Federated Learning-Based Neural Network Modeling Methods and Related Equipment

Abstract

This disclosure provides a neural network modeling method and related equipment based on federated learning, relating to the field of artificial intelligence technology. The method includes distributing model information of a neural network model to be trained to a first participant and a second participant. The neural network model includes an input layer, a hidden layer, and an output layer. The first participant trains the input layer and the hidden layer, and the second participant trains the output layer. The method also includes receiving a first training result returned by the first participant and a second training result returned by the second participant; and aggregating the first and second training results into a trained neural network model. This disclosure performs neural network model segmentation and joint modeling based on data characteristics, protecting the privacy of all parties while utilizing data from multiple sources for training.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence technology, and in particular to a neural network modeling method based on federated learning and related equipment. Background Technology

[0002] With the development and advancement of artificial intelligence technology, neural network models are being researched and applied in various fields, such as smart homes and wearable devices. Federated learning is an emerging foundational AI technology applicable to neural network models. Its design goal is to enable efficient machine learning among multiple participants or computing nodes while ensuring information security during big data exchange, protecting terminal and personal data privacy, and guaranteeing legal compliance.

[0003] Current neural network model training mainly focuses on a single data source. When training with data from multiple sources, it can lead to the leakage of privacy data from all parties. Even with existing federated neural network modeling methods, there are still significant limitations, making them unsuitable for vertical scenarios where feature data and label data are separated.

[0004] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] This disclosure provides a neural network modeling method and related device based on federated learning, which at least to some extent overcomes the technical problem of privacy data leakage when calling data from multiple parties for neural network training in related technologies.

[0006] Other features and advantages of this disclosure will become apparent from the following detailed description, or may be learned in part from practice of this disclosure.

[0007] According to one aspect of this disclosure, a method for modeling neural networks based on federated learning is provided, comprising:

[0008] The model information of the neural network model to be trained is sent to the first participant and the second participant. The neural network model includes an input layer, a hidden layer and an output layer. The first participant is used to train the input layer and the hidden layer, and the second participant is used to train the output layer.

[0009] Receive the first training result returned by the first participant and the second training result returned by the second participant;

[0010] The first training result and the second training result are aggregated into a trained neural network model.

[0011] In one embodiment of this disclosure, the first participant is a feature data provider, and the second participant is a tag data provider.

[0012] In one embodiment of this disclosure, receiving a first training result returned by the first participant and a second training result returned by the second participant specifically includes:

[0013] The system receives the neural network structure and neuron weights of the input and hidden layers after training by the first participant, as well as the neural network structure and neuron weights of the output layer after training by the second participant.

[0014] In one embodiment of this disclosure, before aggregating the first training result and the second training result into a trained neural network model, the method further includes:

[0015] Obtain the connection information between the first participant and the second participant, wherein the connection information includes the neuronal connection relationships and connection weights between the hidden layer of the first participant and the output layer of the second participant.

[0016] In one embodiment of this disclosure, the first training result and the second training result are aggregated into a trained neural network model, specifically including:

[0017] Based on the connection information, the first training result and the second training result are concatenated and spliced ​​together to aggregate into a trained neural network model.

[0018] In one embodiment of this disclosure, the model information of the neural network model to be trained is distributed to the first participant and the second participant, specifically including:

[0019] The input layer and hidden layer information of the neural network model to be trained are sent to the first participant, and the output layer information of the neural network model to be trained is sent to the second participant. The hidden layer of the first participant and the output layer of the second participant are encrypted and connected in a differential privacy manner.

[0020] In one embodiment of this disclosure, the method further includes:

[0021] When the number of the first participants is not unique, the hidden layers of multiple first participants are respectively encrypted with the output layer of the second participants in a differential privacy manner.

[0022] According to another aspect of this disclosure, a neural network modeling apparatus based on federated learning is provided, comprising:

[0023] The model distribution module is used to distribute the model information of the neural network model to be trained to the first participant and the second participant. The neural network model includes an input layer, a hidden layer and an output layer. The first participant is used to train the input layer and the hidden layer, and the second participant is used to train the output layer.

[0024] The parameter acquisition module is used to receive the first training result returned by the first participant and the second training result returned by the second participant.

[0025] The model aggregation module is used to aggregate the first training result and the second training result into a trained neural network model.

[0026] According to another aspect of this disclosure, an electronic device is provided, comprising:

[0027] Processor; and

[0028] A memory for storing executable instructions of the processor; wherein the processor is configured to execute the above-described federated learning-based neural network modeling method by executing the executable instructions.

[0029] According to another aspect of this disclosure, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the above-described federated learning-based neural network modeling method.

[0030] The embodiments of this disclosure provide a federated learning-based neural network modeling method and related equipment. This method segments the neural network model according to data characteristics in scenarios where feature data and label data belong to different parties. The feature data provider trains the input and hidden layers of the neural network model, while the label data provider trains the output layer of the neural network model. The training results of the two parties are then used to jointly model the model, which can protect the privacy data of each party while using data from multiple parties for training.

[0031] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0032] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0033] Figure 1 A schematic diagram illustrating the framework of federated learning in related technologies is shown;

[0034] Figure 2 This shows a data feature map of vertical federated learning in related technologies;

[0035] Figure 3 A schematic diagram of the execution device for the neural network modeling method based on federated learning in an embodiment of this disclosure is shown;

[0036] Figure 4 A flowchart of a neural network modeling method based on federated learning is shown in an embodiment of this disclosure;

[0037] Figure 5 This diagram illustrates the execution process of a neural network modeling method based on federated learning in an embodiment of this disclosure.

[0038] Figure 6 This diagram illustrates a neural network modeling apparatus based on federated learning, as shown in an embodiment of this disclosure; and

[0039] Figure 7 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown. Detailed Implementation

[0040] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that this disclosure will be more comprehensive and complete, and will fully convey the concept of the exemplary embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0041] Furthermore, the accompanying drawings are merely illustrative of this disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0042] To facilitate understanding, the following is an explanation of some of the terms used in this disclosure:

[0043] Federated learning is a distributed machine learning framework with privacy protection and secure encryption technology. It aims to enable decentralized participants to collaborate on training machine learning models without disclosing their private data to other participants.

[0044] like Figure 1As shown, the training process of the classic federated learning framework can be simply summarized as follows:

[0045] Coordinator (e.g.) Figure 1 The central server establishes a basic model and informs all participating parties of the model's basic structure and parameters (e.g., ...). Figure 1 Participants A and B); each participant utilizes local data (e.g., Figure 1 The model is trained on samples A and B, and the results are returned to the coordinator. Parameters are exchanged between the participants and between the samples through a homomorphic encryption mechanism. The coordinator summarizes the models of all participants to build a more accurate global model, thereby improving the overall model performance and effectiveness.

[0046] The federated learning framework incorporates various technologies, such as traditional machine learning model training techniques, algorithms for coordinating parameter integration, communication technologies for efficient transmission between coordinators and participants, and privacy-preserving encryption technologies. Furthermore, the federated learning framework includes an incentive mechanism, allowing all data holders to participate and ensuring that benefits are universally shared.

[0047] Based on the different ways in which each party contributes to the dataset, federated learning can be specifically divided into horizontal federated learning, vertical federated learning, and federated transfer learning, with each sub-technique corresponding to different scenarios.

[0048] Horizontal federated learning is suitable for scenarios where data holders have similar business types, possess numerous user features, and have little or no overlap in their user spaces. For example, different shopping malls in different regions may have largely similar customer shopping information, but their user demographics differ. Horizontal federated learning is guided by the feature dimensions of the data, extracting the parts where the participants share similar features but the users are not entirely identical, and then jointly training them. In this process, by jointly training samples from each participant, the training sample space is expanded, thereby improving the model's accuracy and generalization ability.

[0049] Research on neural network models based on federated learning in related technologies is mostly based on horizontal federated learning.

[0050] Vertical federated learning is suitable for scenarios where there is significant overlap in user spaces among the participants, but little or no overlap in feature spaces. For example, banks and shopping malls in a region may have overlapping user spaces due to their similar geographical locations, but their user features differ significantly because of their different business types. Vertical federated learning is data alignment-oriented, extracting the parts of the participants' users that are identical but have slightly different features for joint training. Therefore, during joint training, it is necessary to first align the samples of each participant's data to obtain overlapping user data, and then each participant trains on the selected dataset. Furthermore, to ensure the security of non-overlapping data, sample alignment is performed at the system level, and each participant only has a model trained on its local data.

[0051] For details on the specific data composition of vertical federated learning, please refer to [link / reference]. Figure 2 This means that there is significant sample overlap between data A and data B, but relatively little feature overlap. The neural network modeling method based on federated learning provided in this disclosure applies the aforementioned longitudinal federated learning scenario.

[0052] Federated transfer learning complements horizontal and vertical federated learning, and is suitable for scenarios where the user spaces and feature spaces of the participating parties have little overlap. For example, banks and shopping malls in different regions have little overlap in their user spaces and virtually no overlap in their feature spaces. In this scenario, horizontal federated learning may produce a worse model than training alone, while vertical federated learning may result in negative transfer. Federated transfer learning transfers the model learned in the source domain to the target domain based on the similarity between the data or models of the participating parties. It mostly uses labels from the source domain to predict the accuracy of labels in the target domain.

[0053] Before introducing the technical solution of this disclosure, we will first use a specific application scenario of this disclosure to introduce the problems existing in the technology and the technical conception process of this disclosure.

[0054] In practical applications, due to the holistic nature of neural network models, current neural network model training mainly focuses on a single data source, making distributed model training with multi-source data difficult. Existing federated neural network modeling methods are mostly based on horizontal federated learning scenarios, with limited coverage of vertical scenarios where feature data and label data are separated. For example, a bank and an e-commerce platform may have user groups that likely encompass a large portion of the local population, resulting in significant user overlap. However, since the bank's records may contain user feature data, while the e-commerce platform may retain user label data, federated neural networks in related technologies struggle to effectively integrate these two sets of data when using neural networks to predict user consumption habits. Furthermore, federated neural networks in related technologies often rely on homomorphic encryption for privacy protection, requiring substantial computation and key negotiation and transmission processes, which limits their application with large amounts of image data.

[0055] To address the aforementioned issues, this disclosure distributes the model information of the neural network model to be trained to the feature data provider and the label data provider. The neural network model includes an input layer, a hidden layer, and an output layer. The feature data provider is used to train the input layer and the hidden layer, and the label data provider is used to train the output layer. The training results returned by the feature data provider and the label data provider are received, and then the training results are aggregated into a trained neural network model.

[0056] Based on the above scheme, this disclosure protects the privacy data of various parties, such as feature data providers and tag data providers, while enabling multiple feature data providers and tag providers to jointly model neural networks, thus breaking down data silos among multiple data partners.

[0057] It should be noted that the execution device for the federated learning-based neural network modeling method provided in this disclosure can be a terminal device, a server, or other electronic device with computing capabilities, and this disclosure does not limit it.

[0058] Terminal devices can be various electronic devices, including but not limited to smartphones, tablets, laptops, desktop computers, wearable devices, augmented reality devices, virtual reality devices, etc.

[0059] Optionally, the client for the application installed on different terminal devices can be the same, or the client for the same type of application based on different operating systems. Depending on the terminal platform, the specific form of the application client can also differ; for example, the application client can be a mobile client, a PC client, etc.

[0060] A server can be a server that provides various services, such as a backend management server that supports the devices operated by users through terminal devices. The backend management server can analyze and process received requests and other data, and then feed the processing results back to the terminal device.

[0061] Optionally, the server can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The terminal can be a smartphone, tablet, laptop, desktop computer, smart speaker, smartwatch, etc., but is not limited to these. The terminal and server can be directly or indirectly connected via wired or wireless communication, which is not limited herein.

[0062] For example, Figure 3 A schematic diagram of an execution device for a neural network modeling method based on federated learning according to an embodiment of the present disclosure is shown.

[0063] like Figure 3 As shown, the execution device may include: a processor 301, such as a CPU; a network interface 304; a user interface 303; a memory 305; and a communication bus 302. The communication bus 302 is used to enable communication between these components. The user interface 303 may include a display screen or an input unit such as a keyboard; optionally, the user interface 303 may also include a standard wired interface or a wireless interface. The network interface 304 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 305 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 305 may also be a storage device independent of the aforementioned processor 301.

[0064] The memory 305, serving as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a federated learning-based neural network modeling program. The operating system is a program that manages and controls the device's hardware and software resources, supporting the operation of the federated learning-based neural network modeling program and other software or programs.

[0065] Those skilled in the art will understand that Figure 3The device structure shown does not constitute a limitation on the execution device of the federated learning-based neural network modeling method, and may include more or fewer components than shown, or combine certain components, or have different component arrangements. This disclosure does not limit the scope of the invention.

[0066] The following detailed description of this exemplary implementation method is provided in conjunction with the accompanying drawings and embodiments.

[0067] First, this disclosure provides a neural network modeling method based on federated learning, which can be executed by any electronic device with computing power.

[0068] Figure 4 This illustration shows a flowchart of a neural network modeling method based on federated learning, as shown in an embodiment of the present disclosure. Figure 4 As shown, the neural network modeling method based on federated learning provided in this embodiment can be executed by the central server of federated learning, and includes the following steps:

[0069] S402, the model information of the neural network model to be trained is sent to the first participant and the second participant. The neural network model includes an input layer, a hidden layer and an output layer. The first participant is used to train the input layer and the hidden layer, and the second participant is used to train the output layer.

[0070] It should be noted that the first participant is the feature data provider, and the second participant is the label data provider.

[0071] It should be noted that the neural network model in this disclosure can be a neural network model applicable to any field, wherein the model information may include all model information or may include some model information, and any model information distribution method that can achieve the above effect is applicable to the embodiments of this disclosure.

[0072] In one embodiment of this disclosure, the input layer information and hidden layer information of the neural network model to be trained can be sent to a first participant, and the output layer information of the neural network model to be trained can be sent to a second participant, wherein the hidden layer of the first participant and the output layer of the second participant are connected.

[0073] In another embodiment of this disclosure, the output layer, hidden layer, and all information of the neural network model to be trained can be sent to the first participant and the second participant respectively, and the hidden layer of the first participant and the output layer of the second participant can be connected.

[0074] Those skilled in the art will understand that the first participant is used to train the input layer and hidden layer of the neural network model, and the second participant is used to train the output layer of the neural network. The specific structure of the neural network model and the connection method of the neurons can be adjusted according to actual needs, and this disclosure does not limit this.

[0075] Optionally, the hidden layer of the first participant and the output layer of the second participant are encrypted and connected using differential privacy.

[0076] Specifically, in this embodiment of the disclosure, the forward propagation excitation value weighted calculation result and the backpropagation gradient between the hidden layer of the first participant and the output layer of the second participant can be increased with noise through differential privacy to avoid gradient backpropagation.

[0077] It should be noted that differential privacy achieves its privacy protection effect by adding noise to the data through anonymization, perturbation, and obfuscation. Furthermore, data processed using differential privacy does not affect the output of any subsequent calculations. The implementation of differential privacy is well-known to those skilled in the art, and will not be elaborated upon here.

[0078] Those skilled in the art will understand that differential privacy assumes that an attacker can obtain all other record information except the target record. The sum of this information can be understood as the maximum background knowledge that the attacker can possess. Under this strong assumption, differential privacy protection does not need to consider any possible background knowledge possessed by the attacker.

[0079] Based on the above principles, this disclosure solves the problem of cross-domain gradient privacy protection during the backpropagation training process of longitudinal federated learning neural networks through differential privacy technology. When gradient backpropagation is involved between cross-domain layers, differential privacy processing that conforms to the privacy budget needs to be added to the backpropagation gradient to achieve privacy protection of gradient data between cross-domains.

[0080] In one embodiment of this disclosure, when the number of first participants is not unique, the hidden layers of multiple first participants are respectively encrypted with the output layer of the second participants in a differential privacy manner.

[0081] S404, Receive the first training result returned by the first participant and the second training result returned by the second participant;

[0082] Specifically, this disclosure differentiates the allocation of neural network model layers based on whether the data provider is a feature data provider or a label data provider. The feature data provider is responsible for training the input and hidden layers of the neural network, while the label data provider is responsible for training the output layer. The training results are received and aggregated through a central server.

[0083] More specifically, the central server can receive the neural network structure and neuron weights of the input and hidden layers returned by the first participant after training, as well as the neural network structure and neuron weights of the output layer returned by the second participant after training.

[0084] S406, aggregate the first training result and the second training result into a trained neural network model.

[0085] Specifically, before aggregating the first training result and the second training result into a trained neural network model, embodiments of this disclosure may also obtain connection information between the first participant and the second participant, wherein the connection information includes the neuron connection relationship and connection weight between the hidden layer of the first participant and the output layer of the second participant.

[0086] More specifically, in embodiments of this disclosure, the first training result and the second training result can be concatenated and spliced ​​together according to the above connection information to aggregate into a trained neural network model.

[0087] In one embodiment of this disclosure, S404 and S406 can be omitted, that is, it is not necessary to use the central server of federated learning to obtain the first training result and the second training result and complete the aggregation.

[0088] Specifically, in the actual application of the trained model, the first participant in this embodiment can continue to serve as the input layer and hidden layer of the neural network model, and the second participant can continue to serve as the output layer of the neural network model. At this time, the hidden layer of the first participant still maintains a connection with the output layer of the second participant, thereby forming a complete neural network model.

[0089] Those skilled in the art will understand that the trained neural network model in the embodiments of this disclosure can be invoked in a distributed manner.

[0090] Optionally, in this disclosure, cross-domain communication is protected by an encrypted channel using the Transport Layer Security (TLS) protocol to prevent man-in-the-middle attacks and channel sniffing.

[0091] In practical applications, the methods provided in this disclosure can be applied to fields such as epidemic prevention and control, financial risk control, public security and public opinion monitoring, and medical imaging diagnosis and treatment, thereby achieving the corresponding technical effects.

[0092] For example, in epidemic prevention and control, the method provided in this disclosure can achieve joint prevention and control of the epidemic supported by big data while ensuring the protection of personal information. Through data privacy sharing and circulation, it can achieve joint investigation and precise positioning across institutions, while fully protecting the privacy and security of residents.

[0093] In financial risk control, the method provided in this disclosure can help financial institutions train a highly accurate user financial risk scoring model to predict potential financial defaults and fraudulent activities, effectively reduce bad debt rates, and provide risk warnings.

[0094] In public security intelligence analysis, the method provided in this disclosure can effectively predict and prevent public safety incidents through the fusion analysis of telecommunications service data, internet behavior data, and social media sentiment. It also achieves multi-party, two-way privacy protection in public security intelligence analysis.

[0095] In medical imaging diagnosis and treatment, the method provided in this disclosure can achieve intelligent diagnosis and treatment system by jointly modeling expert medical diagnosis and treatment label data and related medical imaging data of the patients.

[0096] Figure 5 This is a schematic diagram illustrating the execution process of the federated learning-based neural network modeling method provided in this embodiment of the disclosure.

[0097] like Figure 5 As shown, this embodiment of the disclosure provides a parameter server (i.e., the central server for federated learning), a first participant (providing image data), and a second participant (providing label data). Based on the characteristics of the data held by different participants, the participants are distinguished into feature data (image data in the figure) providers and label data providers. The feature data provider is used to establish the neural network hierarchy structure of the input and hidden layers, while the label data provider is used to establish the neural network hierarchy structure of the output layer.

[0098] Specifically, this disclosure embodiment performs model segmentation and joint modeling based on data features to perform neural network hierarchical segmentation in scenarios where feature data and label data belong to different participants.

[0099] In one embodiment of this disclosure, the parameter server and multiple participating party model training containers uniformly constitute a distributed machine learning framework. This enables multiple parties to train user models locally based on local data, while simultaneously interacting with the parameter server to achieve secure aggregation of local models from multiple parties and construct a global model.

[0100] Specifically, the embodiments of this disclosure solve the problem of difficult data interconnection between tag data providers and feature data providers, break down data silos, and promote unified management and secure flow of data at the upper layer.

[0101] In one embodiment of this disclosure, the weighted calculation result of the forward propagation incentive value and the backpropagation gradient between the first participant and the second participant are both increased with noise through differential privacy to avoid gradient backpropagation.

[0102] Optionally, cross-domain communication in this disclosure is protected by a secure TLS encryption channel.

[0103] This disclosure combines differential privacy technology with longitudinal federated neural network learning technology, which fully leverages the advantage of federated learning data not leaving the domain, while simultaneously addressing the risk of privacy leakage during gradient transmission through differential privacy.

[0104] Those skilled in the art will know that Figure 5 The number of participants is merely illustrative; the number may be more or fewer depending on actual needs. This disclosure does not limit the number of participants.

[0105] Based on the same inventive concept, this disclosure also provides a neural network modeling device based on federated learning, as described in the following embodiments. Since the principle by which this device solves the problem is similar to that of the method embodiments described above, the implementation of this device embodiment can refer to the implementation of the method embodiments described above, and repeated details will not be elaborated further.

[0106] Figure 6 This illustration shows a schematic diagram of a neural network modeling apparatus based on federated learning, as shown in an embodiment of this disclosure. Figure 6 As shown, the device 600 includes:

[0107] The model distribution module 601 is used to distribute the model information of the neural network model to be trained to the first participant and the second participant. The neural network model includes an input layer, a hidden layer and an output layer. The first participant is used to train the input layer and the hidden layer, and the second participant is used to train the output layer.

[0108] The parameter acquisition module 602 is used to receive the first training result returned by the first participant and the second training result returned by the second participant;

[0109] The model aggregation module 603 is used to aggregate the first training result and the second training result into a trained neural network model.

[0110] Optionally, the first participant is the feature data provider, and the second participant is the label data provider.

[0111] Optionally, when the number of first participants is not unique, the hidden layers of multiple first participants are respectively encrypted with the output layer of the second participants in a differential privacy manner.

[0112] Optionally, the parameter acquisition module 602 described above is specifically used for:

[0113] The system receives the neural network structure and neuron weights of the input and hidden layers after training by the first participant, as well as the neural network structure and neuron weights of the output layer after training by the second participant.

[0114] Optionally, the model aggregation module 603 described above can also be used for:

[0115] Obtain the connection information between the first participant and the second participant. The connection information includes the neuronal connection relationships and connection weights between the hidden layer of the first participant and the output layer of the second participant.

[0116] Optionally, the above-mentioned model aggregation module 603 is specifically used for:

[0117] Based on the connection information, the first training result and the second training result are concatenated and spliced ​​together to form a trained neural network model.

[0118] Optionally, the model distribution module 601 described above is specifically used for:

[0119] The input layer and hidden layer information of the neural network model to be trained are sent to the first participant, and the output layer information of the neural network model to be trained is sent to the second participant. The hidden layer of the first participant and the output layer of the second participant are encrypted and connected in a differential privacy manner.

[0120] Those skilled in the art will understand that the first participant is used to train the input layer and hidden layer of the neural network model, and the second participant is used to train the output layer of the neural network. The specific structure of the neural network model and the connection method of the neurons can be adjusted according to actual needs, and this disclosure does not limit this.

[0121] Optionally, the hidden layer of the first participant and the output layer of the second participant are encrypted and connected using differential privacy.

[0122] Specifically, in this embodiment of the disclosure, the forward propagation excitation value weighted calculation result and the backpropagation gradient between the hidden layer of the first participant and the output layer of the second participant can be increased with noise through differential privacy to avoid gradient backpropagation.

[0123] Based on the above principles, this disclosure solves the problem of cross-domain gradient privacy protection during the backpropagation training process of longitudinal federated learning neural networks through differential privacy technology. When gradient backpropagation is involved between cross-domain layers, differential privacy processing that conforms to the privacy budget needs to be added to the backpropagation gradient to achieve privacy protection of gradient data between cross-domains.

[0124] It should be noted that the above embodiments of the federated learning-based neural network modeling device are only illustrated by the division of the above functional modules during neural network modeling. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the federated learning-based neural network modeling device and the federated learning-based neural network modeling method embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0125] Those skilled in the art will understand that various aspects of this disclosure can be implemented as a system, method, or program product. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0126] The following reference Figure 7 To describe an electronic device 700 according to such an embodiment of the present disclosure. Figure 7 The electronic device 700 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments disclosed herein.

[0127] like Figure 7 As shown, the electronic device 700 is manifested in the form of a general-purpose computing device. The components of the electronic device 700 may include, but are not limited to: at least one processing unit 710, at least one storage unit 720, and a bus 730 connecting different system components (including storage unit 720 and processing unit 710).

[0128] The storage unit stores program code that can be executed by the processing unit 710, causing the processing unit 710 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. For example, the processing unit 710 can perform the following steps of the above method embodiment: distributing model information of a neural network model to be trained to a first participant and a second participant, wherein the neural network model includes an input layer, a hidden layer, and an output layer, the first participant is used to train the input layer and the hidden layer, and the second participant is used to train the output layer; receiving a first training result returned by the first participant and a second training result returned by the second participant; and aggregating the first training result and the second training result into a trained neural network model.

[0129] Optionally, the first participant is the feature data provider, and the second participant is the label data provider.

[0130] Optionally, the above-mentioned processing unit 710 may specifically perform:

[0131] The system receives the neural network structure and neuron weights of the input and hidden layers after training by the first participant, as well as the neural network structure and neuron weights of the output layer after training by the second participant.

[0132] Optionally, the processing unit 710 may also perform:

[0133] Obtain the connection information between the first participant and the second participant. The connection information includes the neuronal connection relationships and connection weights between the hidden layer of the first participant and the output layer of the second participant.

[0134] Optionally, the above-mentioned processing unit 710 may specifically perform:

[0135] Based on the connection information, the first training result and the second training result are concatenated and spliced ​​together to form a trained neural network model.

[0136] Optionally, the above-mentioned processing unit 710 may specifically perform:

[0137] The input layer and hidden layer information of the neural network model to be trained are sent to the first participant, and the output layer information of the neural network model to be trained is sent to the second participant. The hidden layer of the first participant and the output layer of the second participant are encrypted and connected in a differential privacy manner.

[0138] Optionally, the above-mentioned processing unit 710 may specifically perform:

[0139] When the number of first participants is not unique, the hidden layers of multiple first participants are encrypted with the output layers of second participants using differential privacy.

[0140] Those skilled in the art will understand that the first participant is used to train the input layer and hidden layer of the neural network model, and the second participant is used to train the output layer of the neural network. The specific structure of the neural network model and the connection method of the neurons can be adjusted according to actual needs, and this disclosure does not limit this.

[0141] Optionally, the hidden layer of the first participant and the output layer of the second participant are encrypted and connected using differential privacy.

[0142] Specifically, in this embodiment of the disclosure, the forward propagation excitation value weighted calculation result and the backpropagation gradient between the hidden layer of the first participant and the output layer of the second participant can be increased with noise through differential privacy to avoid gradient backpropagation.

[0143] Based on the above principles, this disclosure solves the problem of cross-domain gradient privacy protection during the backpropagation training process of longitudinal federated learning neural networks through differential privacy technology. When gradient backpropagation is involved between cross-domain layers, differential privacy processing that conforms to the privacy budget needs to be added to the backpropagation gradient to achieve privacy protection of gradient data between cross-domains.

[0144] Storage unit 720 may include a readable medium in the form of a volatile storage unit, such as random access memory (RAM) 7201 and / or cache memory 7202, and may further include a read-only memory (ROM) 7203.

[0145] The storage unit 720 may also include a program / utility 7204 having a set (at least one) program module 7205, such program module 7205 including but not limited to: an operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0146] Bus 730 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0147] Electronic device 700 can also communicate with one or more external devices 740 (e.g., keyboard, pointing device, Bluetooth device, etc.), and with one or more devices that enable a user to interact with electronic device 700, and / or with any device that enables electronic device 700 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 750. Furthermore, electronic device 700 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 760. As shown, network adapter 760 communicates with other modules of electronic device 700 via bus 730. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 700, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0148] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0149] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, which may be a readable signal medium or a readable storage medium. A program product capable of implementing the methods described above is stored thereon. In some possible implementations, various aspects of this disclosure may also be implemented as a program product including program code, which, when run on a terminal device, causes the terminal device to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure.

[0150] More specific examples of computer-readable storage media in this disclosure may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, 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), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0151] In this disclosure, a computer-readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of transmitting, propagating, or transmitting a program for use by or in connection with an instruction execution system, apparatus, or device.

[0152] Optionally, the program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0153] In practical implementation, program code for performing the operations of this disclosure can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java and C++, and conventional procedural programming languages ​​such as C or similar languages. The program code can execute entirely on the user's computing device, partially on the user's device, as a standalone software package, partially on the user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0154] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0155] Furthermore, although the steps of the method in this disclosure are described in a specific order in the accompanying drawings, this does not require or imply that the steps must be performed in that specific order, or that all the steps shown must be performed to achieve the desired result. Additional or alternative steps may be omitted, multiple steps may be combined into one step, and / or a step may be broken down into multiple steps.

[0156] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, mobile terminal, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0157] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This disclosure is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the appended claims.

Claims

1. A neural network modeling method based on federated learning, characterized in that, The method includes: The model information of the neural network model to be trained is sent to the first participant and the second participant. The neural network model includes an input layer, a hidden layer and an output layer. The first participant is used to train the input layer and the hidden layer, and the second participant is used to train the output layer. Receive the first training result returned by the first participant and the second training result returned by the second participant; The first training result and the second training result are aggregated into a trained neural network model; In this context, the first participant is the feature data provider, and the second participant is the tag data provider; Specifically, the process of distributing the model information of the neural network model to be trained to the first participant and the second participant includes: distributing the input layer information and hidden layer information of the neural network model to be trained to the first participant, and distributing the output layer information of the neural network model to be trained to the second participant. The hidden layer of the first participant and the output layer of the second participant are encrypted and connected using differential privacy.

2. The neural network modeling method based on federated learning according to claim 1, characterized in that, Receiving the first training result returned by the first participant and the second training result returned by the second participant specifically includes: The system receives the neural network structure and neuron weights of the input and hidden layers after training by the first participant, as well as the neural network structure and neuron weights of the output layer after training by the second participant.

3. The neural network modeling method based on federated learning according to claim 1, characterized in that, Before aggregating the first training result and the second training result into a trained neural network model, the method further includes: Obtain the connection information between the first participant and the second participant, wherein the connection information includes the neuronal connection relationships and connection weights between the hidden layer of the first participant and the output layer of the second participant.

4. The neural network modeling method based on federated learning according to claim 3, characterized in that, The first training result and the second training result are aggregated into a trained neural network model, specifically including: Based on the connection information, the first training result and the second training result are concatenated and spliced ​​together to aggregate into a trained neural network model.

5. The neural network modeling method based on federated learning according to claim 1, characterized in that, The method further includes: When the number of the first participants is not unique, the hidden layers of multiple first participants are respectively encrypted with the output layer of the second participants in a differential privacy manner.

6. A neural network modeling device based on federated learning, characterized in that, include: The model distribution module is used to distribute the model information of the neural network model to be trained to the first participant and the second participant. The neural network model includes an input layer, a hidden layer and an output layer. The first participant is used to train the input layer and the hidden layer, and the second participant is used to train the output layer. The parameter acquisition module is used to receive the first training result returned by the first participant and the second training result returned by the second participant. The model aggregation module is used to aggregate the first training result and the second training result into a trained neural network model. In this context, the first participant is the feature data provider, and the second participant is the tag data provider; Specifically, the model distribution module is used to: distribute the input layer information and hidden layer information of the neural network model to be trained to the first participant, and distribute the output layer information of the neural network model to be trained to the second participant, wherein the hidden layer of the first participant and the output layer of the second participant are encrypted and connected in a differential privacy manner.

7. An electronic device, characterized in that, include: processor; as well as Memory for storing the executable instructions of the processor; The processor is configured to execute the neural network modeling method based on federated learning according to any one of claims 1 to 5 by executing the executable instructions.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the neural network modeling method based on federated learning as described in any one of claims 1 to 5.

Images