Federal learning-based automatic driving privacy data protection method and device

Abstract

The invention discloses an automatic driving privacy data protection method based on federal learning, and the method comprises a first device and at least two second devices which are applied to a collaboration party, and each second device is applied to a target vehicle enterprise. The first device is in communication connection with the second device; the method is applied to first equipment. Comprising the following steps: for any target vehicle enterprise in at least two target vehicle enterprises, taking an automatic driving database corresponding to the target vehicle enterprise and model parameters of a vehicle enterprise pre-training model as a privacy database corresponding to the target vehicle enterprise; performing federal co-creation encryption processing on the at least two private databases to generate an encrypted sample database; each private database corresponds to one encrypted sample; and performing federated learning on the collaboration party pre-training model based on the encrypted sample database to generate a federated encryption model. Therefore, federated modeling is carried out on the premise of ensuring that vehicle enterprises of all parties do not share the privacy database, the problem of data islands is solved, privacy protection is completed, and machine learning efficiency is improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of vehicle autonomous driving, and in particular relates to a method and device for protecting privacy data of autonomous driving based on federated learning. Background Art

[0002] At present, the focus of artificial intelligence at home and abroad is still on using the new generation of information technology represented by artificial intelligence to solve the digitalization and intelligence of functions, applications, and services in various fields, and has not fully considered the safe use and sharing of private data. In order to solve the problem of data islands and protect data privacy at the same time, researchers have proposed federated learning. Federated learning allows multiple participants to build a common and robust machine learning model without sharing data, thereby solving the privacy leakage problem caused by the centralization of data. It has become the most widely used privacy data protection solution.

[0003] Most existing federated learning research adopts a single-key homomorphic encryption scheme, that is, each client shares the same key pair. When a client actively intercepts encrypted data sent by other clients, it can directly use its own key to decrypt, resulting in the leakage of the intercepted client's model and local privacy data. There are also a few multi-key federated learning schemes, but there are still problems such as the complex ciphertext structure causing the ciphertext size to grow exponentially with the number of participants, and the number of participants needs to be set in advance, which cannot meet the flexible multi-hop requirements. Summary of the invention

[0004] In view of the above-mentioned problems existing in the prior art, an embodiment of the present invention provides a method and device for protecting privacy data of autonomous driving based on federated learning. The method can realize the available but invisible privacy databases of multiple automobile companies in the creation of a federated encryption model, thereby ensuring that each participating automobile company performs federal modeling without sharing a privacy database, thereby solving the problem of data islands in the federated modeling process, completing privacy protection, and improving the efficiency of machine learning.

[0005] According to a first aspect of an embodiment of the present invention, there is provided a method for protecting privacy data in autonomous driving based on federated learning, characterized in that it comprises a first device applied to a collaborating party and at least two second devices, each of which is applied to a target automobile enterprise; the first device and the second device are communicatively connected; the method is applied to the first device; comprising: for any target automobile enterprise among at least two target automobile enterprises: taking the autonomous driving database corresponding to the target automobile enterprise and the model parameters of the automobile enterprise's pre-trained model as the privacy database corresponding to the target automobile enterprise; wherein the automobile enterprise's pre-trained model is used to indicate a model generated by the second device corresponding to the target automobile enterprise by performing machine learning and / or deep learning based on the autonomous driving database; the autonomous driving database comprises at least: the automobile enterprise name, the user ID, and the regulations of the autonomous driving related institutions; performing federated co-creation encryption processing on the at least two privacy databases to generate an encrypted sample database; wherein the encrypted sample database comprises at least two encrypted samples; each of the privacy databases corresponds to one encrypted sample; performing federated learning on the collaborative party's pre-trained model based on the encrypted sample database to generate a federated encrypted model.

[0006] Optionally, the at least two privacy databases are subjected to federal co-creation encryption processing to generate an encrypted sample database; including: for any privacy database among the at least two privacy databases: performing vector conversion processing on the privacy database to generate a converted vector; centralizing the converted vector to output a centralized vector; the centralized vector and the remaining converted vectors of each of the privacy databases together constitute a candidate sample data group corresponding to the privacy database; weighting the at least two candidate sample data groups to generate a quasi-sample data group; wherein the quasi-sample data group includes the database vector corresponding to the autonomous driving database of each target automobile company and the model parameter vector corresponding to the automobile company's pre-trained model; obtaining the model parameter vector corresponding to each target automobile company from the quasi-sample data group, and calculating the average value of the model parameter vectors of all target automobile companies to obtain the total model parameters; based on the total model parameters and the database vector corresponding to each of the target automobile companies, an encrypted sample database is generated.

[0007] Optionally, when the at least two candidate sample data groups include a first candidate sample data group corresponding to a first target automobile enterprise, a second candidate sample data group corresponding to a second target automobile enterprise, and a third candidate sample data group corresponding to a third target automobile enterprise; the first candidate sample data group includes a first centering vector, a second converted vector, and a third converted vector; the second candidate sample data group includes a first converted vector, a second centering vector, and a third converted vector; the third candidate sample data group includes a first converted vector, a second converted vector, and a third centering vector;

[0008] The method of weighting at least two of the candidate sample data groups to generate a quasi-sample data group includes: determining a first weight corresponding to the first candidate sample data group, a second weight corresponding to the second candidate sample data group, and a third weight corresponding to the third candidate sample data group based on an attention mechanism or a preset rule; applying a first weight to the first centering vector in the first candidate sample data group, a second weight to the first converted vector in the second candidate sample data group, and a third weight to the first converted vector in the third candidate sample data group, and performing summing processing to output a first database vector and a first model parameter vector corresponding to the first target car company; applying a second weight to the second centering vector in the second candidate sample data group, and applying a first weight to the second converted vector in the first candidate sample data group, so as to obtain a first target car company. and the second converted vector in the third candidate sample data group is applied with a third weight, and summed up to output the second database vector and the second model parameter vector corresponding to the second target vehicle company; the third centralization vector in the third candidate sample data group is applied with a third weight, the third converted vector in the first candidate sample data group is applied with a first weight, and the third converted vector in the second candidate sample data group is applied with a second weight, and summed up to output the third database vector and the third model parameter vector corresponding to the third target vehicle company; the first database vector and the first model parameter vector corresponding to the first target vehicle company, the second database vector and the second model parameter vector corresponding to the second target vehicle company, and the third database vector and the third model parameter vector corresponding to the third target vehicle company are determined as quasi-sample data groups.

[0009] Optionally, before performing vector conversion processing on the privacy database, it also includes: performing homomorphic encryption processing on the privacy database and outputting homomorphic encrypted data; performing multi-party computing on the homomorphic encrypted data and outputting the privacy database after multi-party computing.

[0010] Optionally, the privacy database sent by the second device is encrypted based on the federal encryption model to output encrypted data; and an encrypted database is generated based on a number of the encrypted data.

[0011] Optionally, the method also includes: receiving a data request sent by a second device of a requesting vehicle company; determining the requested vehicle company based on the data request; sending the data request to a second device corresponding to the requested vehicle company; upon receiving a consent sharing instruction sent by the requested vehicle company, selecting encrypted data corresponding to the data request from the encrypted database; and sending the encrypted data to the second device corresponding to the requesting vehicle company.

[0012] According to a second aspect of an embodiment of the present invention, a method for protecting privacy data in autonomous driving based on federated learning is provided, which is applied to a second device of a target automobile enterprise; the method comprises: performing compliance data processing on an autonomous driving database corresponding to the target automobile enterprise, and outputting compliant federated data; performing machine learning and / or deep learning on the compliant federated data, and generating a pre-trained model for the automobile enterprise; and sending model parameters corresponding to the pre-trained model of the automobile enterprise and the autonomous driving database to the first device.

[0013] Optionally, the method also includes: receiving encrypted data sent by the first device; optimizing the vehicle enterprise pre-trained model based on the encrypted data to generate a vehicle enterprise quasi-model.

[0014] Optionally, the autonomous driving database corresponding to the target vehicle company is subjected to compliance data processing, and compliant federal data is output; including: based on the target vehicle company signing a local data normalization compliance agreement of all parties, multi-party data normalization processing is performed on the autonomous driving database corresponding to the target vehicle company, and a normalized database is output; local preprocessing is performed on the normalized database, and preprocessed data is output; federal view normalization compliance processing is performed on the preprocessed data, and compliant federal data is output.

[0015] According to a third aspect of an embodiment of the present invention, there is also provided an autonomous driving privacy data protection device based on federated learning, comprising a first device applied to a collaborating party and at least two second devices, each of the second devices being applied to a target automobile company; the first device and the second device are communicatively connected; the device is applied to the first device; comprising: a first determination module, for targeting any one of at least two target automobile companies: using the autonomous driving database corresponding to the target automobile company and the model parameters of the automobile company's pre-trained model as the privacy database corresponding to the target automobile company; wherein the automobile company's pre-trained model is used to indicate a model generated by the second device corresponding to the target automobile company by performing machine learning and / or deep learning based on the autonomous driving database; the autonomous driving database includes at least: the automobile company name, user ID, and regulations of autonomous driving related institutions; an encryption processing module, for performing federated co-creation encryption processing on the privacy databases corresponding to at least two target automobile companies to generate an encrypted sample database; a model generation module, for performing federated learning on the collaborative party's pre-trained model based on the encrypted sample database to generate a federated encryption model.

[0016] According to a fourth aspect of an embodiment of the present invention, a federated learning-based autonomous driving privacy data protection device is applied to a second device of a target automobile company; it includes: a compliance data module, which is used to perform compliance data processing on the autonomous driving database corresponding to the target automobile company, and output compliant federal data; a pre-training module, which is used to perform machine learning and / or deep learning on the compliant federal data to generate a car company pre-training model; a sending module, which is used to send the model parameters corresponding to the car company pre-training model and the autonomous driving database to the first device.

[0017] According to the fifth aspect of the embodiments of the present invention, there is also provided an electronic device, comprising: a processor; a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the method described in the first aspect or the second aspect.

[0018] According to a sixth aspect of an embodiment of the present invention, there is further provided a computer-readable medium on which a computer program is stored, and when the program is executed by a processor, the method described in the first aspect or the second aspect is implemented.

[0019] An embodiment of the present invention provides a method for protecting privacy data of autonomous driving based on federated learning, comprising a first device applied to a collaborating party and at least two second devices, each of which is applied to a target automobile company; the first device and the second device are communicatively connected; the method is applied to the first device; comprising: for any target automobile company among at least two target automobile companies: using the autonomous driving database corresponding to the target automobile company and the model parameters of the automobile company's pre-trained model together as the privacy database corresponding to the target automobile company; wherein the automobile company's pre-trained model is used to indicate a model generated by the second device corresponding to the target automobile company based on the autonomous driving database through machine learning and / or deep learning; the autonomous driving database includes at least: the automobile company name, user ID, and autonomous driving-related institutional regulations; performing federated co-creation encryption processing on the at least two privacy databases to generate an encrypted sample database; wherein the encrypted sample database includes at least two encrypted samples; each of the privacy databases corresponds to one encrypted sample; performing federated learning on the collaborative party's pre-trained model based on the encrypted sample database to generate a federated encrypted model. This embodiment uses federal co-creation encryption processing technology to encrypt the privacy databases generated by multiple target automobile companies during the autonomous driving process, and performs federated learning based on the encrypted sample database generated by the encryption processing. This ensures that federal modeling is performed on the premise that the participating automobile companies do not share the privacy database, which not only solves the data island problem but also completes privacy protection and improves the efficiency of machine learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Hereinafter, some specific embodiments of the present invention will be described in detail in an exemplary and non-limiting manner with reference to the accompanying drawings. The same reference numerals in the accompanying drawings indicate the same or similar components or parts. It should be understood by those skilled in the art that these drawings are not necessarily drawn to scale. In the accompanying drawings:

[0021] Figure 1 A schematic diagram of a process flow of a method for protecting privacy data in autonomous driving based on federated learning provided in one embodiment of the present invention;

[0022] Figure 2 A schematic diagram of a process for generating an encrypted sample database provided by an embodiment of the present invention;

[0023] Figure 3 A schematic diagram of a flow chart of a method for protecting privacy data in autonomous driving based on federated learning provided in another embodiment of the present invention;

[0024] Figure 4 A schematic diagram of a federated encryption model framework based on federated co-creation provided by an embodiment of the present invention;

[0025] Figure 5 A schematic diagram of the structure of a privacy data protection device for autonomous driving based on federated learning provided in one embodiment of the present invention;

[0026] Figure 6 A schematic diagram of the structure of an autonomous driving privacy data protection device based on federated learning provided in another embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0028] In 2016, Google proposed the concept of federated learning (Konen et al., 2016). Federated learning is a distributed computing method that allows multiple participants to jointly carry out efficient machine learning while ensuring data exchange security and personal privacy. Its characteristics are that the data of the participants is available but not visible, and the local data is trained, the updated parameters are uploaded to the server, and then the server aggregates to obtain the total parameters. Compared with traditional machine learning methods, federated learning can not only improve learning efficiency, but also ensure that all participants can perform federated modeling without sharing data, thus solving the problem of data islands and achieving privacy protection.

[0029] Assume that there are N car companies that store and are eligible to retrieve autonomous driving user data to participate in modeling. Each car company participant Ui has its own data set Di. Federated learning is no longer about integrating all data together for unified training to obtain a model. Instead, each car company participant trains local data based on the initial parameters wG transmitted from the server to obtain new parameters wGi′, and then transmits the updated parameter value δi=wGi′-wG to the server. The server aggregates and obtains new overall parameters wG′=wG+f(δ1;δ2;···;δn). After multiple iterations, the overall training model MFed is finally obtained. Federated learning needs to ensure that the gap between the effect VFed of MFed and the effect VSum of the model MSum is small enough, that is, |VFed-VSum|<ε, where ε is an arbitrarily small positive value (Yang et al., 2019).

[0030] like Figure 1 As shown, it is a flowchart of a method for protecting privacy data in autonomous driving based on federated learning provided in one embodiment of the present invention.

[0031] A method for protecting privacy data of autonomous driving based on federated learning, comprising a first device applied to a collaborating party and at least two second devices, each of which is applied to a target automobile enterprise; the first device and the second device are in communication connection; the method is applied to the first device; and at least comprises the following steps:

[0032] S101, for any one of at least two target automobile companies: using the autonomous driving database corresponding to the target automobile company and the model parameters of the automobile company's pre-trained model together as the privacy database corresponding to the target automobile company; wherein the automobile company's pre-trained model is used to indicate a model generated by a second device corresponding to the target automobile company through machine learning and / or deep learning based on the autonomous driving database; the autonomous driving database at least includes: the automobile company name, user ID, and autonomous driving related institutional regulations;

[0033] S102, performing federated co-creation encryption processing on at least two privacy databases to generate an encrypted sample database; wherein the encrypted sample database includes at least two encrypted samples; and each privacy database corresponds to one encrypted sample;

[0034] S103, performing federated learning on the pre-trained model of the collaborating party based on the encrypted sample database to generate a federated encryption model.

[0035] In S101, the regulations of autonomous driving agencies include at least: data privacy protection laws, relevant autonomous driving regulations, and traffic laws. The second device of the target car company stores the privacy database locally and is eligible to access its own privacy database. In order to protect privacy data and meet the requirements of the current "Personal Information Protection Law (Draft)", the target car companies usually do not share or actively exchange their own privacy databases. Therefore, the target car companies used for federated learning maintain the independence of their privacy databases.

[0036] It should be noted that the car company pre-training model corresponding to each target car company can be the same or different. For example: the car company pre-training model corresponding to the first target car company is used to indicate the model generated by the first target car company based on machine learning to train the autonomous driving database corresponding to the first target car company. The car company pre-training model corresponding to the second target car company is used to indicate the model generated by the second target car company based on machine learning and deep learning to train the autonomous driving database corresponding to the second target car company.

[0037] In S102 and S103 , each target automobile enterprise has a corresponding second device.

[0038] There is no limitation on the specific process of federated co-creation encryption processing; as long as it can prevent the target car companies of both parties or multiple target car companies from guessing the original privacy database. For example: the first device implements federated co-creation encryption processing by aligning and encrypting the privacy databases of multiple target car companies. Through federated co-creation encryption processing, the first device not only encrypts the autonomous driving databases of each target car company, but also encrypts the model parameters of each car company.

[0039] The collaborator's pre-trained model can be an initialized random model or a model trained based on the autonomous driving database of at least one target car company.

[0040] The classification of federated learning considers the overlap between the user dimension and the feature dimension, and is proposed based on the data distribution in the encrypted sample database, including horizontal federated learning, vertical federated learning, and federated transfer learning. Vertical federated learning is based on different data sets. When there is a lot of overlap in data features and less overlap in users, the data set is segmented according to the user dimension and the data with the same data features but different users are taken out for training. Horizontal federated learning is based on different data sets. When there is a lot of overlap in users and less overlap in data features, the data set is segmented according to the data feature dimension and the data with the same user but different data features are taken out for training. Federated transfer learning is based on the situation where there is little overlap in users and data features of multiple data sets, and the data is not segmented, but transfer learning is used to overcome the situation of insufficient data or labels. In the practice of protecting user privacy data in autonomous driving, the application scenarios of federated learning can also be classified into ToB and ToC. For ToB scenarios, its main service objects are car companies and related institutions; for ToC scenarios, the federated modeling participants are mainly edge computing devices, and usually this type of federated learning has a large number of participants and low computing power.

[0041] In this embodiment, the first device performs federal co-creation encryption processing on the privacy databases corresponding to at least two target automobile companies; uses the generated encrypted sample database to perform federal learning on the pre-trained model of the collaborating party to generate a federal encryption model; thereby, while ensuring that all participating automobile companies perform federal modeling without sharing privacy databases, it not only solves the data island problem but also completes privacy protection.

[0042] like Figure 2 FIG. 1 is a flow chart of generating an encrypted sample database according to an embodiment of the present invention.

[0043] Performing federated co-creation encryption processing on the at least two privacy databases to generate an encrypted sample database; including:

[0044] S201, for any privacy database among at least two privacy databases: perform vector conversion processing on the privacy database to generate a converted vector; perform centralization processing on the converted vector to output a centralized vector; and use the centralized vector and the converted vectors of each remaining privacy database to form a candidate sample data group corresponding to the privacy database;

[0045] S202, performing weighted processing on at least two candidate sample data groups to generate a quasi-sample data group; wherein the quasi-sample data group includes a database vector corresponding to the autonomous driving database of each target automobile enterprise and a model parameter vector corresponding to the automobile enterprise's pre-trained model;

[0046] S203, obtaining a model parameter vector corresponding to each target automobile enterprise from the quasi-sample data group, and calculating an average value of the model parameter vectors of all target automobile enterprises to obtain a total model parameter;

[0047] S204: Generate an encrypted sample database based on the total model parameters and the database vector corresponding to each target automobile company.

[0048] In this embodiment, there are at least two target automobile companies for federated learning.

[0049] For example: when the at least two candidate sample data groups include a first candidate sample data group corresponding to a first target car company, a second candidate sample data group corresponding to a second target car company, and a third candidate sample data group corresponding to a third target car company; the first candidate sample data group includes a first centering vector, a second transformed vector, and a third transformed vector; the second candidate sample data group includes a first transformed vector, a second centering vector, and a third transformed vector; the third candidate sample data group includes a first transformed vector, a second transformed vector, and a third centering vector;

[0050] The method of weighting at least two of the candidate sample data groups to generate a quasi-sample data group includes: determining a first weight corresponding to the first candidate sample data group, a second weight corresponding to the second candidate sample data group, and a third weight corresponding to the third candidate sample data group based on an attention mechanism or a preset rule; applying a first weight to the first centering vector in the first candidate sample data group, a second weight to the first converted vector in the second candidate sample data group, and a third weight to the first converted vector in the third candidate sample data group, and performing summing processing to output a first database vector and a first model parameter vector corresponding to the first target car company; applying a second weight to the second centering vector in the second candidate sample data group, and applying a first weight to the second converted vector in the first candidate sample data group, so as to obtain a first target car company. The third weight is applied to the second converted vector in the third candidate sample data group, and the sum is performed to output the second database vector and the second model parameter vector corresponding to the second target car company; the third centralization vector in the third candidate sample data group is applied with the third weight, the third converted vector in the first candidate sample data group is applied with the first weight, and the third converted vector in the second candidate sample data group is applied with the second weight, and the sum is performed to output the third database vector and the third model parameter vector corresponding to the third target car company; the first database vector and the first model parameter vector corresponding to the first target car company, the second database vector and the second model parameter vector corresponding to the second target car company, and the third database vector and the third model parameter vector corresponding to the third target car company are determined as a quasi-sample data group. The first model parameter vector, the second model parameter vector, and the third model parameter vector are averaged to obtain the total model parameters. Based on the first database vector, the second database vector, the third database vector, and the total model parameters, a federal co-creation encrypted database is generated.

[0051] Here, "federated co-creation encryption processing" is also called "centralized data exchange fusion weighted average federated algorithm" (CWA-FL).

[0052] This embodiment uses federal co-creation encryption processing to encrypt the privacy databases of multiple target car companies. This can effectively protect data privacy, prevent the two target car companies and multiple target car companies from deducing and guessing the original privacy database, and conduct in-depth analysis and protection of distributed data. It also improves the accuracy of the training of the federal encryption model.

[0053] In a preferred implementation manner of this embodiment, before the privacy database is subjected to vector conversion processing, it also includes: performing homomorphic encryption processing on the privacy database and outputting homomorphically encrypted data; performing multi-party computation on the homomorphic encrypted data and outputting the privacy database after multi-party computation.

[0054] Therefore, the combination of homomorphic encryption, multi-party computing, and federal co-creation encryption processing can effectively protect the privacy databases of all target car companies, prevent the target car companies from obtaining the original privacy databases through reverse deduction technology, improve the level of protection of the original privacy database, and make the federal encryption model available but invisible to the privacy databases of all car companies during training.

[0055] In a preferred implementation manner of this embodiment, the method further includes: encrypting the privacy database sent by the second device based on the federal encryption model and outputting encrypted data; and generating an encrypted database based on a plurality of the encrypted data.

[0056] After receiving the privacy database sent by the second device, the first device directly encrypts the privacy database and outputs the encrypted data. Since each privacy database corresponds to one encrypted data, at least two encrypted data form an encrypted database.

[0057] In a preferred implementation of this embodiment, the method also includes: receiving a data request sent by a second device of a requesting vehicle company; determining the requested vehicle company based on the data request; sending the data request to a second device corresponding to the requested vehicle company; upon receiving a consent sharing instruction sent by the requested vehicle company, selecting encrypted data corresponding to the data request from the encrypted database; and sending the encrypted data to the second device corresponding to the requesting vehicle company.

[0058] Here, since the requested car company is used to indicate other target car companies for federated learning in addition to the requesting car company, the encrypted data corresponding to the data request also does not include the encrypted data corresponding to the requesting car company.

[0059] In this embodiment, the first device sends the encrypted data corresponding to the data request to the requested car company, so that the requested car company can use the encrypted data but not see it, thereby effectively protecting the privacy databases of all car companies and enabling them to share training data, which is beneficial to the training of accurate models of all car companies, solves the problem of data islands when conducting model training, and improves the efficiency of machine learning.

[0060] like Figure 3 As shown, it is a flowchart of an autonomous driving privacy data protection method based on federated learning provided by another embodiment of the present invention.

[0061] A method for protecting privacy data in autonomous driving based on federated learning, applied to a second device of a target car company; including:

[0062] S301: Perform compliance data processing on the autonomous driving database corresponding to the target car company and output compliant federal data;

[0063] S302, performing machine learning and / or deep learning on compliant federal data to generate a pre-trained model for the automobile enterprise;

[0064] S303: Send the model parameters corresponding to the car manufacturer's pre-trained model and the autonomous driving database to the first device.

[0065] Specifically, first, the second device performs multi-party data normalization processing on the autonomous driving database corresponding to the target car company based on the local data normalization compliance agreement signed by the target car company, and outputs a normalized database; secondly, the second device performs local preprocessing on the normalized database and outputs preprocessed data; after that, the second device performs federal view normalization compliance processing on the preprocessed data and outputs compliant federal data. Here, local preprocessing includes: deduplication, standardization, and error correction.

[0066] This solves the problem of data islands during federated encryption model training and improves the accuracy of federated encryption model training.

[0067] In a preferred implementation manner of this embodiment, the method further includes: receiving encrypted data sent by the first device; optimizing the car company pre-trained model based on the encrypted data to generate a car company quasi-model.

[0068] After receiving the encrypted data sent by the first device, the second device optimizes or trains the car company's pre-trained model based on the encrypted data to generate a car company's accurate model; thereby, the second device can re-train the car company's pre-trained model based on the privacy databases of multiple target car companies, thereby improving the prediction coverage capability of the car company's accurate model.

[0069] A system for protecting privacy data of autonomous driving based on federated learning includes a first device applied to a collaborating party and at least two second devices, each of which is applied to a target automobile enterprise; the first device and the second device are in communication connection; and at least the following steps are included:

[0070] S1, the second device performs compliance data processing on the autonomous driving database corresponding to the target automobile enterprise, and outputs compliant federal data; performs machine learning and / or deep learning on the compliant federal data to generate a pre-trained model for the automobile enterprise; and sends the model parameters corresponding to the pre-trained model of the automobile enterprise and the autonomous driving database to the first device.

[0071] S2, for any one of at least two target automobile companies: the first device uses the autonomous driving database corresponding to the target automobile company and the model parameters of the automobile company's pre-trained model as the privacy database corresponding to the target automobile company; wherein the automobile company's pre-trained model is used to indicate the model generated by the second device corresponding to the target automobile company through machine learning and / or deep learning based on the autonomous driving database; the autonomous driving database includes at least: the automobile company name, user ID, and regulations of autonomous driving related institutions.

[0072] S3, for any privacy database among at least two privacy databases: the first device performs vector conversion processing on the privacy database to generate a converted vector; performs centralization processing on the converted vector and outputs a centralized vector; the centralized vector and the remaining converted vectors of each privacy database together form a candidate sample data group corresponding to the privacy database; at least two of the candidate sample data groups are weighted to generate a quasi-sample data group; wherein the quasi-sample data group includes the database vector corresponding to the autonomous driving database of each target car company and the model parameter vector corresponding to the car company's pre-trained model; the model parameter vector corresponding to each target car company is obtained from the quasi-sample data group, and the average value of the model parameter vectors of all target car companies is calculated to obtain the total model parameters; based on the total model parameters and the database vector corresponding to each target car company, an encrypted sample database is generated. wherein the encrypted sample database includes at least two encrypted samples; each privacy database corresponds to one encrypted sample.

[0073] S4, the first device performs federated learning on the pre-trained model of the collaborating party based on the encrypted sample database to generate a federated encryption model.

[0074] S5, the first device encrypts the privacy database sent by the second device based on the federal encryption model and outputs encrypted data; based on a number of the encrypted data, an encrypted database is generated.

[0075] S6, the first device receives a data request sent by the second device of the requesting car company; determines the requested car company based on the data request; sends the data request to the second device corresponding to the requested car company; when receiving the consent sharing instruction sent by the requested car company, selects encrypted data corresponding to the data request from the encrypted database; and sends the encrypted data to the second device corresponding to the requesting car company.

[0076] S7, the second device receives the encrypted data sent by the first device; and optimizes the car company's pre-trained model based on the encrypted data to generate a car company's quasi-model.

[0077] In this embodiment, the first device interacts with at least two second devices, and the first device fully manages the entire federated learning process. When processing large-scale distributed data, CWA-FL can avoid direct transmission of sensitive data by pre-training and updating the model in a local privacy database, thereby protecting the user's data privacy.

[0078] The following is a detailed description of the autonomous driving privacy data protection method based on federated learning provided in this embodiment in conjunction with specific application scenarios.

[0079] A method for protecting privacy data in autonomous driving based on federated learning. The specific operation process includes:

[0080] (1) Initialization: All second devices initialize the random model at the beginning of the task to obtain the initial task model;

[0081] (2) The first device selects the second device: a portion of the second devices are randomly selected from all the second devices for model pre-training;

[0082] (3) Model pre-training: The selected second device uses the initial task model to train on its local autonomous driving database until the respective gradient directions converge, generating a car company pre-training model;

[0083] (4) Model update and upload: Each second device uploads the trained car company pre-trained model to the first device;

[0084] (5) Federal co-creation encryption processing: After the pre-training of the first device is completed, the first device vectorizes all uploaded pre-trained models of automobile companies, and uses the centralized data exchange fusion weighted average federation algorithm to group all second devices and corresponding models for calculation;

[0085] (6) Model tuning: The first device performs federated learning on the pre-trained model of the collaborating party based on the encrypted sample database generated after the federated co-creation encryption process, and adjusts its parameters to optimize performance;

[0086] (7) Model convergence detection: The first device checks whether the pre-trained model of the first device's collaborator has converged on the data of the new task. If it has converged, a federated encryption model is generated; and the first device will start training for the next task. Otherwise, return to step (2) and continue the training process.

[0087] This embodiment provides an efficient learning strategy based on federated learning, the main goal of which is to effectively protect data privacy, prevent two or more parties from guessing the original data, and conduct in-depth analysis and protection of distributed data. CWA-FL adopts a centralized federated learning architecture, in which the first device interacts with multiple second devices and fully manages the entire federated learning process.

[0088] like Figure 4 As shown, it is a schematic diagram of a federated encryption model framework based on federal co-creation provided by an embodiment of the present invention.

[0089] In the practice of the solution for protecting the privacy data of autonomous driving users, we focus on the underlying code → algorithm optimization → intelligent platform construction, through independent research and development innovation and intelligent management. The model for protecting the privacy data of autonomous driving users based on the idea of ​​federal co-creation mainly includes a customized framework of the parent framework (one-stop intelligent collaboration system) and the sub-framework (federal co-creation modeling framework platform):

[0090] Mother Framework: A one-stop intelligent collaboration system.

[0091] The one-stop intelligent collaboration system realizes intelligent prediction and explainable innovative attempts based on federated co-creation learning. The collaboration chain is consolidated from the data layer, federation layer, algorithm layer, and optimization layer. In terms of practical functions, it is specifically implemented through the [privacy and security module] connecting and paralleling the [multi-party data source module], [multi-party federated co-creation module], [local pre-training and formal training module], and [model optimization and update module]. The first device includes: multi-party data source module, multi-party federated co-creation module, local pre-training and formal training module, privacy and security module, and model optimization and update module; the second device includes: multi-party data source module, multi-party federated co-creation module, local pre-training and formal training module, and model optimization and update module.

[0092] The autonomous driving data in the multi-party data source module is divided into structured data and unstructured data; the autonomous driving data in the multi-party data source module comes from car companies, autonomous driving users, and related structural laws and regulations. The multi-party data source module sends the autonomous driving database formed by the autonomous driving data to the multi-party federation co-creation module.

[0093] The multi-party federated co-creation module performs compliance data processing on the autonomous driving database; compliance data processing includes: multi-party data standardization, local preprocessing, and federated view normalization compliance, etc. For example: first, based on the signed local data normalization compliance agreement of all parties, multi-party data standardization processing is performed on the autonomous driving database to obtain normalized data; second, based on deduplication, standardization, and error correction methods, normalized data is preprocessed locally to obtain preprocessed data; finally, the preprocessed data is processed in a federal view normalization compliance manner through federated normalization to output compliant federated data. The multi-party federated co-creation module delivers compliant federated data to the local pre-training and formal training modules.

[0094] The local pre-training and formal training modules include first performing machine learning on compliant federal data; then performing deep learning on the machine learning results to output the car company's pre-training model; among them, machine learning includes: functionalization, regression, and multi-classification, etc.; deep learning includes CNN, RNN, and BERT, etc.

[0095] The local pre-training and formal training modules send the car company's pre-training model or the collaborating party's pre-training model to the model optimization and update module.

[0096] The model optimization and update processing includes several units; the several units include: asymmetric encryption of model information, compliant transmission, collaborative integration of gradient parameters, optimization of the federated co-creation learning model, and updates of various models.

[0097] The privacy and security module includes homomorphic encryption, multi-party computing, and centralized data exchange fusion weighted average federation algorithm.

[0098] Sub-framework: Federal co-creation modeling framework platform

[0099] In the federated co-creation modeling framework platform, through parallel encryption algorithms, asynchronous computing, federated co-creation learning and other technologies, the learning efficiency is improved under the premise of ensuring data security, and the ability to integrate billions of scale data is gradually achieved. This framework uses and mines the data value of all parties more efficiently and accurately, achieving a win-win situation for all parties. There are three major characteristics. The first characteristic is that in terms of data and model privacy, the data of different active and passive participants (see the figure database 1-data worker 1; database 2-data worker 2; ... database q-data worker q) do not directly exchange local autonomous driving databases and model parameters, but exchange intermediate values ​​required to update parameters. The second characteristic is that in terms of information transmission, the concept of "centralized data exchange fusion weighted average algorithm" is proposed, and the collaborating parties complete the data processing, so that the data exchange is independent of the participants and it is difficult for other collaborating parties to deduce and guess the data; the third characteristic is the use of an asynchronous computing framework to improve the training speed of the federated encryption model. The parent framework is the basis of the data system, and the sub-framework completes the federated co-creation modeling calculation. The parent framework and the sub-framework work together to complete the task of protecting the privacy data of autonomous driving users.

[0100] This embodiment shares the privacy databases of all automobile companies under the premise of effectively protecting their privacy databases, thereby solving the problem of data islands during the training of federal encryption models and automobile companies' quasi-models, and improving the accuracy of model training.

[0101] like Figure 5 As shown, it is a structural diagram of an autonomous driving privacy data protection device based on federated learning provided by one embodiment of the present invention.

[0102] A device for protecting privacy data of autonomous driving based on federated learning, comprising a first device applied to a collaborating party and at least two second devices, each of which is applied to a target automobile enterprise; the first device and the second device are communicatively connected; the device 500 is applied to the first device; comprising: a first determination module 501, for targeting any one of at least two target automobile enterprises: using the autonomous driving database corresponding to the target automobile enterprise and the model parameters of the automobile enterprise's pre-trained model as the privacy database corresponding to the target automobile enterprise; wherein the automobile enterprise's pre-trained model is used to indicate a model generated by a second device corresponding to the target automobile enterprise by machine learning and / or deep learning based on the autonomous driving database; the autonomous driving database includes at least: automobile enterprise name, user ID, and autonomous driving-related institutional regulations; an encryption processing module 502, for performing federated co-creation encryption processing on the privacy databases corresponding to at least two target automobile enterprises to generate an encrypted sample database; a model generation module 503, for performing federated learning on the collaborative party's pre-trained model based on the encrypted sample database to generate a federated encryption model.

[0103] In a preferred implementation manner of this embodiment, the encryption module includes: a first encryption unit, which is used to: for any privacy database among at least two privacy databases: perform vector conversion processing on the privacy database to generate a converted vector; perform centralization processing on the converted vector to output a centralized vector; the centralized vector and the remaining converted vectors of each of the privacy databases together constitute a candidate sample data group corresponding to the privacy database; a second encryption unit, which is used to perform weighted processing on at least two of the candidate sample data groups to generate a quasi-sample data group; wherein the quasi-sample data group includes a database vector corresponding to the autonomous driving database of each target automobile company and a model parameter vector corresponding to the automobile company's pre-trained model; a third encryption unit, which is used to obtain the model parameter vector corresponding to each target automobile company from the quasi-sample data group, and calculate the average value of the model parameter vectors of all target automobile companies to obtain the total model parameters; a fourth encryption unit, which is used to generate an encrypted sample database based on the total model parameters and the database vector corresponding to each of the target automobile companies.

[0104] In a preferred implementation of this embodiment, when the at least two candidate sample data groups include a first candidate sample data group corresponding to the first target car company, a second candidate sample data group corresponding to the second target car company, and a third candidate sample data group corresponding to the third target car company; the first candidate sample data group includes a first centering vector, a second converted vector, and a third converted vector; the second candidate sample data group includes a first converted vector, a second centering vector, and a third converted vector; the third candidate sample data group includes a first converted vector, a second converted vector, and a third centering vector;

[0105] The second encryption unit includes: a first determining subunit, which is used to determine, based on an attention mechanism or a preset rule, a first weight corresponding to the first candidate sample data group, a second weight corresponding to the second candidate sample data group, and a third weight corresponding to the third candidate sample data group; a first weighting subunit, which is used to apply a first weight to the first centering vector in the first candidate sample data group, a second weight to the first converted vector in the second candidate sample data group, and a third weight to the first converted vector in the third candidate sample data group, and perform summing processing to output a first database vector and a first model parameter vector corresponding to the first target car company; a second weighting subunit, which is used to apply a second weight to the second centering vector in the second candidate sample data group, a first weight to the second converted vector in the first candidate sample data group, and a third weight to the first converted vector in the third candidate sample data group. A third weight is applied to the second converted vector in the sample data group, and summed up, and a second database vector and a second model parameter vector corresponding to the second target vehicle company are output; a third weighting subunit is used to apply a third weight to the third centralizing vector in the third candidate sample data group, a first weight to the third converted vector in the first candidate sample data group, and a second weight to the third converted vector in the second candidate sample data group, and sum them up, and output a third database vector and a third model parameter vector corresponding to the third target vehicle company; a second determination unit is used to determine the first database vector and the first model parameter vector corresponding to the first target vehicle company, the second database vector and the second model parameter vector corresponding to the second target vehicle company, and the third database vector and the third model parameter vector corresponding to the third target vehicle company as a quasi-sample data group.

[0106] In a preferred implementation of this embodiment, the encryption module also includes: a homomorphic encryption unit, which is used to perform homomorphic encryption processing on the privacy database and output homomorphic encrypted data; a multi-party computing unit, which is used to perform multi-party computing on the homomorphic encrypted data and output the privacy database after multi-party computing.

[0107] In a preferred implementation manner of this embodiment, the device also includes: an encryption prediction module, which is used to encrypt the privacy database sent by the second device based on the federal encryption model and output encrypted data; an encrypted database generation module, which is used to generate an encrypted database based on a number of the encrypted data.

[0108] In a preferred implementation of this embodiment, the device also includes: a receiving module, used to receive a data request sent by a second device of a requesting vehicle company; a second determination module, used to determine the requested vehicle company based on the data request; a first sending module, used to send the data request to the second device corresponding to the requested vehicle company; a selection module, used to select encrypted data corresponding to the data request from the encrypted database when receiving a consent sharing instruction sent by the requested vehicle company; and a second sending module, used to send the encrypted data to the second device corresponding to the requesting vehicle company.

[0109] like Figure 6 , which is a schematic diagram of the structure of an autonomous driving privacy data protection device based on federated learning provided by another embodiment of the present invention.

[0110] A federated learning-based autonomous driving privacy data protection device is applied to a second device of a target automobile company; the device 600 includes: a compliance data module 601, which is used to perform compliance data processing on the autonomous driving database corresponding to the target automobile company and output compliant federal data; a pre-training module 602, which is used to perform machine learning and / or deep learning on the compliant federal data to generate a pre-trained model for the automobile company; a sending module 603, which is used to send the model parameters corresponding to the pre-trained model of the automobile company and the autonomous driving database to the first device.

[0111] In a preferred implementation manner of this embodiment, the device also includes: a receiving module for receiving encrypted data sent by the first device; and a model optimization module for optimizing the car company's pre-trained model based on the encrypted data to generate a car company's quasi-model.

[0112] In a preferred implementation manner of this embodiment, the compliance data module includes: a multi-party data normalization unit, which is used to perform multi-party data normalization processing on the autonomous driving database corresponding to the target vehicle company based on the local data normalization compliance agreement signed by the target vehicle company, and output a normalized database; a local preprocessing unit, which is used to perform local preprocessing on the normalized database and output preprocessed data; and a compliance processing unit, which is used to perform federal view normalization compliance processing on the preprocessed data and output compliant federal data.

[0113] The above-mentioned device can execute the autonomous driving privacy data protection method based on federated learning provided in one embodiment of the present invention, and has the corresponding functional modules and beneficial effects of executing the autonomous driving privacy data protection method based on federated learning. For technical details not fully described in this embodiment, please refer to the autonomous driving privacy data protection method based on federated learning provided in one embodiment of the present invention.

[0114] The present invention also provides an electronic device, comprising: a processor; a memory for storing executable instructions of the processor; the processor is used to read the executable instructions from the memory and execute the instructions to implement the autonomous driving privacy data protection method based on federated learning described in the present invention.

[0115] In addition to the above-mentioned methods and devices, an embodiment of the present application may also be a computer program product, which includes computer program instructions, which, when executed by a processor, enable the processor to execute the steps of the method according to various embodiments of the present application described in the above-mentioned "Exemplary Method" section of this specification.

[0116] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages, such as Java, C++, etc., and conventional procedural programming languages, such as "C" language or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0117] In addition, an embodiment of the present application may also be a computer-readable storage medium on which computer program instructions are stored. When the computer program instructions are executed by a processor, the processor executes the steps of the method according to the following embodiments of the present application described in the above "Exemplary Method" section of this specification.

[0118] The computer readable storage medium can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0119] The basic principles of the present application are described above in conjunction with specific embodiments. However, it should be noted that the advantages, strengths, effects, etc. mentioned in the present application are only examples and not limitations, and it cannot be considered that these advantages, strengths, effects, etc. are required by each embodiment of the present application. In addition, the specific details disclosed above are only for the purpose of illustration and ease of understanding, not for limitation, and the above details do not limit the present application to being implemented by adopting the above specific details.

[0120] The block diagrams of the devices, apparatuses, equipment, and systems involved in this application are only illustrative examples and are not intended to require or imply that they must be connected, arranged, and configured in the manner shown in the block diagram. As will be appreciated by those skilled in the art, these devices, apparatuses, equipment, and systems can be connected, arranged, and configured in any manner. Words such as "including", "comprising", "having", etc. are open words, referring to "including but not limited to", and can be used interchangeably with them. The words "or" and "and" used here refer to the words "and / or" and can be used interchangeably with them, unless the context clearly indicates otherwise. The words "such as" used here refer to the phrase "such as but not limited to", and can be used interchangeably with them.

[0121] It should also be noted that in the apparatus, device and method of the present application, each component or each step can be decomposed and / or recombined. Such decomposition and / or recombination should be regarded as equivalent solutions of the present application.

[0122] The above description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the general principles defined herein may be applied to other aspects without departing from the scope of the present application. Therefore, the present application is not intended to be limited to the aspects shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

[0123] The above description has been given for the purpose of illustration and description. In addition, this description is not intended to limit the embodiments of the present application to the forms disclosed herein. Although multiple example aspects and embodiments have been discussed above, those skilled in the art will recognize certain variations, modifications, changes, additions and sub-combinations thereof.

[0124] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0125] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0126] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for protecting privacy data in autonomous driving based on federated learning, characterized in that: It includes a first device applied to a collaborating party and at least two second devices, each of which is applied to a target automobile enterprise; The first device is communicatively connected with the second device; The method is applied to a first device; comprising: For any one of at least two target automobile companies: the autonomous driving database corresponding to the target automobile company and the model parameters of the automobile company's pre-trained model are used together as the privacy database corresponding to the target automobile company; wherein the automobile company's pre-trained model is used to indicate a model generated by a second device corresponding to the target automobile company through machine learning and / or deep learning based on the autonomous driving database; the autonomous driving database at least includes: the automobile company name, user ID, and autonomous driving related institutional regulations; Performing federated co-creation encryption processing on the at least two privacy databases to generate an encrypted sample database; wherein the encrypted sample database includes at least two encrypted samples; each of the privacy databases corresponds to one encrypted sample; Based on the encrypted sample database, the pre-trained model of the collaborating party is federated learned to generate a federated encryption model.

2. The method according to claim 1, characterized in that The step of performing a federated co-creation encryption process on the at least two privacy databases to generate an encrypted sample database comprises: For any privacy database among at least two privacy databases: performing vector conversion processing on the privacy database to generate a converted vector; performing centralization processing on the converted vector to output a centralized vector; and combining the centralized vector and the remaining converted vectors of each of the privacy databases to form a candidate sample data group corresponding to the privacy database; Performing weighted processing on at least two of the candidate sample data groups to generate a quasi-sample data group; wherein the quasi-sample data group includes a database vector corresponding to the autonomous driving database of each target automobile enterprise and a model parameter vector corresponding to the automobile enterprise's pre-trained model; Obtaining a model parameter vector corresponding to each target automobile enterprise from the quasi-sample data group, and calculating an average value of the model parameter vectors of all target automobile enterprises to obtain a total model parameter; Based on the total parameters of the model and the database vector corresponding to each of the target automobile companies, an encrypted sample database is generated.

3. The method according to claim 2, characterized in that When the at least two candidate sample data groups include a first candidate sample data group corresponding to a first target car company, a second candidate sample data group corresponding to a second target car company, and a third candidate sample data group corresponding to a third target car company; the first candidate sample data group includes a first centering vector, a second converted vector, and a third converted vector; the second candidate sample data group includes a first converted vector, a second centering vector, and a third converted vector; the third candidate sample data group includes a first converted vector, a second converted vector, and a third centering vector; The step of performing weighted processing on at least two candidate sample data groups to generate a quasi-sample data group comprises: Based on an attention mechanism or a preset rule, determine a first weight corresponding to the first candidate sample data group, a second weight corresponding to the second candidate sample data group, and a third weight corresponding to the third candidate sample data group; Apply a first weight to the first centralized vector in the first candidate sample data group, apply a second weight to the first converted vector in the second candidate sample data group, and apply a third weight to the first converted vector in the third candidate sample data group, and perform summation processing to output a first database vector and a first model parameter vector corresponding to the first target car company; Apply a second weight to the second centralized vector in the second candidate sample data group, apply a first weight to the second converted vector in the first candidate sample data group, and apply a third weight to the second converted vector in the third candidate sample data group, and perform summation processing to output a second database vector and a second model parameter vector corresponding to the second target car company; Apply a third weight to the third centralized vector in the third candidate sample data group, apply a first weight to the third converted vector in the first candidate sample data group, and apply a second weight to the third converted vector in the second candidate sample data group, and perform summation processing to output a third database vector and a third model parameter vector corresponding to the third target car company; The first database vector and the first model parameter vector corresponding to the first target car manufacturer, the second database vector and the second model parameter vector corresponding to the second target car manufacturer, and the third database vector and the third model parameter vector corresponding to the third target car manufacturer are determined as a quasi-sample data group.

4. The method according to claim 2, characterized in that: Before performing vector conversion processing on the privacy database, the method further includes: Performing homomorphic encryption processing on the privacy database and outputting homomorphic encrypted data; Perform multi-party computation on the homomorphic encrypted data and output a privacy database after the multi-party computation.

5. The method according to claim 1, characterized in that Also includes: Encrypting the privacy database sent by the second device based on the federated encryption model, and outputting encrypted data; Based on the plurality of encrypted data, an encrypted database is generated.

6. The method according to claim 5, characterized in that Also includes: receiving a data request sent by a second device of a requesting automobile enterprise; Determine the requested vehicle manufacturer according to the data request; Sending the data request to a second device corresponding to the requested automobile enterprise; Upon receiving the consent sharing instruction sent by the requested automobile enterprise, selecting the encrypted data corresponding to the data request from the encrypted database; The encrypted data is sent to a second device corresponding to the requesting vehicle manufacturer.

7. A method for protecting privacy data in autonomous driving based on federated learning, characterized in that: The second equipment used in the target car companies includes: Perform compliance data processing on the autonomous driving database corresponding to the target car company and output compliant federal data; Perform machine learning and / or deep learning on the compliant federal data to generate a pre-trained model for the automobile enterprise; The model parameters corresponding to the car manufacturer's pre-trained model and the autonomous driving database are sent to the first device.

8. The method according to claim 7, characterized in that Also includes: receiving encrypted data sent by the first device; The car company's pre-trained model is optimized based on the encrypted data to generate a car company's accurate model.

9. The method according to claim 7, characterized in that: The step of performing compliance data processing on the autonomous driving database corresponding to the target vehicle enterprise and outputting compliant federal data includes: Based on the target car company signing a local data normalization compliance agreement between all parties, the target car company's corresponding autonomous driving database is subjected to multi-party data normalization processing, and a normalized database is output; Performing local preprocessing on the normalized database and outputting preprocessed data; Perform federated view normalization and compliance processing on the preprocessed data to output compliant federated data.

10. An autonomous driving privacy data protection device based on federated learning, characterized in that: It includes a first device applied to a collaborating party and at least two second devices, each of which is applied to a target automobile enterprise; The first device is communicatively connected with the second device; The device is applied to a first device; and comprises: A first determination module is used for: for any one of at least two target automobile companies: using the autonomous driving database corresponding to the target automobile company and the model parameters of the automobile company's pre-trained model as the privacy database corresponding to the target automobile company; wherein the automobile company's pre-trained model is used to indicate a model generated by a second device corresponding to the target automobile company through machine learning and / or deep learning based on the autonomous driving database; the autonomous driving database at least includes: the automobile company name, user ID, and autonomous driving related institutional regulations; An encryption processing module is used to perform federal co-creation encryption processing on the privacy databases corresponding to at least two target automobile companies to generate an encrypted sample database; The model generation module is used to perform federated learning on the pre-trained model of the collaborating party based on the encrypted sample database to generate a federated encryption model.

11. An autonomous driving privacy data protection device based on federated learning, characterized in that: The second equipment used in the target car companies includes: A compliance data module, used to perform compliance data processing on the autonomous driving database corresponding to the target vehicle enterprise and output compliant federal data; A pre-training module, used to perform machine learning and / or deep learning on the compliant federal data to generate a pre-training model for the automobile enterprise; A sending module is used to send the model parameters corresponding to the car manufacturer's pre-trained model and the autonomous driving database to the first device.

12. A computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method according to any one of claims 1 to 6 or any one of claims 7 to 9 is implemented.

Images