight matrix, and L2 regularization based on endophytic correlation distribution of European space of the weight matrix is carried out on the weight matrix, so that the semantic dependency degree of the weight space on a specific mode expressed by the feature is reflected irrespective of the numerical distribution of the feature to be weighted and the numerical value according to the position, the transmission effect of the intrinsic knowledge of the extracted feature is reflected by the weight space, the convergence of the weight matrix is accelerated, and the overall training speed of the model is improved. Therefore, the evaluation accuracy of the fused internet financial risk evaluation model can be improved, so that the risk control of internet finance is optimized, a global model based on virtual fusion data is constructed, and the balance of data privacy protection and data sharing calculation is realized.
Based on this, the application proposes an internet financial risk control system based on federal learning, which includes: the data acquisition module is used for acquiring intra-domain training data provided by each financial service node, wherein the intra-domain training data comprises pedestrian credit data, three-party credit data, judicial information, industrial and commercial information, tax information, invoice information, public accumulation information, public transaction information and transaction information; the internet financial risk assessment model training module is used for respectively training the internet financial risk assessment model with a plurality of weight matrixes by taking the intra-domain training data provided by each financial service node as training data so as to obtain a plurality of trained internet financial risk assessment models; the channel weight calculation module is used for calculating global average values of a plurality of weight matrixes of the trained internet financial risk assessment models respectively to obtain channel weight feature vectors composed of a plurality of global average values so as to obtain a plurality of channel weight feature vectors; the weight global association module is used for arranging the channel weight feature vectors into a two-dimensional feature matrix and then obtaining a global weight feature matrix through a convolutional neural network model serving as a feature extractor; the query module is used for taking the weight feature vectors of the channels as query feature vectors, and calculating the matrix product between the query feature vectors and the global weight feature matrix to obtain a plurality of classification feature vectors; the weight probability value calculation module is used for enabling the classification feature vectors to pass through a classifier to obtain a plurality of probability values, and normalizing the probability values so that the sum of the probability values is 1; and the model integration module is used for fusing the plurality of trained internet financial risk assessment models by taking the plurality of probability values as weights.
Having described the basic principles of the present application, various non-limiting embodiments of the present application will now be described in detail with reference to the accompanying drawings.
Exemplary System: fig. 1 is a block diagram of an internet financial risk control system based on federal learning according to an embodiment of the present application. As shown in fig. 1, the federally-learned based internet financial risk control system 300 according to an embodiment of the present application includes an inference module, wherein the inference module includes: a data acquisition module 310; an internet financial risk assessment model training module 320; a channel weight calculation module 330; a weight global association module 340; a query module 350; a weight probability value calculation module 360; and, a model integration module 370.
The data collection module 310 is configured to obtain intra-domain training data provided by each financial service node, where the intra-domain training data includes pedestrian credit data, three-party credit data, judicial information, industrial and commercial information, tax information, invoice information, public deposit information, public transaction information, and transaction information; the training module 320 of the internet financial risk assessment model is configured to train the internet financial risk assessment model with a plurality of weight matrices to obtain a plurality of trained internet financial risk assessment models, respectively, with the intra-domain training data provided by the financial service nodes as training data; the channel weight calculation module 330 is configured to calculate global average values of a plurality of weight matrices of the trained internet financial risk assessment models respectively to obtain channel weight feature vectors composed of a plurality of global average values, so as to obtain a plurality of channel weight feature vectors; the weight global association module 340 is configured to arrange the multiple channel weight feature vectors into a two-dimensional feature matrix, and then obtain a global weight feature matrix through a convolutional neural network model serving as a feature extractor; the query module 350 is configured to calculate a matrix product between the query module 350 and the global weight feature matrix by using the channel weight feature vectors as query feature vectors, so as to obtain a plurality of classification feature vectors; the weight probability value calculation module 360 is configured to pass the plurality of classification feature vectors through a classifier to obtain a plurality of probability values, and normalize the plurality of probability values such that a sum of the plurality of probability values is 1; and the model integration module 370 is configured to fuse the trained internet financial risk assessment models with the probability values as weights.
Fig. 3 is a system architecture diagram of an inference module in an internet financial risk control system based on federal learning according to an embodiment of the present application. As shown in fig. 3, in the system architecture of the federal learning-based internet financial risk control system 300, in the inference process, first, intra-domain training data provided by each financial service node is acquired through the data acquisition module 310, where the intra-domain training data includes pedestrian credit data, three-party credit data, judicial information, industrial and commercial information, tax information, invoice information, public transaction information, and transaction information; then, the training module 320 of the internet financial risk assessment model trains the internet financial risk assessment model with a plurality of weight matrices by using the intra-domain training data provided by each financial service node acquired by the data acquisition module 310 as training data, so as to obtain a plurality of trained internet financial risk assessment models; the channel weight calculation module 330 calculates global average values of a plurality of weight matrices of each trained internet financial risk assessment model obtained by the internet financial risk assessment model training module 320 to obtain a channel weight feature vector composed of a plurality of global average values, so as to obtain a plurality of channel weight feature vectors; then, the weight global association module 340 arranges the channel weight feature vectors calculated by the channel weight calculation module 330 into a two-dimensional feature matrix, and then obtains a global weight feature matrix through a convolutional neural network model serving as a feature extractor; the query module 350 uses each channel weight feature vector calculated by the channel weight calculation module 330 as a query feature vector, and calculates a matrix product between the query feature vector and the global weight feature matrix obtained by the weight global association module 340 to obtain a plurality of classification feature vectors; the weight probability value calculation module 360 passes the classification feature vectors obtained by the query module 350 through a classifier to obtain a plurality of probability values, and normalizes the plurality of probability values so that the sum of the plurality of probability values is 1; further, the model integration module 370 fuses the plurality of trained internet financial risk assessment models with the plurality of probability values as weights.
Specifically, during operation of the federal learning-based internet financial risk control system 300, the data collection module 310 is configured to obtain intra-domain training data provided by each financial service node, where the intra-domain training data includes pedestrian credit data, three-party credit data, judicial information, industrial and commercial information, tax information, invoice information, public transaction information, and transaction information. It should be understood that in the process of actually integrating the parameters of each model, the weight matrix of each trained internet financial risk assessment model may be subjected to weighted fusion, and considering that the trained internet financial risk assessment model is obtained through intra-domain data training of each financial service node, performance of the trained internet financial risk assessment model may be relatively poor in other intra-domain data, that is, contribution degree of each internet financial risk assessment model degree to the final internet financial risk assessment model based on global data is different, so in the technical scheme of the application, the implicit correlation characteristic information between the weight matrices of each internet financial risk assessment model is mined, and therefore each internet financial risk assessment model is fused, so that assessment accuracy of the fused internet financial risk assessment model is improved. In one specific example of the present application, first, intra-domain training data provided by respective financial service nodes is obtained, wherein the intra-domain training data includes pedestrian credit data, three-party credit data, judicial information, business information, tax information, invoice information, public deposit information, public transaction information, and transaction information.
Specifically, during the operation of the federal learning-based internet financial risk control system 300, the internet financial risk assessment model training module 320 is configured to train the internet financial risk assessment model with a plurality of weight matrices to obtain a plurality of trained internet financial risk assessment models, respectively, with the intra-domain training data provided by the respective financial service nodes as training data. That is, training of the internet financial risk control model is performed based on the data of each node through federal learning, that is, specifically, the internet financial risk assessment model having a plurality of weight matrices is respectively trained with the intra-domain training data provided by each financial service node as training data to obtain a plurality of trained internet financial risk assessment models.
Specifically, during the operation of the federal learning-based internet financial risk control system 300, the channel weight calculation module 330 is configured to calculate global averages of a plurality of weight matrices of each trained internet financial risk assessment model respectively to obtain a channel weight feature vector composed of a plurality of global averages, so as to obtain a plurality of channel weight feature vectors. For each trained internet financial risk assessment model, the weight matrixes of the trained internet financial risk assessment models have a correlative relation, that is, the trained internet financial risk assessment models are interrelated, and form a risk assessment model whole. Therefore, if the trained internet financial risk assessment models are to be better integrated, so as to optimize the risk assessment accuracy of the models, the correlation characteristics between the weight matrixes of the trained internet financial risk assessment models need to be fully expressed. Specifically, in the technical scheme of the application, first, global average values of a plurality of weight matrixes of each trained internet financial risk assessment model are calculated respectively, so that the weight matrixes are converted into global average values, and a channel weight feature vector formed by the global average values is obtained. And further, obtaining a plurality of channel weight feature vectors for each trained internet financial risk assessment model.
Specifically, during the operation of the federal learning-based internet financial risk control system 300, the weight global association module 340 is configured to arrange the plurality of channel weight feature vectors into a two-dimensional feature matrix, and then obtain a global weight feature matrix through a convolutional neural network model serving as a feature extractor. That is, after the channel weight feature vectors are arranged into a two-dimensional feature matrix, feature mining of the two-dimensional feature matrix is performed by using a convolutional neural network model which is a feature extractor and has excellent performance in terms of implicit associated feature extraction, so as to extract the implicit associated feature distribution information among the weights of the trained internet financial risk assessment models, thereby obtaining a global weight feature matrix. In one particular example, the convolutional neural network includes a plurality of neural network layers that are cascaded with one another, wherein each neural network layer includes a convolutional layer, a pooling layer, and an activation layer. In the coding process of the convolutional neural network, each layer of the convolutional neural network carries out convolutional processing based on a convolutional kernel on input data by using the convolutional layer in the forward transmission process of the layer, carries out pooling processing on a convolutional feature map output by the convolutional layer by using the pooling layer and carries out activation processing on the pooling feature map output by the pooling layer by using the activation layer.
Fig. 5 is a flowchart of convolutional neural network coding in an internet financial risk control system based on federal learning according to an embodiment of the present application. As shown in fig. 5, in the encoding process of the convolutional neural network, the method includes: each layer of the convolutional neural network model using the feature extractor performs, in forward transfer of the layer, input data: s210, carrying out convolution processing on input data to obtain a convolution characteristic diagram; s220, pooling the convolution feature map along the channel dimension to obtain a pooled feature map; s230, carrying out nonlinear activation on the pooled feature map to obtain an activated feature map; the output of the last layer of the convolutional neural network serving as the feature extractor is the global weight feature matrix, and the input of the first layer of the convolutional neural network serving as the feature extractor is the two-dimensional feature matrix of the arrangement of the plurality of channel weight feature vectors.
Specifically, during the operation of the federal learning-based internet financial risk control system 300, the query module 350 is configured to calculate a matrix product between the channel weight feature vectors and the global weight feature matrix to obtain a plurality of classification feature vectors by using the channel weight feature vectors as query feature vectors. That is, after obtaining the implicit association features between the weights of the trained internet financial risk assessment models, the channel weight feature vectors are used as query feature vectors, and the matrix product between the channel weight feature vectors and the global weight feature matrix is calculated to obtain a plurality of classification feature vectors. That is, the weight characteristics of the respective trained internet financial risk assessment models are extracted with implicit correlation characteristics between the respective weights based on the respective trained internet financial risk assessment models as a base background.
Specifically, during the operation of the federal learning-based internet financial risk control system 300, the weight probability value calculation module 360 and the model integration module 370 are configured to pass the plurality of classification feature vectors through a classifier to obtain a plurality of probability values, and normalize the plurality of probability values such that a sum of the plurality of probability values is 1; and fusing the plurality of trained internet financial risk assessment models by taking the plurality of probability values as weights. In the technical scheme of the application, the classification feature vectors are classified by a classifier, namely, the weight features of the internet financial risk assessment model which are completed by each training under the implicit association features among the weights are used for obtaining a plurality of probability values. To facilitate subsequent weight weighting, the plurality of probability values are further normalized such that the plurality of probabilitiesThe sum of the values is 1. Specifically, the plurality of classification feature vectors are processed using the classifier to obtain a plurality of probability values in a classification formula:
Figure SMS_28
wherein- >
Figure SMS_29
To->
Figure SMS_30
Is a weight matrix>
Figure SMS_31
To the point of
Figure SMS_32
For the bias vector +.>
Figure SMS_33
For a plurality of classification feature vectors, < >>
Figure SMS_34
Representing the plurality of probability values. Accordingly, in a specific example of the present application, the normalization processing of the plurality of probability values may be performed by means of average normalization. And then, fusing the trained internet financial risk assessment models by taking the probability values as weights. Therefore, the evaluation accuracy of the fused Internet financial risk evaluation model can be improved, and a global model based on virtual fusion data can be constructed, so that the balance of data privacy protection and data sharing calculation is realized.
It should be appreciated that the convolutional neural network model as a feature extractor and the classifier need to be trained prior to the inference using the neural network model described above. That is, in the internet financial risk control system based on federal learning of the present application, a training module is further included for training the convolutional neural network model as the feature extractor and the classifier. The training of deep neural networks mostly adopts a back propagation algorithm, and the back propagation algorithm updates the parameters of the current layer through errors transmitted by the later layer by using a chained method, which can suffer from the problem of gradient disappearance or more broadly, the problem of unstable gradient when the network is deep.
Fig. 2 is a block diagram of an internet financial risk control system based on federal learning according to an embodiment of the present application. As shown in fig. 2, the federal learning-based internet financial risk control system 300 according to an embodiment of the present application further includes a training module 400 including: a training data acquisition module 410; training a channel weight calculation module 420; training the weight global association module 430; training a query module 440; a classification loss module 450; the training module 460 is optimized.
The training data acquisition module 410 is configured to acquire optimized training data, where the optimized training data includes a plurality of training weight matrices of the internet financial risk assessment model that are completed by the respective training, and true values of the plurality of probability values; the training channel weight calculation module 420 is configured to calculate training global average values of a plurality of training weight matrices of the trained internet financial risk assessment model respectively to obtain training channel weight feature vectors composed of the plurality of training global average values, so as to obtain a plurality of training channel weight feature vectors; the training weight global association module 430 is configured to arrange the training channel weight feature vectors into a training two-dimensional feature matrix, and then obtain a training global weight feature matrix through the convolutional neural network model serving as the feature extractor; the training query module 440 is configured to calculate a matrix product between the training channel weight feature vectors and the training global weight feature matrix to obtain a plurality of training classification feature vectors by using the training channel weight feature vectors as query feature vectors; the classification loss module 450 is configured to pass the plurality of training classification feature vectors through the classifier to obtain a plurality of classification loss function values; the optimization training module 460 is configured to train the convolutional neural network model as the feature extractor and the classifier based on the multiple classification loss function values and propagating in a gradient descent direction, where in each iteration of the training, a spatial regularization constraint iteration of a weight matrix is performed on a weight matrix of the classifier.
Fig. 4 is a system architecture diagram of a training module in an internet financial risk control system based on federal learning according to an embodiment of the present application. As shown in fig. 4, in the system architecture of the federal learning-based internet financial risk control system 300, in a training module 400, first, optimized training data is acquired through the training data acquisition module 410, where the optimized training data includes a plurality of training weight matrices of the respective trained internet financial risk assessment models, and true values of the plurality of probability values; next, the training channel weight calculation module 420 calculates training global average values of a plurality of training weight matrices of each trained internet financial risk assessment model acquired by the training data acquisition module 410 respectively to obtain training channel weight feature vectors composed of the plurality of training global average values, so as to obtain a plurality of training channel weight feature vectors; the training weight global association module 430 arranges the training channel weight feature vectors calculated by the training channel weight calculation module 420 into a training two-dimensional feature matrix, and then obtains a training global weight feature matrix through the convolutional neural network model serving as the feature extractor; then, the training query module 440 uses the training channel weight feature vectors calculated by the training channel weight calculation module 420 as query feature vectors, and calculates a matrix product between the query feature vectors and the training global weight feature matrix obtained by the training global weight correlation module 430 to obtain a plurality of training classification feature vectors; the classification loss module 450 passes the training classification feature vectors obtained by the training query module 440 through the classifier to obtain a plurality of classification loss function values; the optimization training module 460 trains the convolutional neural network model as a feature extractor and the classifier based on the plurality of classification loss function values and propagating through the direction of gradient descent, wherein in each round of iteration of the training, a spatial regularization constraint iteration of a weight matrix is performed on the weight matrix of the classifier.
In the technical scheme of the application, when the channel weight feature vector is used as a query feature vector and a matrix product between the query feature vector and the global weight feature matrix is calculated to obtain the classification feature vector, matrix-sample two-dimensional cross correlation features of channel weights represented by the weight matrix expressed by the global weight feature matrix are mapped into feature distribution of the weight matrix expressed by the channel weight feature vector, so that the weight matrix feature vector expresses not only the weight matrix feature distribution of a single model, but also feature expression of the weight matrix feature vector under the global weight matrix feature distribution of each model, and the feature expression capability of the weight matrix feature vector is enhanced. On the other hand, the feature distribution of the classification feature vector is higher than the discretization degree of the feature distribution of the channel weight feature vector due to the feature representation under the feature distribution of the global weight matrix fused with each model, so that the convergence speed of the weight matrix of the classifier is slow in the training process, and the overall training speed of the model is influenced when the classification of the classification feature vectors through the classifier is respectively carried out. Therefore, in the technical solution of the present application, the applicant of the present application performs spatial regularization constraint of the weight matrix at each iteration of the weight matrix of each classifier, expressed as:
Figure SMS_36
Wherein->
Figure SMS_38
Is a weight matrix of the classifier, +.>
Figure SMS_43
Frobenius norms of the matrix are represented, < >>
Figure SMS_37
Transpose of the weight matrix representing the classifier, < >>
Figure SMS_39
Is a bias matrix, +.>
Figure SMS_41
Representing matrix multiplication +.>
Figure SMS_44
Representing matrix addition, ++>
Figure SMS_35
Representing the multiplication by the position point,
Figure SMS_40
an exponential operation representing a matrix representing the calculation of a natural exponential function value raised to a power by a characteristic value at each position in the matrix,/v>
Figure SMS_42
And representing the weight matrix of the classifier after iteration. The spatial regularization constraint of the weight matrix is based on an endophytic correlation matrix obtained by spatial embedding the weight matrix with the transpose of the weight matrix, and L2 regularization based on endophytic correlation distribution of European space of the weight matrix is carried out on the weight matrix, so that the semantic dependency degree of the weight space on a specific mode expressed by the feature is reflected irrespective of the numerical distribution of the feature to be weighted and the numerical value according to the position, the transmission effect of the intrinsic knowledge of the extracted feature is reflected by the weight space, the convergence of the weight matrix is accelerated, and the overall training speed of the model is improved. Therefore, the evaluation accuracy of the fused internet financial risk evaluation model can be improved, so that the risk control of internet finance is optimized, a global model based on virtual fusion data is constructed, and the balance of data privacy protection and data sharing calculation is realized.
In summary, the federal learning-based internet financial risk control system 300 according to the embodiment of the present application is illustrated, which uses a deep learning-based neural network model to mine implicit association feature information between weight matrices of each internet financial risk assessment model, so as to fuse each internet financial risk assessment model, so as to improve the assessment accuracy of the fused internet financial risk assessment model.
As described above, the federally learning-based internet financial risk control system according to the embodiment of the present application may be implemented in various terminal devices. In one example, the federally-learned based internet financial risk control system 300 according to embodiments of the present application may be integrated into a terminal device as a software module and/or hardware module. For example, the federally-learned internet financial risk control system 300 may be a software module in the operating system of the terminal device, or may be an application developed for the terminal device; of course, the federally learning-based internet financial risk control system 300 can also be one of a number of hardware modules of the terminal device.
Alternatively, in another example, the federally-learning-based internet financial risk control system 300 and the terminal device may be separate devices, and the federally-learning-based internet financial risk control system 300 may be connected to the terminal device through a wired and/or wireless network and transmit interactive information in a agreed data format.
An exemplary method is: fig. 6 is a flowchart of an internet financial risk control method based on federal learning according to an embodiment of the present application. As shown in fig. 6, the internet financial risk control method based on federal learning according to an embodiment of the present application includes the steps of: s110, acquiring intra-domain training data provided by each financial service node, wherein the intra-domain training data comprises pedestrian credit data, three-party credit data, judicial information, industrial and commercial information, tax information, invoice information, public deposit information, public transaction information and transaction information; s120, training the Internet financial risk assessment model with a plurality of weight matrixes by taking intra-domain training data provided by each financial service node as training data so as to obtain a plurality of trained Internet financial risk assessment models; s130, calculating global average values of a plurality of weight matrixes of the trained internet financial risk assessment models respectively to obtain channel weight feature vectors composed of the global average values so as to obtain a plurality of channel weight feature vectors; s140, arranging the channel weight feature vectors into a two-dimensional feature matrix, and then obtaining a global weight feature matrix through a convolutional neural network model serving as a feature extractor; s150, taking the weight feature vectors of all channels as query feature vectors, and calculating a matrix product between the query feature vectors and the global weight feature matrix to obtain a plurality of classification feature vectors; s160, enabling the classification feature vectors to pass through a classifier to obtain a plurality of probability values, and normalizing the probability values so that the sum of the probability values is 1; and S170, fusing the trained internet financial risk assessment models by taking the probability values as weights.
In one example, in the above-mentioned federal learning-based internet financial risk control method, the step S140 includes: each layer of the convolutional neural network model using the feature extractor performs, in forward transfer of the layer, input data: carrying out convolution processing on input data to obtain a convolution characteristic diagram; pooling the convolution feature map along a channel dimension to obtain a pooled feature map; performing nonlinear activation on the pooled feature map to obtain an activated feature map; the output of the last layer of the convolutional neural network serving as the feature extractor is the global weight feature matrix, and the input of the first layer of the convolutional neural network serving as the feature extractor is the two-dimensional feature matrix of the arrangement of the plurality of channel weight feature vectors.
In one example, in the above-mentioned federal learning-based internet financial risk control method, the step S150 includes: calculating a matrix product between each channel weight feature vector and the global weight feature matrix to obtain a plurality of classification feature vectors according to the following formula; wherein, the formula is:
Figure SMS_45
wherein->
Figure SMS_46
Representing the weight feature vector of each channel, +. >
Figure SMS_47
Representing the global weight feature matrix, +.>
Figure SMS_48
Representing the plurality of classification feature vectors, +.>
Figure SMS_49
Representing a matrix multiplication.
In one example, in the above-mentioned federal learning-based internet financial risk control method, the step S160 includes: processing the plurality of classification feature vectors using the classifier to obtain a plurality of probability values in a classification formula, wherein the formula is:
Figure SMS_50
wherein->
Figure SMS_51
To->
Figure SMS_52
Is a weight matrix>
Figure SMS_53
To->
Figure SMS_54
For the bias vector +.>
Figure SMS_55
For a plurality of classification feature vectors, < >>
Figure SMS_56
Representing the plurality of probability values.
In summary, the internet financial risk control method based on federal learning according to the embodiment of the application is explained, and by adopting a neural network model based on deep learning to mine implicit association characteristic information among weight matrixes of all internet financial risk assessment models, all internet financial risk assessment models are fused, so that the assessment accuracy of the fused internet financial risk assessment models is improved.
Exemplary electronic device: next, an electronic device according to an embodiment of the present application is described with reference to fig. 7.
Fig. 7 illustrates a block diagram of an electronic device according to an embodiment of the present application.
As shown in fig. 7, the electronic device 10 includes one or more processors 11 and a memory 12.
The processor 11 may be a Central Processing Unit (CPU) or other form of processing unit having data processing and/or instruction execution capabilities, and may control other components in the electronic device 10 to perform desired functions.
Memory 12 may include one or more computer program products that may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory may include, for example, random Access Memory (RAM) and/or cache memory (cache), and the like. The non-volatile memory may include, for example, read Only Memory (ROM), hard disk, flash memory, and the like. One or more computer program instructions may be stored on the computer readable storage medium that can be executed by the processor 11 to implement the functions in the federally learned internet financial risk control system and/or other desired functions of the various embodiments of the present application described above. Various contents such as channel weight feature vectors may also be stored in the computer readable storage medium.
In one example, the electronic device 10 may further include: an input device 13 and an output device 14, which are interconnected by a bus system and/or other forms of connection mechanisms (not shown).
The input means 13 may comprise, for example, a keyboard, a mouse, etc.
The output device 14 may output various information including a probability value or the like to the outside. The output means 14 may include, for example, a display, speakers, a printer, and a communication network and remote output devices connected thereto, etc.
Of course, only some of the components of the electronic device 10 that are relevant to the present application are shown in fig. 7 for simplicity, components such as buses, input/output interfaces, etc. are omitted. In addition, the electronic device 10 may include any other suitable components depending on the particular application.
Exemplary computer program product and computer readable storage Medium: in addition to the methods and apparatus described above, embodiments of the present application may also be a computer program product comprising computer program instructions which, when executed by a processor, cause the processor to perform steps in the functions of the federally learning-based internet financial risk control method according to various embodiments of the present application described in the "exemplary systems" section of this specification.
The computer program product may write program code for performing the operations of embodiments of the present application in any combination of one or more programming languages, including an object oriented programming language such as Java, C++ or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device, partly on a remote computing device, or entirely on the remote computing device or server.
Furthermore, embodiments of the present application may also be a computer-readable storage medium, having stored thereon computer program instructions that, when executed by a processor, cause the processor to perform steps in the functions of the federally learning-based internet financial risk control method according to various embodiments of the present application described in the "exemplary systems" section of the present specification.
The computer readable storage medium may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may include, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium would include the following: an electrical connection having one or more wires, a portable disk, a hard disk, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
The basic principles of the present application have been described above in connection with specific embodiments, however, it should be noted that the advantages, benefits, effects, etc. mentioned in the present application are merely examples and not limiting, and these advantages, benefits, effects, etc. are not to be considered as necessarily possessed by the various embodiments of the present application. Furthermore, the specific details disclosed herein are for purposes of illustration and understanding only, and are not intended to be limiting, as the application is not intended to be limited to the details disclosed herein as such.
The block diagrams of the devices, apparatuses, devices, systems referred to in this application are only illustrative examples and are not intended to require or imply that the connections, arrangements, configurations must be made in the manner shown in the block diagrams. As will be appreciated by one of skill in the art, the devices, apparatuses, devices, systems may be connected, arranged, configured in any manner. Words such as "including," "comprising," "having," and the like are words of openness and mean "including but not limited to," and are used interchangeably therewith. The terms "or" and "as used herein refer to and are used interchangeably with the term" and/or "unless the context clearly indicates otherwise. The term "such as" as used herein refers to, and is used interchangeably with, the phrase "such as, but not limited to.
It is also noted that in the apparatus, devices and methods of the present application, the components or steps may be disassembled and/or assembled. Such decomposition and/or recombination should be considered as equivalent to the present application.
The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the application to the form disclosed herein. Although a number of example aspects and embodiments have been discussed above, a person of ordinary skill in the art will recognize certain variations, modifications, alterations, additions, and subcombinations thereof.

Claims (10)

1. An internet financial risk control system based on federal learning, comprising: the data acquisition module is used for acquiring intra-domain training data provided by each financial service node, wherein the intra-domain training data comprises pedestrian credit data, three-party credit data, judicial information, industrial and commercial information, tax information, invoice information, public accumulation information, public transaction information and transaction information; the internet financial risk assessment model training module is used for respectively training the internet financial risk assessment model with a plurality of weight matrixes by taking the intra-domain training data provided by each financial service node as training data so as to obtain a plurality of trained internet financial risk assessment models; the channel weight calculation module is used for calculating global average values of a plurality of weight matrixes of the trained internet financial risk assessment models respectively to obtain channel weight feature vectors composed of a plurality of global average values so as to obtain a plurality of channel weight feature vectors; the weight global association module is used for arranging the channel weight feature vectors into a two-dimensional feature matrix and then obtaining a global weight feature matrix through a convolutional neural network model serving as a feature extractor; the query module is used for taking the weight feature vectors of the channels as query feature vectors, and calculating the matrix product between the query feature vectors and the global weight feature matrix to obtain a plurality of classification feature vectors; the weight probability value calculation module is used for enabling the classification feature vectors to pass through a classifier to obtain a plurality of probability values, and normalizing the probability values so that the sum of the probability values is 1; and the model integration module is used for fusing the plurality of trained internet financial risk assessment models by taking the plurality of probability values as weights.
2. The federally learned based internet financial risk control system according to claim 1, wherein the weight global association module is configured to: each layer of the convolutional neural network model using the feature extractor performs, in forward transfer of the layer, input data: carrying out convolution processing on input data to obtain a convolution characteristic diagram; pooling the convolution feature map along a channel dimension to obtain a pooled feature map; non-linear activation is carried out on the pooled feature map so as to obtain an activated feature map; the output of the last layer of the convolutional neural network serving as the feature extractor is the global weight feature matrix, and the input of the first layer of the convolutional neural network serving as the feature extractor is the two-dimensional feature matrix of the arrangement of the plurality of channel weight feature vectors.
3. The federally learned based internet financial risk control system according to claim 2, wherein the query module is configured to: calculating a matrix product between each channel weight feature vector and the global weight feature matrix to obtain a plurality of classification feature vectors according to the following formula; wherein, the formula is:
Figure QLYQS_1
Wherein->
Figure QLYQS_2
Representing the weight feature vector of each channel, +.>
Figure QLYQS_3
Representing the global weight feature matrix, +.>
Figure QLYQS_4
Representing the plurality of classification feature vectors, +.>
Figure QLYQS_5
Representing a matrix multiplication.
4. The federally learned based internet financial risk control system according to claim 3, wherein the weight probability value calculation module is configured to: processing the plurality of classification feature vectors using the classifier to obtain a plurality of probability values in a classification formula, wherein the formula is:
Figure QLYQS_6
wherein->
Figure QLYQS_7
To->
Figure QLYQS_8
Is a weight matrix>
Figure QLYQS_9
To the point of
Figure QLYQS_10
For the bias vector +.>
Figure QLYQS_11
For a plurality of classification feature vectors, < >>
Figure QLYQS_12
Representing the plurality of probability values.
5. The federally-learned, internet financial risk control system according to claim 4, further comprising an optimization training module for training the convolutional neural network model as a feature extractor and the classifier.
6. The federally-learned, internet financial risk control system according to claim 5, wherein the optimization training module comprises: the training data acquisition module is used for acquiring optimized training data, wherein the optimized training data comprises a plurality of training weight matrixes of the internet financial risk assessment models with which the training is completed and the true values of the probability values; the training channel weight calculation module is used for calculating training global average values of a plurality of training weight matrixes of the internet financial risk assessment models after the training is completed respectively to obtain training channel weight feature vectors composed of the training global average values so as to obtain a plurality of training channel weight feature vectors; the training weight global association module is used for arranging the training channel weight feature vectors into a training two-dimensional feature matrix and obtaining a training global weight feature matrix through the convolutional neural network model serving as the feature extractor; the training query module is used for taking the weight feature vectors of the training channels as query feature vectors, and calculating the matrix product between the weight feature vectors and the training global weight feature matrix to obtain a plurality of training classification feature vectors; the classification loss module is used for enabling the training classification feature vectors to pass through the classifier to obtain a plurality of classification loss function values; and the optimization training module is used for training the convolutional neural network model serving as the feature extractor and the classifier based on the plurality of classification loss function values and through gradient descent direction propagation, wherein in each round of iteration of training, the spatial regularization constraint iteration of the weight matrix of the classifier is carried out.
7. The federally-learned, internet financial risk control system according to claim 6, wherein the categorical loss module comprises: performing full-connection coding on the training classification feature vectors by using a plurality of full-connection layers of the classifier to obtain coded classification feature vectors; and passing the encoded classification feature vector through the classifier
Figure QLYQS_13
Classifying the function to obtain the plurality of classification loss function values.
8. The federally learned based internet financial risk control system according to claim 7, wherein in each iteration of the training, the weight matrix of the classifier is iterated with a spatial regularization constraint of the weight matrix with the following optimization formula; wherein, the optimization formula is:
Figure QLYQS_16
wherein->
Figure QLYQS_19
Is a weight matrix of the classifier, +.>
Figure QLYQS_21
Frobenius norms of the matrix are represented, < >>
Figure QLYQS_15
Transpose of the weight matrix representing the classifier, < >>
Figure QLYQS_18
Is a bias matrix, +.>
Figure QLYQS_20
Representing matrix multiplication +.>
Figure QLYQS_23
Representing matrix addition, ++>
Figure QLYQS_14
Representing the multiplication by the position point,
Figure QLYQS_17
an exponential operation representing a matrix, the exponential operation representing a calculationNatural exponential function value raised to power by eigenvalues at each position in matrix, < > >
Figure QLYQS_22
And representing the weight matrix of the classifier after iteration.
9. An internet financial risk control method based on federal learning, which is characterized by comprising the following steps: acquiring intra-domain training data provided by each financial service node, wherein the intra-domain training data comprises pedestrian credit data, three-party credit data, judicial information, industrial and commercial information, tax information, invoice information, public deposit information, public transaction information and transaction information; training the internet financial risk assessment model with a plurality of weight matrixes by taking intra-domain training data provided by each financial service node as training data to obtain a plurality of trained internet financial risk assessment models; respectively calculating global average values of a plurality of weight matrixes of the trained internet financial risk assessment models to obtain channel weight feature vectors composed of the global average values so as to obtain a plurality of channel weight feature vectors; the channel weight feature vectors are arranged into a two-dimensional feature matrix and then a global weight feature matrix is obtained through a convolutional neural network model serving as a feature extractor; taking each channel weight feature vector as a query feature vector, and calculating a matrix product between the query feature vector and the global weight feature matrix to obtain a plurality of classification feature vectors; passing the plurality of classification feature vectors through a classifier to obtain a plurality of probability values, and normalizing the plurality of probability values so that the sum of the plurality of probability values is 1; and fusing the plurality of trained internet financial risk assessment models with the plurality of probability values as weights.
10. The method of claim 9, wherein the step of arranging the plurality of channel weight feature vectors into a two-dimensional feature matrix and then obtaining a global weight feature matrix by using a convolutional neural network model as a feature extractor comprises the steps of: each layer of the convolutional neural network model using the feature extractor performs, in forward transfer of the layer, input data: carrying out convolution processing on input data to obtain a convolution characteristic diagram; pooling the convolution feature map along a channel dimension to obtain a pooled feature map; non-linear activation is carried out on the pooled feature map so as to obtain an activated feature map; the output of the last layer of the convolutional neural network serving as the feature extractor is the global weight feature matrix, and the input of the first layer of the convolutional neural network serving as the feature extractor is the two-dimensional feature matrix of the arrangement of the plurality of channel weight feature vectors.
CN202310419480.4A 2023-04-19 2023-04-19 Internet financial risk control system and method based on federal learning Active CN116167833B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310419480.4A CN116167833B (en) 2023-04-19 2023-04-19 Internet financial risk control system and method based on federal learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310419480.4A CN116167833B (en) 2023-04-19 2023-04-19 Internet financial risk control system and method based on federal learning

Publications (2)

Publication Number Publication Date
CN116167833A CN116167833A (en) 2023-05-26
CN116167833B true CN116167833B (en) 2023-06-30

Family

ID=86422151

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310419480.4A Active CN116167833B (en) 2023-04-19 2023-04-19 Internet financial risk control system and method based on federal learning

Country Status (1)

Country Link
CN (1) CN116167833B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116580029B (en) * 2023-07-12 2023-10-13 浙江海威汽车零件有限公司 Quality inspection control system and method for aluminum alloy casting finished product

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109767322A (en) * 2018-12-20 2019-05-17 平安科技（深圳）有限公司 Suspicious transaction analysis method, apparatus and computer equipment based on big data
CN110751286A (en) * 2018-07-23 2020-02-04 第四范式（北京）技术有限公司 Training method and training system of neural network model
CN110751261A (en) * 2018-07-23 2020-02-04 第四范式（北京）技术有限公司 Training method and system and prediction method and system of neural network model
CN112464925A (en) * 2020-11-11 2021-03-09 湖北省楚建易网络科技有限公司 Mobile terminal account opening data bank information automatic extraction method based on machine learning
CN114565041A (en) * 2022-02-28 2022-05-31 上海嘉甲茂技术有限公司 Payment big data analysis system based on internet finance and analysis method thereof
CN115908022A (en) * 2022-12-05 2023-04-04 中信银行股份有限公司 Abnormal transaction risk early warning method and system based on network modeling

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110751286A (en) * 2018-07-23 2020-02-04 第四范式（北京）技术有限公司 Training method and training system of neural network model
CN110751261A (en) * 2018-07-23 2020-02-04 第四范式（北京）技术有限公司 Training method and system and prediction method and system of neural network model
CN109767322A (en) * 2018-12-20 2019-05-17 平安科技（深圳）有限公司 Suspicious transaction analysis method, apparatus and computer equipment based on big data
CN112464925A (en) * 2020-11-11 2021-03-09 湖北省楚建易网络科技有限公司 Mobile terminal account opening data bank information automatic extraction method based on machine learning
CN114565041A (en) * 2022-02-28 2022-05-31 上海嘉甲茂技术有限公司 Payment big data analysis system based on internet finance and analysis method thereof
CN115908022A (en) * 2022-12-05 2023-04-04 中信银行股份有限公司 Abnormal transaction risk early warning method and system based on network modeling

Also Published As

Publication number Publication date
CN116167833A (en) 2023-05-26

Similar Documents

Publication Publication Date Title
CN111274134B (en) Vulnerability identification and prediction method, system, computer equipment and storage medium based on graph neural network
CN108959482B (en) Single-round dialogue data classification method and device based on deep learning and electronic equipment
CN115796173B (en) Data processing method and system for supervising reporting requirements
CN113596007B (en) Vulnerability attack detection method and device based on deep learning
US20220382553A1 (en) Fine-grained image recognition method and apparatus using graph structure represented high-order relation discovery
CN110602113A (en) Hierarchical phishing website detection method based on deep learning
CN110175248B (en) Face image retrieval method and device based on deep learning and Hash coding
CN116167833B (en) Internet financial risk control system and method based on federal learning
CN115834433B (en) Data processing method and system based on Internet of things technology
WO2023179429A1 (en) Video data processing method and apparatus, electronic device, and storage medium
CN115827257B (en) CPU capacity prediction method and system for processor system
CN116579618B (en) Data processing method, device, equipment and storage medium based on risk management
CN116015837A (en) Intrusion detection method and system for computer network information security
CN116859830B (en) Production management control system for electronic grade ammonium fluoride production
CN116308754B (en) Bank credit risk early warning system and method thereof
CN115759658A (en) Enterprise energy consumption data management system suitable for smart city
CN113628059A (en) Associated user identification method and device based on multilayer graph attention network
CN114611672A (en) Model training method, face recognition method and device
CN114529085A (en) Resident income prediction system based on big data and prediction method thereof
CN112749737A (en) Image classification method and device, electronic equipment and storage medium
US20240046065A1 (en) System, devices and/or processes for defining a search space for neural network processing device architectures
CN116722992A (en) Fraud website identification method and device based on multi-mode fusion
CN116467485A (en) Video image retrieval construction system and method thereof
CN116307624A (en) Resource scheduling method and system of ERP system
CN115620342A (en) Cross-modal pedestrian re-identification method, system and computer

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant