CN110688419A - Federated modeling system and federated modeling method - Google Patents
Federated modeling system and federated modeling method Download PDFInfo
- Publication number
- CN110688419A CN110688419A CN201910955276.8A CN201910955276A CN110688419A CN 110688419 A CN110688419 A CN 110688419A CN 201910955276 A CN201910955276 A CN 201910955276A CN 110688419 A CN110688419 A CN 110688419A
- Authority
- CN
- China
- Prior art keywords
- data
- modeling
- trained
- model
- unit
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/25—Integrating or interfacing systems involving database management systems
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/602—Providing cryptographic facilities or services
Abstract
The disclosure relates to the technical field of data modeling, in particular to a federated modeling system and a method thereof, wherein the federated modeling system comprises: a plurality of software and hardware integrated modeling devices and a central decryption device; wherein, software and hardware integration modeling equipment includes: the acquisition module is used for encrypting the local data acquired in the local database to obtain encrypted data; the modeling module is used for acquiring sample data and performing iterative update on parameters of the model to be trained according to the sample data to acquire a trained model; the central decryption device is used for receiving and decrypting the local data corresponding to each modeling module to obtain corresponding decrypted data, judging whether the training of the model to be trained is finished or not based on the integrated calculation of each decrypted data, and feeding back the corresponding decrypted data to each modeling device when the training is not finished. The method and the device can separate the data encryption process and the modeling process, reduce the development workload of a data party for ensuring the data interaction safety, and reduce the technical difficulty of federal modeling.
Description
Technical Field
The disclosure relates to the technical field of data modeling, in particular to a joint modeling system and a joint modeling method.
Background
With the rapid development of machine learning, machine learning is applied to various fields, for example, data mining, data classification, image recognition, and the like. In practical applications, a process of training a machine learning model may require a large amount of sample data, but sometimes the sample data may not belong to a single data party, so that model training is often required by means of federal modeling.
It is to be noted that the information disclosed in the above background section is only for enhancement of understanding of the background of the present disclosure, and thus may include information that does not constitute prior art known to those of ordinary skill in the art.
Disclosure of Invention
The purpose of the disclosure is to provide a joint modeling system and a joint modeling method, so as to reduce the technical difficulty of the joint modeling at least to a certain extent.
Additional features and advantages of the disclosure will be set forth in the detailed description which follows, or in part will be obvious from the description, or may be learned by practice of the disclosure.
According to a first aspect of the disclosure, a plurality of software and hardware integrated modeling devices and a center decryption device of a joint modeling system are provided;
wherein, the software and hardware integration modeling device is deployed in a data side, and comprises:
the acquisition module is used for encrypting the local data acquired in the local database of the data side to obtain encrypted data;
the modeling module is used for acquiring sample data and performing iterative update on parameters of the model to be trained according to the sample data to acquire a trained model; wherein the sample data comprises the encrypted data, external data and decrypted data;
the central decryption device is used for receiving and decrypting the local data corresponding to each modeling module to obtain corresponding decrypted data, judging whether the model to be trained is trained to be finished or not based on the integrated calculation of each decrypted data, and feeding back the corresponding decrypted data to each modeling device when the training is judged not to be finished;
and the external data is encryption parameters which are sent by other modeling modules in the federal modeling system and are used for jointly calculating the gradient.
In an exemplary embodiment of the disclosure, based on the foregoing scheme, the local data corresponding to the modeling module includes a loss function and a gradient calculated by the modeling module.
In an exemplary embodiment of the present disclosure, based on the foregoing solution, the modeling module includes:
the internal interface unit is used for acquiring the encrypted data sent by the acquisition module;
a storage unit for storing the encrypted data received by the internal interface unit;
the external interface unit is used for carrying out data interaction with other modeling equipment and the central decryption equipment;
and the data modeling unit is used for acquiring the sample data and performing iterative update on the parameters of the model to be trained according to the sample data to acquire the trained model.
In an exemplary embodiment of the present disclosure, based on the foregoing solution, the modeling module further includes:
and the safety verification unit is used for carrying out safety verification on the data modeling unit before the data modeling unit acquires the encrypted data from the storage unit.
In an exemplary embodiment of the present disclosure, based on the foregoing scheme, the external interface unit is further configured to issue the trained model to a model user.
In an exemplary embodiment of the present disclosure, based on the foregoing scheme, the acquisition module includes:
the data acquisition unit is used for acquiring the local data in the local database according to a preset acquisition standard;
and the encryption calculation unit is used for encrypting the local data according to a preset algorithm so as to obtain encrypted data.
In an exemplary embodiment of the disclosure, based on the foregoing scheme, the central decryption device is further configured to feed back modeling end information to each modeling device when it is determined that training is ended, so that each modeling device stops updating parameters of the model to be trained.
In an exemplary embodiment of the present disclosure, based on the foregoing scheme, the central decryption device includes:
the decryption calculation unit is used for carrying out decryption and integration calculation on the received local data to obtain decrypted data;
the result judging unit is used for judging whether the model is trained to be finished or not according to the decrypted data;
and the data feedback unit is used for respectively feeding the decrypted data back to each software and hardware integrated modeling device when the training is judged not to be finished.
In an exemplary embodiment of the present disclosure, based on the foregoing scheme, the central decryption device and any one of the software and hardware integrated modeling devices are deployed on the same data side.
In an exemplary embodiment of the present disclosure, based on the foregoing scheme, the central decryption device is deployed independently from a third party.
According to a second aspect of the present disclosure, a federated modeling method is provided, which is applied to the federated modeling system described in any one of the exemplary embodiments in the first aspect above; the federal modeling method comprises the following steps:
the following steps are executed through software and hardware integrated modeling equipment:
encrypting the local data acquired in the local database of the data side by using an acquisition module to obtain encrypted data;
acquiring sample data by using a modeling module, and performing iterative updating on parameters of a model to be trained according to the sample data to acquire a trained model; wherein the sample data comprises the encrypted data, external data and decrypted data;
receiving and decrypting local data corresponding to each modeling module by using central decryption equipment to obtain corresponding decrypted data, judging whether the model to be trained is trained to be finished or not based on the integrated calculation of each decrypted data, and feeding back the corresponding decrypted data to each modeling equipment when the training is judged not to be finished;
and the external data is encryption parameters which are sent by other modeling modules in the federal modeling system and are used for jointly calculating the gradient.
In an exemplary embodiment of the present disclosure, based on the foregoing scheme, the method further includes: and feeding back modeling end information to each modeling device by using a central decryption device when the training end is judged so as to enable each modeling device to stop updating the parameters of the model to be trained.
The technical scheme provided by the embodiment of the disclosure can have the following beneficial effects:
on one hand, the encryption process and the modeling process in the federal modeling are separated from the operation equipment through the software and hardware integrated equipment and are operated by the software and hardware integrated equipment, so that the workload of development for ensuring data interaction safety in the related technology is reduced, and the technical difficulty of the federal modeling is reduced; on the other hand, the central decryption device of the federal modeling system can directly realize the decryption step which is originally required to be carried out by a third party, thereby avoiding potential safety hazards caused by data transmission to the third party; in addition, because each data side is provided with the same software and hardware integrated modeling equipment, the specification of the acquired data can be controlled through an acquisition module of the modeling equipment, and training deviation caused by different data specifications is avoided.
It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and together with the description, serve to explain the principles of the disclosure. It is to be understood that the drawings in the following description are merely exemplary of the disclosure, and that other drawings may be derived from those drawings by one of ordinary skill in the art without the exercise of inventive faculty. In the drawings:
FIG. 1 is a schematic diagram illustrating a deployment of a federated modeling system in an exemplary embodiment of the present disclosure;
FIG. 2 schematically illustrates a composition diagram of a hardware-software integrated modeling apparatus in an exemplary embodiment of the present disclosure;
FIG. 3 schematically illustrates a schematic composition diagram of an acquisition module in an exemplary embodiment of the disclosure;
FIG. 4 schematically illustrates a compositional schematic of a modeling module in an exemplary embodiment of the disclosure;
FIG. 5 is a schematic diagram illustrating the deployment and data flow of a modeling module in an exemplary embodiment of the present disclosure;
FIG. 6 schematically illustrates a compositional schematic of another modeling module in an exemplary embodiment of the disclosure;
FIG. 7 is a schematic representation of the deployment and data flow of another modeling module in an exemplary embodiment of the present disclosure;
fig. 8 schematically illustrates a composition diagram of a central decryption device in an exemplary embodiment of the present disclosure;
fig. 9 schematically illustrates a deployment manner of a joint modeling system in an exemplary embodiment of the present disclosure.
In the figure: 1. soft and hard integrated modeling equipment; 11. an acquisition module; 111. a data acquisition unit; 112. an encryption calculation unit; 12. a modeling module; 121. an internal interface unit; 122. a storage unit; 123. an external interface unit; 124. a data modeling unit; 125. a security verification unit; 2. a central decryption device; 21. a decryption calculation unit; 22. a result judgment unit; 23. and a data feedback unit.
Detailed Description
Example embodiments will now be described more fully with reference to the accompanying drawings. Example embodiments may, however, be embodied in many different forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example 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.
The terms "a," "an," "the," and "said" are used to indicate the presence of one or more elements/components/etc.; the terms "comprising" and "having" are intended to be inclusive and mean that there may be additional elements/components/etc. other than the listed elements/components/etc.
Furthermore, the drawings are merely schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus their repetitive description will be omitted. Some of the block diagrams shown in the figures are functional entities and do not necessarily correspond to physically or logically separate entities. These functional entities may be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different networks and/or processor devices and/or microcontroller devices.
In the related federal modeling technology, when a data party performs the process of federal modeling, a set of complete software is generally required to be developed according to the data modeling technology and the data encryption technology, and then a plurality of data parties are combined to perform federal modeling. However, in the actual modeling process, since there may be differences in the software developed by the respective data parties, there may be differences in the data specifications provided by each data party. Meanwhile, because the operation of the development software completely depends on the operation equipment, the safety guarantee of the development software also completely depends on the safety mechanism of the operation equipment, and the interaction of data cannot be guaranteed, some safety mechanisms are often required to be separately developed in order to guarantee the safety of data transmission in the operation process of the development software.
In view of the above problems, in the exemplary embodiment, a federated modeling system is first provided, which may be applied to a process of federated modeling performed by multiple data parties. The above-mentioned federal modeling system may include a plurality of software and hardware integrated modeling apparatuses 1 and a central decryption apparatus 2, wherein:
the software and hardware integrated modeling device 1 is deployed on a data side and comprises:
the acquisition module 11 is configured to encrypt local data acquired in a local database of the data side to obtain encrypted data;
the modeling module 12 is configured to obtain sample data, and iteratively update parameters of the model to be trained according to the sample data to obtain a trained model; wherein the sample data comprises the encrypted data, external data and decrypted data;
the central decryption device 2 is configured to receive and decrypt local data corresponding to each modeling module to obtain corresponding decrypted data, determine whether the model to be trained is trained over based on integrated calculation of each decrypted data, and feed back corresponding decrypted data to each modeling device when it is determined that training is not over;
and the external data is encryption parameters which are sent by other modeling modules in the federal modeling system and are used for jointly calculating the gradient.
According to the federal modeling system provided in the exemplary embodiment, the encryption process and the modeling process in the federal modeling are separated from the operation equipment through the software and hardware integrated equipment 1 and are independently operated by the software and hardware integrated equipment 1, so that the workload of development for guaranteeing the data interaction safety in the related technology is reduced, and the technical difficulty of the federal modeling is reduced; on the other hand, the central decryption device 2 of the federal modeling system can directly realize decryption steps which are originally required to be carried out by a third party, so that potential safety hazards caused by data transmission to the third party are avoided; in addition, as the same software and hardware integrated modeling equipment 1 is deployed on each data side, the specification of the acquired data can be controlled by the acquisition module 11 of the software and hardware integrated modeling equipment 1, and training deviation caused by different data specifications is avoided.
In the following, the parts of the federal modeling system in the exemplary embodiment will be described in detail in conjunction with the accompanying drawings and embodiments:
the federal modeling system comprises a plurality of software and hardware integrated modeling devices 1 and a central decryption device 2. The number of the software and hardware integrated modeling devices 1 can be configured according to the actual number of the data parties, a corresponding software and hardware integrated modeling device 1 is deployed for each data party, and communication connection is established between the software and hardware integrated modeling devices and a local database of the data party. For example, referring to fig. 1, when 3 data parties model together, corresponding software and hardware integrated modeling devices a1, B1, and C1 may be deployed on the 3 data parties, respectively, and each software and hardware integrated modeling device establishes a communication connection with a local database of the corresponding data party.
The central decryption device 2 may be deployed together with any one software and hardware integrated modeling device 1 to a corresponding data party, or may be deployed separately to a unit of a third party; meanwhile, the central decryption device can perform data interaction with each software and hardware integrated modeling device in front. For example, the central decryption device 2 and one of the software and hardware integrated modeling devices a1 may be deployed together with the corresponding data party a; as another example, referring to fig. 1, when 3 data parties collectively model, the central decryption device 2 may be deployed to a third party other than the data parties.
Referring to fig. 2, the software and hardware integrated modeling apparatus 1 may include an acquisition module 11 and a modeling module 12. Wherein: the acquisition module 11 is configured to encrypt local data acquired in a local database of the data party to obtain encrypted data; the modeling module 12 is configured to obtain sample data, and iteratively update parameters of the model to be trained according to the sample data to obtain a trained model.
In one embodiment, referring to fig. 3, the collection module 11 may be deployed in a cloud computing unit on the data side to provide a supporting data collection service for the modeling module 12. Specifically, the acquisition module may include a data acquisition unit 111 and an encryption calculation unit 112. The data acquisition unit 111 is configured to acquire local data required for modeling in a local database of a data side according to a preset acquisition standard; the encryption calculation unit 112 is configured to encrypt the local data according to a preset algorithm to obtain encrypted data.
By setting the acquisition module 11 to encrypt the data required by modeling, a preprocessing process of independently encrypting the data required by modeling through development software or development codes in the related art can be avoided, and the workload of developers is reduced.
In an embodiment, the preset collection standard may be set according to different data types, or may be defined according to a requirement of a data party, which is not particularly limited in this disclosure.
In an embodiment, the sample data comprises encrypted data, external data and decrypted data. The encrypted data may be obtained by encrypting the acquired local data according to the acquisition module 11; the external data can be encryption parameters which are sent by other modeling modules 12 in the federal modeling system and are used for jointly calculating the gradient, and can comprise output parameters of models to be trained in other modeling modules 12; the decrypted data may be decryption parameters fed back to the software and hardware integrated modeling device 1 by the central decryption device 2 for updating parameters of the model to be trained, and may include a gradient of an objective function of the model to be trained to the modeling device. For example, when 2 data parties perform joint modeling, for a data party a, the external data refers to an output parameter B obtained by inputting encrypted data B of the data party B into a model B to be trained, which is sent by the data party B.
In one embodiment, referring to fig. 4, the modeling module 12 includes an internal interface unit 121, a storage unit 122, an external interface unit 123, and a data modeling unit 124. The internal interface unit 121 is configured to obtain encrypted data sent by the acquisition module; the storage unit 122 is used for storing the encrypted data received by the internal interface unit; the external interface unit 123 is used for data interaction with other modeling devices and the central decryption device; the data modeling unit 124 is configured to obtain the sample data and iteratively update parameters of the model to be trained according to the sample data to obtain the trained model.
Specifically, in the federal modeling process, the deployment and data flow of the modeling module 12 can be referred to as shown in fig. 5. With the modeling module of one data party as a view, after receiving the encrypted data sent by the acquisition module, the internal interface data 121 sends the encrypted data to the storage unit for storage, the data modeling unit 124 reads the encrypted data in the storage unit 122 to train the model to be trained, and sends the model output (external data for other data parties) to other data parties through the external interface unit 123; meanwhile, the external interface unit 123 acquires external data transmitted from other data parties, and the data modeling unit 124 calculates local data (loss function and gradient data) of the model to be trained based on the model output and the external data, and transmits the local data to the central decryption device 2 through the external interface unit 123 so that the central decryption device performs decryption and calculation based on the local data. Subsequently, the external interface data 123 receives the decryption data fed back by the central decryption device 2, and the data modeling unit 124 updates the parameters of the model to be trained according to the decryption data, so as to perform further training.
By arranging the independent storage unit 122, local data required in the modeling process can be independent of other local data, and security threats to other local data caused by directly reading the data required for modeling in a local database are avoided.
In addition, after the training of the model to be trained is finished, the external interface unit 123 may be further configured to issue the trained model to the model user.
In one embodiment, referring to fig. 6, the modeling module 12 further includes a security verification unit 125 for performing security verification on the data modeling unit before the data modeling unit obtains the encrypted data from the storage unit.
Specifically, in the federal modeling process, the deployment and data flow of the modeling module 12 can be referred to fig. 7. Taking the modeling module of one data side as a view, after receiving the encrypted data sent by the acquisition module, the internal interface data 121 sends the encrypted data to the storage unit for storage, the data modeling unit 124 performs verification through the safety verification unit 125, when the verification fails, the data modeling unit 124 may have potential safety hazards, and in order to ensure the safety of the encrypted data, the data modeling unit 124 is not allowed to read the data in the storage unit 122; after the verification is passed, the encrypted data in the storage unit 122 is read to train the model to be trained, and the output of the model (external data of other data parties) is sent to other data parties through the external interface unit 123; meanwhile, the external interface unit 123 acquires external data transmitted from other data parties, and the data modeling unit 124 calculates local data (loss function and gradient data) of the model to be trained based on the model output and the external data, and transmits the local data to the central decryption device 2 through the external interface unit 123 so that the central decryption device performs decryption and calculation based on the local data. Subsequently, the external interface data 123 receives the decryption data fed back by the central decryption device 2, and the data modeling unit 124 updates the parameters of the model to be trained according to the decryption data, so as to perform further training.
The security verification unit 125 is arranged in the modeling module 12 to perform security verification on the data modeling unit 124 before the data modeling unit 124 reads the encrypted data in the database, so that the security of the data is further improved.
In one embodiment, the actual architecture of the modeling module 12 may include a hardware layer and a software layer based on the configuration of the modeling module 12 and the units described above. The software layer can comprise a bottom operating system, a middle model compiling tool, safety software, an upper combined modeling algorithm, a federal modeling algorithm and other modeling tools; the system comprises a bottom operating system, a middle model compiling tool, a safety software, a combined modeling algorithm, a federal modeling algorithm and other modeling tools, wherein the bottom operating system is used for operating modeling hardware, the middle model compiling tool is used for compiling a model to be trained, the safety software is used for carrying out safety verification on a gateway, and the combined modeling algorithm, the federal modeling algorithm and other modeling tools are used for training the model to be trained; the hardware layer may include components such as a central processing unit, a graphics processor, a network interface, etc.; the central processor and the graphic processor are respectively used for processing data such as files and images, and the network interface is used for data interaction with other equipment.
The central decryption device 2 is configured to receive and decrypt the local data corresponding to each modeling module 12 to obtain corresponding decrypted data, determine whether the training of the model to be trained is completed based on the integrated calculation of each decrypted data, and feed back the corresponding decrypted data to each modeling device 12 when it is determined that the training is not completed. The local data corresponding to the modeling module 12 may include a loss function and a gradient calculated by the modeling module 12. For example, when 2 data parties perform joint modeling, the local data for the data party a means a loss function and a gradient for the model a to be trained, which are calculated based on the output parameter a and the output parameter B after the data party a acquires external data (the output parameter B output by the model B to be trained).
In an embodiment, the central decryption device is further configured to feed back information of modeling completion to each modeling device when it is determined that training is complete, so that each modeling device stops updating parameters of the model to be trained.
Specifically, decrypting the data means decrypting and integrating the local data sent by each modeling module 12 to obtain a loss function and a gradient corresponding to the federal modeling system. The central decryption device 2 may determine whether the training of the model to be trained is finished according to the decrypted data.
In one embodiment, referring to fig. 8, the central decryption device 2 may include a decryption calculation unit 21, a result judgment unit 22, and a data feedback unit 23. Wherein, the decryption calculation unit 21 may be configured to perform decryption and integration calculation on the received local data to obtain decrypted data; the result judging unit 22 may be configured to judge whether the model training is finished according to the decrypted data; the data feedback unit 23 may be configured to feed the decrypted data back to each software and hardware integrated modeling apparatus 1 when it is determined that training is not finished.
The following describes the modeling process of the federal modeling system according to the present disclosure in detail with reference to fig. 9, taking 3 data parties A, B, C as an example, where the central decryption device is deployed in a third party.
Taking a data party A as an example, an acquisition module A11 acquires data in a local database A of the data party A according to a preset standard, and encrypts the data to obtain encrypted data; after the modeling module A12 obtains the encrypted data, the encrypted data is input into the model A to be trained of the data side A to obtain corresponding output data A.
Before the second step, the data side B and the data side C correspondingly complete the steps in the first step to obtain output data B and output data C, and for the data side A, the output data B and the output data C are external data B and external data C respectively.
Secondly, taking the data side A as an example, the modeling module A12 calculates a loss function and a gradient value corresponding to the model A to be trained through the output data A, the external data B and the external data C, namely local data A; and sending the local data A to the central interface equipment through the communication network.
Before the third step, the data side B and the data side C correspondingly complete the steps in the second step to obtain the local data B and the local data C, and send the local data B and the local data C to the central decryption device 2.
Thirdly, the central decryption device 2 decrypts the received local data a, local data B and local data C to obtain decrypted data a, decrypted data B and decrypted data C. Calculating according to the decrypted data A, the decrypted data B and the decrypted data C to obtain an overall loss function and a gradient value, and judging whether the model training is finished or not according to the overall loss function and the gradient value; when the model training is judged to be finished, feeding back information of finishing modeling to modeling modules A12, B12 and C12; and when the model training is judged not to be finished, feeding back the decrypted data A, the decrypted data B and the decrypted data C to the corresponding modeling modules A12, B12 and C12 respectively.
And fourthly, taking the data side A as an example, after the modeling module A12 receives the decrypted data A, updating parameters of the model A to be trained according to the decrypted data A, and continuing to train the model A to be trained.
The present exemplary embodiment further provides a joint modeling method, which is applied to the above-mentioned joint modeling system, and includes:
the following steps are executed through software and hardware integrated modeling equipment:
encrypting the local data acquired in the local database of the data side by using an acquisition module to obtain encrypted data;
acquiring sample data by using a modeling module, and performing iterative updating on parameters of a model to be trained according to the sample data to acquire a trained model; wherein the sample data comprises the encrypted data, external data and decrypted data;
receiving and decrypting local data corresponding to each modeling module by using central decryption equipment to obtain corresponding decrypted data, judging whether the model to be trained is trained or not based on the integrated calculation of each decrypted data, and feeding back the corresponding decrypted data to each modeling equipment;
and the external data is encryption parameters which are sent by other modeling modules in the federal modeling system and are used for jointly calculating the gradient.
In an exemplary embodiment of the present disclosure, based on the foregoing scheme, the method further includes: and feeding back modeling end information to each modeling device by using a central decryption device when the training end is judged so as to enable each modeling device to stop updating the parameters of the model to be trained.
For details that are not disclosed in the embodiments of the method of the present disclosure, please refer to the embodiments of the federal modeling system described above for details that are not disclosed in the embodiments of the method of the present disclosure, because each step of the federal modeling method of the exemplary embodiments of the present disclosure corresponds to a device of the exemplary embodiments of the federal modeling system described above.
Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains. It is intended that the specification and examples be considered as exemplary only, with a true scope and spirit of the disclosure being indicated by the following claims.
Claims (12)
1. A federated modeling system is characterized in that the federated modeling system comprises a plurality of software and hardware integrated modeling devices and a central decryption device;
wherein, the software and hardware integration modeling device is deployed in a data side, and comprises:
the acquisition module is used for encrypting the local data acquired in the local database of the data side to obtain encrypted data;
the modeling module is used for acquiring sample data and performing iterative update on parameters of the model to be trained according to the sample data to acquire a trained model; wherein the sample data comprises the encrypted data, external data and decrypted data;
the central decryption device is used for receiving and decrypting the local data corresponding to each modeling module to obtain corresponding decrypted data, judging whether the model to be trained is trained to be finished or not based on the integrated calculation of each decrypted data, and feeding back the corresponding decrypted data to each modeling device when the training is judged not to be finished;
and the external data is encryption parameters which are sent by other modeling modules in the federal modeling system and are used for jointly calculating the gradient.
2. The system of claim 1, wherein the local data corresponding to the modeling module comprises a loss function and a gradient calculated by the modeling module.
3. The system of claim 1, wherein the modeling module comprises:
the internal interface unit is used for acquiring the encrypted data sent by the acquisition module;
a storage unit for storing the encrypted data received by the internal interface unit;
the external interface unit is used for carrying out data interaction with other modeling equipment and the central decryption equipment;
and the data modeling unit is used for acquiring the sample data and performing iterative update on the parameters of the model to be trained according to the sample data to acquire the trained model.
4. The system of claim 3, wherein the modeling module further comprises:
and the safety verification unit is used for carrying out safety verification on the data modeling unit before the data modeling unit acquires the encrypted data from the storage unit.
5. The system of claim 3, wherein the external interface unit is further configured to publish the trained model to a model user.
6. The system of claim 1, wherein the acquisition module comprises:
the data acquisition unit is used for acquiring the local data in the local database according to a preset acquisition standard;
and the encryption calculation unit is used for encrypting the local data according to a preset algorithm so as to obtain encrypted data.
7. The system according to claim 1, wherein the central decryption device is further configured to feed back information of modeling completion to each modeling device when training completion is determined, so that each modeling device stops updating parameters of the model to be trained.
8. The system of claim 1, wherein the central decryption device comprises:
the decryption calculation unit is used for carrying out decryption and integration calculation on the received local data to obtain decrypted data;
the result judging unit is used for judging whether the model is trained to be finished or not according to the decrypted data;
and the data feedback unit is used for respectively feeding the decrypted data back to each software and hardware integrated modeling device when the training is judged not to be finished.
9. The system according to claim 1, wherein the central decryption device is deployed on the same data side as any one of the software and hardware integrated modeling devices.
10. The system of claim 1, wherein the central decryption device is deployed independently from a third party.
11. A federated modeling method, which is applied to the federated modeling system of any one of claims 1 to 10; the federal modeling method comprises the following steps:
the following steps are executed through software and hardware integrated modeling equipment:
encrypting the local data acquired in the local database of the data side by using an acquisition module to obtain encrypted data;
acquiring sample data by using a modeling module, and performing iterative updating on parameters of a model to be trained according to the sample data to acquire a trained model; wherein the sample data comprises the encrypted data, external data and decrypted data;
receiving and decrypting local data corresponding to each modeling module by using central decryption equipment to obtain corresponding decrypted data, judging whether the model to be trained is trained to be finished or not based on the integrated calculation of each decrypted data, and feeding back the corresponding decrypted data to each modeling equipment when the training is judged not to be finished;
and the external data is encryption parameters which are sent by other modeling modules in the federal modeling system and are used for jointly calculating the gradient.
12. The method of claim 11, further comprising: and feeding back modeling end information to each modeling device by using a central decryption device when the training end is judged so as to enable each modeling device to stop updating the parameters of the model to be trained.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910955276.8A CN110688419A (en) | 2019-10-09 | 2019-10-09 | Federated modeling system and federated modeling method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910955276.8A CN110688419A (en) | 2019-10-09 | 2019-10-09 | Federated modeling system and federated modeling method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110688419A true CN110688419A (en) | 2020-01-14 |
Family
ID=69111809
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910955276.8A Pending CN110688419A (en) | 2019-10-09 | 2019-10-09 | Federated modeling system and federated modeling method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110688419A (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112801731A (en) * | 2021-01-06 | 2021-05-14 | 广东工业大学 | Federal reinforcement learning method for order taking auxiliary decision |
| CN113051586A (en) * | 2021-03-10 | 2021-06-29 | 北京沃东天骏信息技术有限公司 | Federal modeling system and method, and federal model prediction method, medium, and device |
| CN113592097A (en) * | 2021-07-23 | 2021-11-02 | 京东科技控股股份有限公司 | Federal model training method and device and electronic equipment |
| CN113902137A (en) * | 2021-12-06 | 2022-01-07 | 腾讯科技（深圳）有限公司 | Streaming model training method and device, computer equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2979064A1 (en) * | 2015-03-12 | 2016-09-15 | Fornetix Llc | Client services for applied key management systems and processes |
| CN109167695A (en) * | 2018-10-26 | 2019-01-08 | 深圳前海微众银行股份有限公司 | Alliance Network construction method, equipment and readable storage medium storing program for executing based on federation's study |
| CN109284313A (en) * | 2018-08-10 | 2019-01-29 | 深圳前海微众银行股份有限公司 | Federal modeling method, equipment and readable storage medium storing program for executing based on semi-supervised learning |
| CN110276210A (en) * | 2019-06-12 | 2019-09-24 | 深圳前海微众银行股份有限公司 | Based on the determination method and device of the model parameter of federation's study |
-
2019
- 2019-10-09 CN CN201910955276.8A patent/CN110688419A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2979064A1 (en) * | 2015-03-12 | 2016-09-15 | Fornetix Llc | Client services for applied key management systems and processes |
| CN109284313A (en) * | 2018-08-10 | 2019-01-29 | 深圳前海微众银行股份有限公司 | Federal modeling method, equipment and readable storage medium storing program for executing based on semi-supervised learning |
| CN109167695A (en) * | 2018-10-26 | 2019-01-08 | 深圳前海微众银行股份有限公司 | Alliance Network construction method, equipment and readable storage medium storing program for executing based on federation's study |
| CN110276210A (en) * | 2019-06-12 | 2019-09-24 | 深圳前海微众银行股份有限公司 | Based on the determination method and device of the model parameter of federation's study |
Non-Patent Citations (1)
| Title |
|---|
| 杨惠珍 等: ""基于CPN的联邦概念模型形式化建模与验证"", 《系统仿真学报》 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112801731A (en) * | 2021-01-06 | 2021-05-14 | 广东工业大学 | Federal reinforcement learning method for order taking auxiliary decision |
| CN113051586A (en) * | 2021-03-10 | 2021-06-29 | 北京沃东天骏信息技术有限公司 | Federal modeling system and method, and federal model prediction method, medium, and device |
| CN113592097A (en) * | 2021-07-23 | 2021-11-02 | 京东科技控股股份有限公司 | Federal model training method and device and electronic equipment |
| CN113592097B (en) * | 2021-07-23 | 2024-02-06 | 京东科技控股股份有限公司 | Training method and device of federal model and electronic equipment |
| CN113902137A (en) * | 2021-12-06 | 2022-01-07 | 腾讯科技（深圳）有限公司 | Streaming model training method and device, computer equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110688419A (en) | Federated modeling system and federated modeling method | |
| CN109284313B (en) | Federal modeling method, device and readable storage medium based on semi-supervised learning | |
| CN109255444B (en) | Federal modeling method and device based on transfer learning and readable storage medium | |
| CN109165725B (en) | Neural network federal modeling method, equipment and storage medium based on transfer learning | |
| CN110929886B (en) | Model training and predicting method and system | |
| CN108197891B (en) | Electronic signing device and method based on block chain | |
| CN113159327B (en) | Model training method and device based on federal learning system and electronic equipment | |
| US9232394B2 (en) | Authentication of phone caller identity | |
| CN111582508A (en) | Strategy making method and device based on federated learning framework and electronic equipment | |
| CN106796763B (en) | Secret computing system, relay device, methods therefor, and recording medium | |
| CN111628966B (en) | Data transmission method, system and device and data authorization method, system and device | |
| CN111340247A (en) | Longitudinal federated learning system optimization method, device and readable storage medium | |
| CN110417756B (en) | Cross-network data transmission method and device | |
| CN107368744A (en) | For updating the method and the equipment of measurement and control technology of fastener components | |
| CN112954050B (en) | Distributed management method and device, management equipment and computer storage medium | |
| JP2022033002A (en) | Bidding information processing method based on block chain, system, apparatus, and medium | |
| CN110019075A (en) | Log encryption method and decryption method and device | |
| KR20170082393A (en) | METHOD AND PLATFORM APPARATUS FOR INTERLOCKING WITH IoT TERMINAL AND SERVICE USING GRAPHIC USER INTERFACE | |
| CN110740038A (en) | Block chain and communication method, gateway, communication system and storage medium thereof | |
| CN112381000A (en) | Face recognition method, device, equipment and storage medium based on federal learning | |
| CN105955784A (en) | Elevator program upgrade and backup methods and systems, and mobile terminals | |
| CN112182635B (en) | Method, device, equipment and medium for realizing joint modeling | |
| CN106657348A (en) | File cooperative processing method and system based on two-dimensional bar code | |
| CN116502732B (en) | Federal learning method and system based on trusted execution environment | |
| US20240007287A1 (en) | Machine learning device, machine learning system, and machine learning method |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200114 |