Bank anti-fraud model optimization method and device, equipment and medium
Through TensorFlow Federated technology, the problem of difficulty in data sharing among different institutions is solved, the accuracy of bank anti-fraud models and data privacy protection is improved, data silos are broken, and cross-industry data sharing is realized.
Patent Information
- Application Number
- CN202411552927.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-31
- Publication Date
- 2025-02-11
AI Technical Summary
Due to data privacy protection and business barriers, data between different institutions is difficult to share, limiting the performance improvement of the anti-fraud model.
Adopt the bank anti-fraud model optimization method based on TensorFlow Federated. By determining the data standards of participants, a TensorFlow Federated environment is built, local data is preprocessed, key features are extracted, deep learning models are constructed, and federated learning tasks are performed in the TensorFlow Federated environment. Update the iterative model parameters.
Through federated learning technology, cross-institutional data collaboration is achieved, the model training samples and feature dimensions are enriched, and the accuracy of the anti-fraud model is significantly improved. At the same time, data privacy and security are protected, data silos are broken, and data across industries and institutions are realized. Sharing and maximizing value utilization.
Smart Images
Figure CN119416869A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of bank intelligent anti-fraud technology and financial technology, and in particular to a bank anti-fraud model optimization method, device, equipment and medium based on TensorFlow Federated. Background Art
[0002] With the rapid development of financial services, fraudulent behaviors are becoming increasingly complex and changeable, bringing huge challenges to bank risk management. Traditional anti-fraud methods mainly rely on expert rules, blacklist sharing and other methods, which have problems such as high misjudgment rate and untimely updates. At the same time, due to data privacy protection and commercial barriers, data between different institutions is difficult to share, which limits the performance improvement of anti-fraud models. Summary of the invention
[0003] The present invention provides a bank anti-fraud model optimization method, device, computer equipment and medium based on TensorFlow Federated to solve the technical problem that data between different institutions is difficult to share due to data privacy protection and commercial barriers, which limits the performance improvement of the anti-fraud model.
[0004] First, a bank anti-fraud model optimization method based on TensorFlow Federated is provided, including:
[0005] Determine the data standards for participants and participants;
[0006] Build a TensorFlow Federated environment based on the data standards of the participants;
[0007] Preprocess the local data of the participants and extract key features;
[0008] Build a deep learning model of key features and initialize the deep learning model parameters in the TensorFlow Federated environment;
[0009] Execute federated learning tasks and update and iterate deep learning model parameters based on federated learning tasks.
[0010] In the second aspect, a bank anti-fraud model optimization device based on TensorFlow Federated is provided, including:
[0011] A determination module is used to determine the data standards of the participants and the participants;
[0012] Build a module for building a TensorFlow Federated environment based on the data standards of the participants;
[0013] The extraction module is used to pre-process the local data of the participants and extract key features;
[0014] The deep learning module is used to build a deep learning model of key features and initialize the deep learning model parameters in the TensorFlow Federated environment;
[0015] The update iteration module is used to execute federated learning tasks and update and iterate the parameters of the deep learning model based on the federated learning tasks.
[0016] In a third aspect, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the above-mentioned bank anti-fraud model optimization method based on TensorFlowFederated are implemented.
[0017] In a fourth aspect, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned bank anti-fraud model optimization method based on TensorFlow Federated are implemented.
[0018] In the scheme implemented by the above-mentioned bank anti-fraud model optimization method, device, computer equipment and storage medium based on TensorFlow Federated, the data standards of the participants and the participants can be determined; the TensorFlow Federated environment can be built based on the data standards of the participants; the local data of the participants can be preprocessed and key features can be extracted; a deep learning model of key features can be constructed, and the deep learning model parameters can be initialized in the TensorFlow Federated environment; the federated learning task can be executed, and the deep learning model parameters can be updated and iterated based on the federated learning task. To improve the accuracy of the anti-fraud model: cross-institutional data collaborative modeling can be achieved through federated learning technology, which enriches the model training samples and feature dimensions and significantly improves the accuracy of the anti-fraud model; protect data privacy and security: during the entire federated learning process, the original data is always retained locally, and only the model update information is transmitted instead of the original data, which ensures the privacy and security of the data provider and complies with the requirements of relevant laws and regulations; break the data island phenomenon: solve the data island problem in traditional anti-fraud methods, and realize cross-industry and cross-institutional data sharing and maximize value utilization. BRIEF DESCRIPTION OF THE DRAWINGS
[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.
[0020] Figure 1 It is a schematic diagram of an application environment of a bank anti-fraud model optimization method based on TensorFlow Federated in one embodiment of the present invention;
[0021] Figure 2 It is a flow chart of a bank anti-fraud model optimization method based on TensorFlow Federated in one embodiment of the present invention;
[0022] Figure 3 It is a structural schematic diagram of a bank anti-fraud model optimization device based on TensorFlow Federated in one embodiment of the present invention;
[0023] Figure 4 is a schematic diagram of a structure of a computer device in one embodiment of the present invention;
[0024] Figure 5 It is another structural schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION
[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0026] The bank anti-fraud model optimization method based on TensorFlow Federated provided in the embodiment of the present invention can be applied in Figure 1In the application environment, the client communicates with the server through the network. The server can determine the data standards between the participants; build a TensorFlow Federated environment based on the data standards of the participants; pre-process the local data of the participants and extract key features; build a deep learning model of key features, and initialize the deep learning model parameters in the TensorFlow Federated environment; execute federated learning tasks, and update and iterate the deep learning model parameters based on the federated learning tasks. To improve the accuracy of the anti-fraud model: cross-institutional data collaborative modeling is achieved through federated learning technology, which enriches the model training samples and feature dimensions and significantly improves the accuracy of the anti-fraud model; protect data privacy and security: during the entire federated learning process, the original data is always retained locally, and only the model update information rather than the original data is transmitted, which ensures the privacy and security of the data provider in accordance with relevant laws and regulations; break the data island phenomenon: solve the data island problem in traditional anti-fraud methods, and realize cross-industry and cross-institutional data sharing and maximize value utilization. The present invention is described in detail below through specific embodiments.
[0027] See also Figure 2 As shown, Figure 2 A flow chart of a bank anti-fraud model optimization method based on TensorFlow Federated provided in an embodiment of the present invention includes the following steps:
[0028] S10: Determine the data standards between participants and participants;
[0029] Among them, data standards include: standard format for data exchange, encryption protocol type, data fields, goals and evaluation indicators.
[0030] Specifically, the participants are determined as follows: First, the various institutions involved in federated learning, such as banks, telecom operators, e-commerce platforms, etc., are clearly defined. These institutions should participate in the same banking project based on common business needs and willingness to cooperate; Data standard formulation: The participants jointly discuss and determine the standard format for data exchange, such as JSON, CSV, etc., to ensure the consistency and processability of the data; Encryption protocol: Determine the protocol for data encryption transmission, such as AES, TLS, etc., to protect the privacy and security during data transmission; Data fields: Clarify the data fields that each participant needs to provide, such as user ID, transaction records, communication behavior, consumption behavior, etc., and unify the field naming and data type; Goals and evaluation indicators: Clarify the goals of the federated learning task and the main indicators for evaluating model performance, such as fraud identification accuracy, recall rate, F1 score, etc.
[0031] S20: Build a TensorFlow Federated environment based on the data standards of the participants;
[0032] TensorFlow Federated (TFF) is an open source framework developed by Google that is specifically designed to support machine learning and federated learning tasks on decentralized data. TFF allows multiple clients to train local models locally and aggregate local model updates to a central server, thereby training a global shared model without directly sharing the original data. This feature provides an effective way to solve the data privacy and data island problems in bank anti-fraud.
[0033] Wherein, step S20 includes:
[0034] Install Python and TensorFlow libraries on the participant's server based on the participant's data standards and configure necessary dependencies; install TensorFlow Federated library in other package management tools and configure dependencies; build a central server, configure the dependencies of TensorFlow Federated library and Python and TensorFlow library dependencies to the central server.
[0035] Specifically, environment preparation: install Python and TensorFlow libraries on the server of each participant, and configure necessary dependencies; install TensorFlow Federated: use pip or other package management tools to install the TensorFlowFederated library and its dependencies, and ensure that the TFF versions of all participants are consistent to avoid compatibility issues; central server construction: build a central server to receive and distribute model update information, configure the dependencies of the TensorFlow Federated library and the Python and TensorFlow library dependencies to the central server. The central server should have sufficient computing resources and storage capabilities to support the processing of large-scale model update information.
[0036] S30: pre-processing the local data of the participants and extracting key features;
[0037] Specifically, by cleaning the local data of the participants, duplicate, erroneous or irrelevant data records are removed; sensitive data in the cleaned local data is desensitized; key features in the desensitized local data are extracted according to anti-fraud business needs; and the extracted feature data is formatted into a format supported by the TFF framework.
[0038] Data cleaning: All participants clean local data to remove duplicate, erroneous or irrelevant data records; Data desensitization: Sensitive data is desensitized to protect user privacy. Desensitization methods may include replacing, encrypting or deleting sensitive fields; Feature extraction: Key features are extracted according to anti-fraud business needs. Feature selection should be based on domain knowledge and data characteristics to ensure that the extracted features are discriminatory for fraudulent behavior; Data formatting: The extracted feature data is formatted into a format supported by the TFF framework, such as TensorFlow tensor or dataset object.
[0039] S40: Build a deep learning model of key features and initialize the deep learning model parameters in the TensorFlow Federated environment;
[0040] Specifically, by determining the model structure of deep learning and building a deep learning model of key features; initializing the deep learning model parameters in the TensorFlow Federated environment; and configuring the hyperparameters of the initialized deep learning model.
[0041] Model selection: Select a suitable deep learning model structure, such as convolutional neural network (CNN), recurrent neural network (RNN) or integrated learning model (such as XGBoost). Model selection should be based on data characteristics and business needs. Model initialization: Initialize deep learning model parameters under the TensorFlow Federated framework. Each participant can share the initial model deep learning parameters or initialize them independently and upload them to the central server for aggregation. The initialized deep learning parameters are usually obtained using random values or algorithms such as mini-batch gradient descent. Model configuration: Configure the hyperparameters of the deep learning model, such as learning rate, batch size, iteration rounds, etc. The selection of hyperparameters should be based on the balance between model performance and computing resources.
[0042] S50: Execute the federated learning task, and update and iterate the deep learning model parameters based on the federated learning task.
[0043] The purpose of federated learning is to ensure that the data of each participant does not leave the local area. In the federated learning system, a virtual shared model can be established through data exchange under an encryption mechanism without violating data privacy protection regulations. This virtual model is like the optimal model established by everyone aggregating data together. However, when establishing the virtual model, the data itself does not move, and it will not leak user privacy or affect data specifications. The established virtual model only serves the local target tasks in the respective areas of the participants. Under such a federated learning mechanism, the identities and status of each participant are the same, and the federated learning system helps everyone establish a "common prosperity" strategy.
[0044] Essence: Federated learning is essentially a distributed machine learning technology, or a machine learning framework.
[0045] Objective: The goal of federated learning is to achieve joint modeling and improve the effectiveness of AI models while ensuring data privacy, security, and legal compliance.
[0046] Definition: When multiple participants Fi (i = 1, ..., N) want to jointly train a machine learning model with their respective local data Di, the traditional approach is to integrate the local data into one party and use the local data D = {Di, i = 1, ..., N} to train and obtain the model MSUM. However, this solution is usually difficult to implement due to legal issues such as privacy and data security. To solve this problem, we introduce federated learning. Federated learning refers to the calculation process in which the participant Fi can perform virtual model training to obtain the model MFED without providing the local data Di, and can ensure that the difference between the effect VFED of the model MFED and the effect VSUM of the model MSUM is small enough, that is, |VFED-VSUM|<δ, where δ is an arbitrarily small positive value.
[0047] There are multiple participants, and the data set Di held by each participant can be represented by a matrix. Each row of the matrix represents a participant, and each column represents a participant feature. Some data sets may also contain labeled data. If you want to build a predictive model for the behavior of the participants, you must have labeled data. We call the key feature X and the label feature Y.
[0048] Taking data involving two parties as an example, data distribution can be divided into the following three cases:
[0049] ● The data features (X1, X2, ...) of the two data sets overlap greatly, while the users (U1, U2, ...) overlap less;
[0050] ● The overlap between the users (U1, U2, ...) of the two data sets is large, while the overlap between the data features (X1, X2, ...) is small;
[0051] ●The overlap between users (U1, U2, …) and data features (X1, X2, …) of the two data sets is relatively small.
[0052] In order to cope with the above three data distribution situations, federated learning is divided into horizontal federated learning, vertical federated learning and federated transfer learning (FTL).
[0053] Horizontal Federated Learning:
[0054] The essence is the union of samples, which is suitable for scenarios where the parties have the same business model but reach different users, that is, there is a lot of feature overlap and little user overlap. For example, banks in different regions have similar businesses (similar features) but different users (different samples).
[0055] Learning process:
[0056] S51: Each participant downloads the latest model from server A;
[0057] S52: Each participant uses local data to train a local model, and uploads the encrypted gradient to the central server A. The central server A aggregates the gradients of each user to update the local model parameters.
[0058] S53: Server A returns the updated deep learning model to each participant;
[0059] S54: Each participant updates its local model.
[0060] Step explanation: In traditional machine learning modeling, the data required for local model training is usually collected in a data center and then the local model is trained and predicted. In horizontal federated learning, it can be seen as sample-based distributed model training, distributing all data to different machines. Each machine downloads the model from the central server, and then uses local data to train the local model, and then returns the data parameters that need to be updated to the central server; the central server aggregates the data parameters returned by each participant, updates the deep learning model, and then feeds back the latest deep learning model to each participant.
[0061] In this process, all participants have the same and complete model, and there is no communication or dependence between the participants. When making predictions, the local models of each participant can also make independent predictions. This process can be regarded as sample-based distributed model training.
[0062] Vertical Federated Learning:
[0063] The essence is the union of features, which is suitable for scenarios with a lot of user overlap and little overlap in data features, such as supermarkets and banks in the same area. The users they reach are all residents of the area (same samples), but their businesses are different (different features).
[0064] Learning process:
[0065] The essence of vertical federated learning is to combine the features of cross-users in different business formats. For example, supermarket A and bank B need to be centralized in a data center in the traditional machine learning modeling process. Then, the features of each user are joined into one piece of data to train the model. This requires that both parties have user intersections (modeling based on the join results) and one party has a label. The learning process is divided into two steps:
[0066] Step 1: Encrypted sample alignment. This is done at the system level, so non-crossing users will not be exposed at the perception level of the participants.
[0067] Step 2: Align samples for model encryption training:
[0068] S55: The third party C sends a public key to A and B to encrypt the data to be transmitted;
[0069] S56: A and B respectively calculate the intermediate results of the data features related to themselves, and encrypt the interaction to obtain their respective gradients and losses;
[0070] S57: A and B respectively calculate their encrypted gradients and add masks to send to C. At the same time, B calculates the encrypted loss and sends it to C.
[0071] S58: C decrypts the gradient and loss and passes them back to A and B. A and B remove the mask and update the local model.
[0072] At present, many machine learning models such as logistic regression models, tree structure models and neural network models have gradually been proven to be able to be built on this federal system.
[0073] Federated transfer learning: Federated transfer learning can be considered when there is little overlap in data features and samples between participants, such as the alliance between banks and supermarkets in different regions. It is mainly applicable to scenarios with deep neural networks as the base model.
[0074] Transfer learning refers to a learning process that uses the similarities between data, tasks, or models to apply the model learned in the source domain to the target domain.
[0075] The core of transfer learning is to find the similarity between the source domain and the target domain. For example, if you know how to play badminton, you can play tennis in a similar way. If you know how to play Go, you can play Gobang in a similar way. Finding tennis and badminton, Gobang and Go is similar to copying. In another embodiment: When driving in mainland China, the driver sits on the left and drives on the right side of the road. This is the basic rule. However, if you drive in the UK, Hong Kong, China and other regions, the driver sits on the right and needs to drive on the left side of the road. So, if you go from mainland China to Hong Kong, how should you quickly adapt to their driving style? The trick is to find the invariant here: no matter which region, the driver is close to the middle of the road. This is the invariant in our driving problem. Finding similarities (invariants) is the core of transfer learning.
[0076] Learning process of federated transfer learning: The steps of federated transfer learning are similar to those of vertical federated learning, except that the intermediate transfer results are different (in fact, the intermediate transfer results of each model are different).
[0077] The relationship between federated learning and privacy protection: The characteristics of federated learning enable it to be used to protect the privacy of user data, but it is still quite different from the commonly used privacy protection theories in the fields of big data and data mining, such as differential privacy and k-anonymity. The principles of federated learning are different from those of traditional privacy protection methods. Federated learning protects user data privacy through parameter exchange under an encryption mechanism. The data and model itself will not be transmitted, and the other party's data cannot be guessed. Therefore, there is no possibility of leakage at the data level, and it does not violate stricter data protection laws such as GDPR. Differential privacy theory, k-anonymity and l-diversity methods add noise to the data, or use generalization methods to blur certain sensitive attributes until a third party cannot distinguish individuals, thereby making the data unrecoverable with a higher probability, thereby protecting user privacy. In comparison, federated learning is more useful for protecting user data privacy.
[0078] The relationship between federated learning and distributed machine learning: The multi-party joint training in horizontal federated learning is partially similar to distributed machine learning. Distributed machine learning covers many aspects, including distributed storage of training data in machine learning, distributed operation of computing tasks, and distributed publishing of model results. Parameter server is a typical example of distributed machine learning. As a tool to accelerate the training process of machine learning models, it stores data on distributed working nodes and allocates data distribution and computing resources through a central scheduling node to obtain the final training model more efficiently. The working nodes in horizontal federated learning represent the data owners of model training. They have full autonomy over local data and can decide when to join federated learning for modeling. Federated learning faces a more complex learning environment. Federated learning emphasizes the data privacy protection of data owners during model training, which is an effective measure to deal with data privacy protection.
[0079] The relationship between federated learning and federated databases: A federated database system is a system that integrates multiple different unit databases and manages the integrated whole. It was proposed to achieve interoperability between multiple independent databases. The federated database system often uses distributed storage for the databases of each participant, and in practice the data in the databases of each participant is heterogeneous. Therefore, it has many similarities with federated learning in terms of data type and storage method.
[0080] Specifically, the local data of the participants are trained based on the deep learning model, and a local model is constructed; the update information of the local model is obtained during the training process; the update information of the deep learning model is encrypted and uploaded to the central server, which aggregates the uploaded update information to form new global model parameters; the update information aggregated by the central server is distributed to the participants, and the participants use the new global model parameters to update the local model and start the next round of training.
[0081] Local training: Each participant uses its own data set to train the model locally. The training process includes steps such as forward propagation, loss calculation, back propagation, and parameter update. Model update information calculation: During the local training process, the gradient information or update information of the model is calculated, which will be used to update the global model parameters. Encrypted upload: The calculated model update information is encrypted to ensure privacy and security during transmission, and the encrypted update information is uploaded to the central server. Aggregate update: The central server receives the update information uploaded by each participant and aggregates it. The aggregation method may include weighted average, weighted average, and regularization. The aggregated update information forms new global model parameters. Distribute update: The central server distributes the aggregated update information back to each participant. Each participant updates the local model with the new global model parameters and starts the next round of training. Multiple rounds of iteration: Repeat the above steps until the model converges or reaches the preset training rounds. During the iteration process, the hyperparameters can be adjusted or the model structure can be optimized according to the model performance evaluation results.
[0082] Among them, after executing the federated learning task and updating and iterating the deep learning model parameters based on the federated learning task, the following steps are also included:
[0083] Use an independent validation dataset to evaluate the jointly trained global model; analyze the evaluation results and identify the performance of the global model under different data types; and optimize the global model based on the evaluation results.
[0084] Model evaluation: Use an independent validation set or test set to evaluate the global model after joint training. Evaluation indicators include accuracy, recall, F1 score and other indicators that can fully reflect the performance of the global model. The evaluation results are used to verify the generalization ability and robustness of the global model. Performance analysis: Conduct in-depth analysis of the evaluation results to identify the performance of the global model in specific scenarios or data types. The analysis results help to find problems and potential improvement directions of the global model. Model optimization: Optimize the global model based on the evaluation results. Optimization measures may include adjusting the global model structure (such as increasing the number of layers, changing the activation function, etc.), modifying hyperparameters (such as learning rate, batch size, etc.) or introducing regularization terms. The optimized global model is retrained and evaluated to verify the improvement effect. Deployment and application: Deploy the optimized global model to actual application scenarios, such as the bank's anti-fraud system. During the deployment process, it is necessary to ensure the compatibility and stability of the global model with the existing system. At the same time, continuously monitor the performance of the global model in actual applications and adjust and optimize it as needed.
[0085] The present invention provides a bank anti-fraud model optimization method based on TensorFlow Federated, which first determines the data standards of the participants and the participants; builds a TensorFlow Federated environment based on the data standards of the participants; pre-processes the local data of the participants and extracts key features; builds a deep learning model of the key features, and initializes the deep learning model parameters in the TensorFlow Federated environment; executes a federated learning task, and updates and iterates the deep learning model parameters based on the federated learning task. The accuracy of the anti-fraud model is improved: cross-institutional data collaborative modeling is achieved through federated learning technology, which enriches the model training samples and feature dimensions and significantly improves the accuracy of the anti-fraud model; protects data privacy and security: during the entire federated learning process, the original data is always retained locally, and only the model update information rather than the original data is transmitted, which ensures the privacy and security of the data provider and complies with the requirements of relevant laws and regulations; breaks the data island phenomenon: solves the data island problem in traditional anti-fraud methods, and realizes cross-industry and cross-institutional data sharing and value maximization.
[0086] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.
[0087] In one embodiment, a bank anti-fraud model optimization device based on TensorFlow Federated is provided, and the bank anti-fraud model optimization device based on TensorFlow Federated corresponds one-to-one to the bank anti-fraud model optimization method based on TensorFlow Federated in the above embodiment. Figure 3 As shown, the bank anti-fraud model optimization device based on TensorFlow Federated includes a determination module 101, a construction module 102, an extraction module 103, a deep learning module 104, and an update iteration module 105. The functional modules are described in detail as follows:
[0088] A determination module 101 is used to determine the data standards of the participants and the participants;
[0089] Building module 102, for building a TensorFlow Federated environment based on the data standards of the participants;
[0090] The extraction module 103 is used to pre-process the local data of the participants and extract key features;
[0091] A deep learning module 104 is used to construct a deep learning model of key features and initialize deep learning model parameters in a TensorFlowFederated environment;
[0092] The update iteration module 105 is used to execute the federated learning task and update and iterate the deep learning model parameters based on the federated learning task.
[0093] In one embodiment, the building module 102 is specifically used to:
[0094] Install Python and TensorFlow libraries on the participant’s server based on the participant’s data standards and configure necessary dependencies;
[0095] Install the TensorFlow Federated library in other package management tools and configure dependencies;
[0096] Build a central server and configure the dependencies of the TensorFlow Federated library and the Python and TensorFlow library dependencies to the central server.
[0097] In one embodiment, the extraction module 103 is specifically configured to:
[0098] Clean the local data of the participants to remove duplicate, erroneous or irrelevant data records;
[0099] Desensitize sensitive data in cleaned local data;
[0100] Extract key features from desensitized local data based on anti-fraud business needs;
[0101] Format the extracted feature data into a format supported by the TFF framework.
[0102] In one embodiment, the deep learning module 104 is specifically configured to:
[0103] Determine the model structure of deep learning and build a deep learning model of key features
[0104] Initialize deep learning model parameters in the TensorFlow Federated environment;
[0105] Configure the hyperparameters of the initialized deep learning model.
[0106] In one embodiment, the update iteration module 105 is specifically configured to:
[0107] Train the local data of the participants based on the deep learning model and build a local model;
[0108] Get updated information of the local model during training;
[0109] Encrypt the update information of the deep learning model and upload the encrypted update information to the central server, which aggregates the uploaded update information to form new global model parameters.
[0110] The updated information aggregated by the central server is distributed to the participants. The participants use the new global model parameters to update the local model and start the next round of training.
[0111] In one embodiment, the device is further used for:
[0112] Evaluate the jointly trained global model using an independent validation dataset;
[0113] Analyze the evaluation results and identify the performance of the global model under different data types;
[0114] The global model is optimized based on the evaluation results.
[0115] The present invention provides a bank anti-fraud model optimization device based on TensorFlow Federated, which first determines the data standards of the participants and the participants; builds a TensorFlow Federated environment based on the data standards of the participants; pre-processes the local data of the participants and extracts key features; builds a deep learning model of the key features, and initializes the deep learning model parameters in the TensorFlow Federated environment; executes a federated learning task, and updates and iterates the deep learning model parameters based on the federated learning task. The accuracy of the anti-fraud model is improved: cross-institutional data collaborative modeling is achieved through the federated learning task technology, which enriches the model training samples and feature dimensions and significantly improves the accuracy of the anti-fraud model; protects data privacy and security: during the entire federated learning task process, the original data is always retained locally, and only the model update information rather than the original data is transmitted, which ensures the privacy and security of the data provider and complies with the requirements of relevant laws and regulations; breaks the data island phenomenon: solves the data island problem in traditional anti-fraud methods, and realizes cross-industry and cross-institutional data sharing and value maximization.
[0116] For the specific definition of the bank anti-fraud model optimization device based on TensorFlow Federated, please refer to the definition of the bank anti-fraud model optimization method based on TensorFlow Federated above, which will not be repeated here. Each module in the above-mentioned bank anti-fraud model optimization device based on TensorFlow Federated can be implemented in whole or in part through software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.
[0117] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, it implements a function or step on the server side of a bank anti-fraud model optimization method based on TensorFlow Federated.
[0118] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 5 As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps on the client side of a bank anti-fraud model optimization method based on TensorFlow Federated
[0119] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the following steps when executing the computer program:
[0120] Determine the data standards for participants and participants;
[0121] Build a TensorFlow Federated environment based on the data standards of the participants;
[0122] Preprocess the local data of the participants and extract key features;
[0123] Build a deep learning model of key features and initialize the deep learning model parameters in the TensorFlow Federated environment;
[0124] Execute federated learning tasks and update and iterate deep learning model parameters based on federated learning tasks.
[0125] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant descriptions on the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.
[0126] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).
[0127] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.
[0128] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.
Claims
1. A bank anti-fraud model optimization method based on TensorFlow Federated, characterized in that: include: Determine the data standards for participants and participants; Build a TensorFlow Federated environment based on the data standards of the participants; Preprocess the local data of the participants and extract key features; Build a deep learning model of key features and initialize the deep learning model parameters in the TensorFlow Federated environment; Execute federated learning tasks and update and iterate deep learning model parameters based on federated learning tasks.
2. The bank anti-fraud model optimization method based on TensorFlow Federated as claimed in claim 1, characterized in that: The data standards include: standard format for data exchange, encryption protocol type, data fields, goals and evaluation indicators.
3. The bank anti-fraud model optimization method based on TensorFlow Federated as claimed in claim 2, characterized in that: The steps of building a TensorFlow Federated environment based on the data standards of the participants include: Install Python and TensorFlow libraries on the participant’s server based on the participant’s data standards and configure necessary dependencies; Install the TensorFlow Federated library in other package management tools and configure dependencies; Build a central server and configure the dependencies of the TensorFlow Federated library and the Python and TensorFlow library dependencies to the central server.
4. The bank anti-fraud model optimization method based on TensorFlow Federated as claimed in claim 2, characterized in that: The steps of preprocessing the local data of the participants and extracting key features include: Clean the local data of the participants to remove duplicate, erroneous or irrelevant data records; Desensitize sensitive data in cleaned local data; Extract key features from desensitized local data based on anti-fraud business needs; Format the extracted feature data into a format supported by the TFF framework.
5. The bank anti-fraud model optimization method based on TensorFlow Federated as claimed in claim 2, characterized in that: The steps of constructing a deep learning model of key features and initializing deep learning model parameters in a TensorFlow Federated environment include: Determine the model structure of deep learning and build a deep learning model of key features; Initialize deep learning model parameters in the TensorFlow Federated environment; Configure the hyperparameters of the initialized deep learning model.
6. The bank anti-fraud model optimization method based on TensorFlow Federated as claimed in claim 2, characterized in that: The steps of executing the federated learning task and updating and iterating the deep learning model parameters based on the federated learning task include: Train the local data of the participants based on the deep learning model and build a local model; Get updated information of the local model during training; Encrypt the update information of the deep learning model and upload the encrypted update information to the central server, which aggregates the uploaded update information to form new global model parameters. The updated information aggregated by the central server is distributed to the participants. The participants use the new global model parameters to update the local model and start the next round of training.
7. The bank anti-fraud model optimization method based on TensorFlow Federated as claimed in claim 1, characterized in that: The step of executing the federated learning task and updating and iterating the deep learning model parameters based on the federated learning task further includes: Evaluate the jointly trained global model using an independent validation dataset; Analyze the evaluation results and identify the performance of the global model under different data types; The global model is optimized based on the evaluation results.
8. A bank anti-fraud model optimization device based on TensorFlow Federated, characterized in that: include: A determination module is used to determine the data standards of the participants and the participants; Build a module for building a TensorFlow Federated environment based on the data standards of the participants; The extraction module is used to pre-process the local data of the participants and extract key features; The deep learning module is used to build a deep learning model of key features and initialize the deep learning model parameters in the TensorFlow Federated environment; The update iteration module is used to execute federated learning tasks and update and iterate the parameters of the deep learning model based on the federated learning tasks.
9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the bank anti-fraud model optimization method based on TensorFlow Federated are implemented as described in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the bank anti-fraud model optimization method based on TensorFlowFederated are implemented as described in any one of claims 1 to 7.