The overall architecture system of the joint learning engine based on the Internet of Things

Abstract

The present invention discloses an overall architecture system of a joint learning engine based on the Internet of Things, which belongs to the field of joint learning technology, and includes a cloud service end and a joint learning engine. The cloud service end is used to mobilize multiple joint participants to the joint learning engine, and the joint learning engine is used to respond to applications based on the joint learning architecture and respond to the needs of the target client through a standard interface. The overall architecture system of the joint learning engine based on the Internet of Things of the present invention, the cloud service provides a stable and reliable foundation for the joint learning engine service, ensures that the joint learning engine service can better systematically mobilize and manage multiple joint participants, and empower the ecology. The joint learning engine service provides intelligent joint learning for the joint learning application through a standard interface. The joint learning application is used in various fields to improve the efficiency of joint learning in various fields, effectively use data and machine learning modeling, and improve the efficiency of joint learning.

Description

Technical Field

[0001] The present invention relates to the field of joint learning technology, and in particular to an overall architecture system of a joint learning engine based on the Internet of Things. Background Art

[0002] Federated learning, also known as federated learning, is a machine learning framework that can effectively help multiple institutions use data and conduct machine learning modeling while meeting the requirements of user privacy protection, data security, and government regulations.

[0003] As a distributed machine learning paradigm, federated learning can effectively solve the problem of data silos, allowing participants to jointly model without sharing data, and can technically break data silos and achieve AI collaboration. Federated learning defines a machine learning framework, under which the problem of different data owners collaborating without exchanging data is solved by designing a virtual model. The virtual model is the optimal model for all parties to aggregate data together, and each region serves local goals based on the model. Federated learning requires that the modeling result should be infinitely close to the traditional model, that is, the result of aggregating the data of multiple data owners in one place for modeling. Under the federation mechanism, the identity and status of each participant are the same, and a shared data strategy can be established. Since the data is not transferred, it will not leak user privacy or affect data specifications. In order to protect data privacy and meet legal and compliance requirements.

[0004] The Internet of Things refers to the use of various devices and technologies such as information sensors, radio frequency identification technology, global positioning system, infrared sensors, laser scanners, etc. to collect any objects or processes that need to be monitored, connected, and interacted in real time, and collect various required information such as sound, light, heat, electricity, mechanics, chemistry, biology, and location. Through various possible network access, it realizes ubiquitous connection between objects and objects and between objects and people, and realizes intelligent perception, identification, and management of objects and processes. The Internet of Things is an information carrier based on the Internet, traditional telecommunications networks, etc. It allows all ordinary physical objects that can be independently addressed to form an interconnected network.

[0005] The overall architecture system of the current joint learning engine based on the Internet of Things cannot effectively mobilize and manage multiple joint participants, making the efficiency of joint learning low. Summary of the invention

[0006] The purpose of the present invention is to provide an overall architecture system of a joint learning engine based on the Internet of Things. The cloud infrastructure provides guarantee for cloud services, and cloud services provide a stable and reliable foundation for joint learning engine services, ensuring that the joint learning engine services can better systematically mobilize and manage multiple joint participants, and finally deposit the jointly trained models in the model library to empower the ecology. The joint learning engine service provides intelligent joint learning for joint learning applications through standard interfaces, and the framework has strong fusion capabilities. Joint learning applications are used in various fields to improve the efficiency of joint learning in various fields. The various architectures cooperate and coordinate with each other to effectively use data and machine learning modeling, which can improve the efficiency of joint learning and solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] The overall architecture system of the federated learning engine based on the Internet of Things includes a cloud server and a federated learning engine.

[0009] The cloud server is used to mobilize multiple federation participants to the federated learning engine;

[0010] The federated learning engine is used to respond to applications based on the federated learning architecture and to meet the needs of target clients through standard interfaces.

[0011] Furthermore, the cloud service end includes a container service, an API gateway and a message queue, wherein: the container service is used to provide high-performance and scalable container application management services, supports application lifecycle management using Docker containers, provides multiple application publishing methods and continuous delivery capabilities and supports microservice architecture. The container service simplifies the construction of container management clusters, integrates Alibaba Cloud virtualization, storage, network and security capabilities, and creates the best operating environment for Docker cloud. The API gateway is used to implement an API gateway to take over all ingress traffic and forward all user requests to the backend cloud server. The API gateway does more than just simple forwarding, and also makes some extensions for traffic. By introducing the API gateway, the client only needs to interact with the API gateway instead of communicating with the interfaces of each business party separately. The message queue is used to save messages during the transmission of messages. The message queue is a linked list of messages. The message can be regarded as a record with a specific format and a specific priority. A process with write permission to the message queue can add new messages to the message queue according to certain rules, and a process with read permission to the message queue can read messages from the message queue.

[0012] Furthermore, the cloud service end also includes a database service and a service pipeline. The database service is used to provide database resources to one or more tenants in the form of standard services. The database service provides users with a unified database access service, which shields the underlying heterogeneous database and provides a simple and convenient database access interface for upper-level applications, isolating the application and the database, reducing coupling, enhancing the flexibility and robustness of the system, and providing the possibility of enhancing data access control. The service pipeline is used to provide reliable services with high concurrency, high reliability, easy operation and maintenance, and consistency.

[0013] Furthermore, the federated learning engine includes federated learning basic algorithms, federated learning key technologies, privacy protection key technologies, core algorithm functional components and service encapsulation and model management, wherein: the federated learning basic algorithm provides corresponding basic algorithms for the federated learning engine services, the federated learning key technologies provide corresponding key technologies for the federated learning engine services, the privacy protection key technologies provide corresponding privacy protection for the federated learning engine services, the core algorithm functional components provide corresponding core algorithms for the federated learning engine services, and the service encapsulation model management provides corresponding service encapsulation models for the federated learning engine services.

[0014] Furthermore, the joint learning engine includes a joint strategy, an aggregation strategy, a security strategy, a distributed exception strategy, engine control, coordination of training tasks, a communication protocol and model library management. The joint strategy is used to provide corresponding policy support for joint learning. The aggregation strategy is used to attach one or more aggregation nodes to the message flow to control the timeout value or event storage at runtime. The security policy is a set of rules for all security-related activities in a certain security area. These rules are established by a security authority established in this security area and are described, implemented or realized by a security control agency. The distributed exception strategy is used to detect anomalies in joint learning. The engine control is used to control the operation of the entire engine system. The coordination of training tasks is used to coordinate the joint learning training tasks to make them fully cooperate. The communication protocol is used to agree on the data transmission control of the communicating parties. The agreement includes unified regulations on data format, synchronization method, transmission speed, transmission steps, error detection and correction method, and control character definition, which must be complied with by the communicating parties. The model library management is used to manage the database.

[0015] Furthermore, the basic algorithm of federated learning includes federated machine learning and federated deep learning. The key technologies of federated learning include supporting distributed remote parallel computing, independent deployment prediction of terminal models, and unified terminal-cloud models. The key technologies of privacy protection include secure multi-party computing MPC, homomorphic encryption HKE, and differential privacy DP. The research on secure multi-party computing is mainly aimed at the problem of how to securely calculate an agreed function without a trusted third party. Secure multi-party computing is the cryptographic basis for many applications such as electronic elections, threshold signatures, and electronic auctions. A secure multi-party computing protocol is called information-theoretic secure or unconditionally secure if it is secure against attackers with unlimited computing power. It is also called information-theoretic secure or unconditionally secure if it is secure against attackers with polynomial computing power. If a message is safe from attackers with strong capabilities, it is called cryptographically secure or conditionally secure. Homomorphic encryption is a cryptographic technique based on the computational complexity theory of mathematical problems. The data that has been homomorphically encrypted is processed to obtain an output, and this output is decrypted. The result is the same as the output obtained by processing the unencrypted original data in the same way. Differential privacy is a method in cryptography that aims to maximize the accuracy of data queries when querying from a statistical database while minimizing the chance of identifying its records. The core algorithm functional components include data loading, general model training, model evaluation components, and application model components. Service encapsulation and model management include various interface APIs, model publishing, model monitoring, and task log management.

[0016] Furthermore, the model library management includes the following steps:

[0017] S10: Collecting the joint learning model through the model collection unit, and transmitting the collected model to the model detection unit;

[0018] S20: The model detection unit detects the model by clustering, conversion, statistical description and causal analysis, and transmits the detection result to the model execution unit;

[0019] S30: If the collected model has been stored in the model database, the model execution unit transmits an instruction to the model cleaning unit, and the collected and stored model is cleared by the model cleaning unit;

[0020] S40: If the collected model is not stored in the model database, the model execution unit transmits an instruction to the model management unit, and the model management unit stores the collected model that has not been stored.

[0021] Furthermore, the model data collected in S40 are sorted in a clustering manner, and the sorted model data are compressed and fused before being stored in the model database.

[0022] Furthermore, the S40 includes the following steps:

[0023] S401: Detecting the model database through the database detection unit, and transmitting the detection result to the database execution unit;

[0024] S402: If the memory of the model database is full, the database execution unit transmits an instruction to the database cleaning unit, and the database cleaning unit clears the model stored in the database;

[0025] S403: If the memory of the model database is not full, the database execution unit transmits an instruction to the database management unit, and the model is stored in the model database through the database management unit.

[0026] Furthermore, in S402, the models stored in the model database are cleared according to a priority algorithm.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] The overall architecture system of the federated learning engine based on the Internet of Things of the present invention, the cloud infrastructure provides guarantee for the cloud service, and the cloud service provides a stable and reliable foundation for the federated learning engine service, ensuring that the federated learning engine service can better systematically mobilize and manage multiple joint participants, and finally deposit the jointly trained model in the model library to empower the ecology. The federated learning engine service provides intelligent federated learning for the federated learning application through a standard interface, and the framework has strong fusion capabilities. The federated learning application is used in various fields to improve the efficiency of federated learning in various fields. The various architectures cooperate and coordinate with each other to effectively use data and machine learning modeling, which can improve the efficiency of federated learning. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Figure 1 It is a framework diagram of the overall architecture system of the Internet of Things-based joint learning engine of the present invention;

[0030] Figure 2 A flowchart of the model library management of the present invention;

[0031] Figure 3 A module diagram for model library management of the present invention;

[0032] Figure 4 An algorithm diagram for model library management of the present invention;

[0033] Figure 5 A flowchart stored for the joint learning model of the present invention;

[0034] Figure 6 A module diagram stored for the joint learning model of the present invention;

[0035] Figure 7An algorithm diagram stored for the joint learning model of the present invention. DETAILED DESCRIPTION

[0036] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are 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 ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0037] Embodiment 1

[0038] See also Figure 1 , the overall architecture system of the federated learning engine based on the Internet of Things, including the cloud server and the federated learning engine,

[0039] The cloud server is used to mobilize multiple federation participants to the federated learning engine;

[0040] The federated learning engine is used to respond to applications based on the federated learning architecture and to meet the needs of target clients through standard interfaces.

[0041] The cloud services include container services, API gateways, message queues, database services, and service pipelines, among which: container services are used to provide high-performance, scalable container application management services, support application lifecycle management with Docker containers, provide multiple application publishing methods and continuous delivery capabilities, and support microservice architecture. Container services simplify the construction of container management clusters, integrate Alibaba Cloud virtualization, storage, network, and security capabilities, and create the best operating environment for Docker in the cloud. API gateways are used to implement an API gateway that takes over all ingress traffic and forwards all user requests to the backend cloud server. API gateways do more than just simple forwarding, but also make some extensions for traffic. By introducing API gateways, clients only need to interact with API gateways instead of communicating with the interfaces of each business party separately. Message The queue is used to save messages during the transmission process of the message. The message queue is a linked list of messages. The message can be regarded as a record with a specific format and a specific priority. The process with write permission to the message queue can add new messages to the message queue according to certain rules, and the process with read permission to the message queue can read messages from the message queue. The database service is used to provide database resources to one or more tenants in the form of standard services. The database service provides users with a unified database access service, which shields the underlying heterogeneous database and provides a simple and convenient database access interface for upper-level applications, isolating the application and database, reducing coupling, enhancing the flexibility and robustness of the system, and providing the possibility of enhancing data access control. The service pipeline is used to provide reliable services with high concurrency, high reliability, easy operation and maintenance, and consistency.

[0042] The federated learning engine service includes federated learning basic algorithms, federated learning key technologies, privacy protection key technologies, core algorithm functional components, service encapsulation and model management, joint strategies, aggregation strategies, security strategies, distributed exception strategies, engine control, coordination of training tasks, communication protocols and model library management, wherein: the federated learning basic algorithm provides the corresponding basic algorithm for the federated learning engine service, the federated learning key technology provides the corresponding key technology for the federated learning engine service, the privacy protection key technology provides the corresponding privacy protection for the federated learning engine service, the core algorithm functional component provides the corresponding core algorithm for the federated learning engine service, the service encapsulation model management provides the corresponding service encapsulation model for the federated learning engine service, the federated strategy is used to provide corresponding strategy support for federated learning, and the aggregation strategy is used to attach one or more aggregations to the message stream. The node is used to control the timeout value or event storage at runtime. The security policy is a set of rules for all security-related activities in a security area (a security area usually refers to a series of processing and communication resources belonging to an organization). These rules are established by a security authority established in this security area and described, implemented or realized by the security control agency. The distributed anomaly strategy is used to detect anomalies in joint learning. The control of the engine is used to control the operation of the entire engine system. The coordination of training tasks is used to coordinate the joint learning training tasks to make them fully cooperate. The communication protocol is used to agree on the data transmission control of the communicating parties. The agreement includes unified regulations on data format, synchronization method, transmission speed, transmission steps, error detection and correction methods, and control character definitions. Both communicating parties must comply with them. Model library management is used to manage the database.

[0043] The basic algorithms of federated learning include federated machine learning and federated deep learning. The key technologies of federated learning include supporting distributed remote parallel computing, independent deployment prediction of models on the end, and unified models on the end and cloud. The key technologies of privacy protection include secure multi-party computing MPC, homomorphic encryption HKE, and differential privacy DP. The research on secure multi-party computing is mainly aimed at the problem of how to securely calculate an agreed function without a trusted third party. Secure multi-party computing is the cryptographic basis for many applications such as electronic elections, threshold signatures, and electronic auctions. A secure multi-party computing protocol is called information-theoretic secure or unconditionally secure if it is secure against attackers with unlimited computing power. It is also called information-theoretic secure or unconditionally secure if it is secure against attackers with polynomial computing power. If it is safe from attackers, it is called cryptographic security or conditional security. Homomorphic encryption is a cryptographic technology based on the computational complexity theory of mathematical problems. The data that has been homomorphically encrypted is processed to obtain an output. This output is decrypted, and the result is the same as the output obtained by processing the unencrypted original data in the same way. Differential privacy is a means in cryptography that aims to provide a method that maximizes the accuracy of data queries when querying from a statistical database while minimizing the chance of identifying its records. The core algorithm functional components include data loading, general model training, model evaluation components, and application model components. Service encapsulation and model management include various interface APIs, model publishing, model monitoring, and task log management.

[0044] See also Figure 2-Figure 4 , the model library management includes the following steps:

[0045] S10: Collecting the joint learning model through the model collection unit, and transmitting the collected model to the model detection unit;

[0046] S20: The model detection unit detects the model by clustering, conversion, statistical description and causal analysis, and transmits the detection result to the model execution unit;

[0047] S30: If the collected model has been stored in the model database, the model execution unit transmits an instruction to the model cleaning unit, and the collected and stored model is cleared by the model cleaning unit;

[0048] S40: If the collected model is not stored in the model database, the model execution unit transmits instructions to the model management unit, and the collected model that has not been stored is stored through the model management unit. The collected model data is sorted in a clustering manner, and the sorted model data is compressed and fused and stored in the model database.

[0049] Embodiment 2

[0050] See also Figure 1, the overall architecture system of the federated learning engine based on the Internet of Things, including the cloud server and the federated learning engine,

[0051] The cloud server is used to mobilize multiple federation participants to the federated learning engine;

[0052] The federated learning engine is used to respond to applications based on the federated learning architecture and to meet the needs of target clients through standard interfaces.

[0053] The cloud services include container services, API gateways, message queues, database services, and service pipelines, among which: container services are used to provide high-performance, scalable container application management services, support application lifecycle management with Docker containers, provide multiple application publishing methods and continuous delivery capabilities, and support microservice architecture. Container services simplify the construction of container management clusters, integrate Alibaba Cloud virtualization, storage, network, and security capabilities, and create the best operating environment for Docker in the cloud. API gateways are used to implement an API gateway that takes over all ingress traffic and forwards all user requests to the backend cloud server. API gateways do more than just simple forwarding, but also make some extensions for traffic. By introducing API gateways, clients only need to interact with API gateways instead of communicating with the interfaces of each business party separately. Message The queue is used to save messages during the transmission process of the message. The message queue is a linked list of messages. The message can be regarded as a record with a specific format and a specific priority. The process with write permission to the message queue can add new messages to the message queue according to certain rules, and the process with read permission to the message queue can read messages from the message queue. The database service is used to provide database resources to one or more tenants in the form of standard services. The database service provides users with a unified database access service, which shields the underlying heterogeneous database and provides a simple and convenient database access interface for upper-level applications, isolating the application and database, reducing coupling, enhancing the flexibility and robustness of the system, and providing the possibility of enhancing data access control. The service pipeline is used to provide reliable services with high concurrency, high reliability, easy operation and maintenance, and consistency.

[0054] The federated learning engine service includes federated learning basic algorithms, federated learning key technologies, privacy protection key technologies, core algorithm functional components, service encapsulation and model management, joint strategies, aggregation strategies, security strategies, distributed exception strategies, engine control, coordination of training tasks, communication protocols and model library management, wherein: the federated learning basic algorithm provides the corresponding basic algorithm for the federated learning engine service, the federated learning key technology provides the corresponding key technology for the federated learning engine service, the privacy protection key technology provides the corresponding privacy protection for the federated learning engine service, the core algorithm functional component provides the corresponding core algorithm for the federated learning engine service, the service encapsulation model management provides the corresponding service encapsulation model for the federated learning engine service, the federated strategy is used to provide corresponding strategy support for federated learning, and the aggregation strategy is used to attach one or more aggregations to the message stream. The node is used to control the timeout value or event storage at runtime. The security policy is a set of rules for all security-related activities in a security area (a security area usually refers to a series of processing and communication resources belonging to an organization). These rules are established by a security authority established in this security area and described, implemented or realized by the security control agency. The distributed anomaly strategy is used to detect anomalies in joint learning. The control of the engine is used to control the operation of the entire engine system. The coordination of training tasks is used to coordinate the joint learning training tasks to make them fully cooperate. The communication protocol is used to agree on the data transmission control of the communicating parties. The agreement includes unified regulations on data format, synchronization method, transmission speed, transmission steps, error detection and correction methods, and control character definitions. Both communicating parties must comply with them. Model library management is used to manage the database.

[0055] The basic algorithms of federated learning include federated machine learning and federated deep learning. The key technologies of federated learning include supporting distributed remote parallel computing, independent deployment prediction of models on the end, and unified models on the end and cloud. The key technologies of privacy protection include secure multi-party computing MPC, homomorphic encryption HKE, and differential privacy DP. The research on secure multi-party computing is mainly aimed at the problem of how to securely calculate an agreed function without a trusted third party. Secure multi-party computing is the cryptographic basis for many applications such as electronic elections, threshold signatures, and electronic auctions. A secure multi-party computing protocol is called information-theoretic secure or unconditionally secure if it is secure against attackers with unlimited computing power. It is also called information-theoretic secure or unconditionally secure if it is secure against attackers with polynomial computing power. If it is safe from attackers, it is called cryptographic security or conditional security. Homomorphic encryption is a cryptographic technology based on the computational complexity theory of mathematical problems. The data that has been homomorphically encrypted is processed to obtain an output. This output is decrypted, and the result is the same as the output obtained by processing the unencrypted original data in the same way. Differential privacy is a means in cryptography that aims to provide a method that maximizes the accuracy of data queries when querying from a statistical database while minimizing the chance of identifying its records. The core algorithm functional components include data loading, general model training, model evaluation components, and application model components. Service encapsulation and model management include various interface APIs, model publishing, model monitoring, and task log management.

[0056] The federated learning engine service includes some basic algorithms for federated learning. Of course, these algorithms are different from traditional machine learning. Instead, they are incorporated into the federated machine learning and federated deep learning of the entire federated learning framework. In addition, they must support key technologies such as distributed parallel computing and unified end-cloud models. In addition, there are some ways to protect privacy. In order to ensure smooth information exchange between the engine and users, a series of joint strategies, aggregation strategies, security strategies, distributed exception strategies, etc. are also provided; in addition, there is engine control, task scheduling and coordination, and model library management, etc. These are all to enable the engine to better systematically mobilize and manage multiple joint participants, and ultimately deposit the jointly trained models in the model library to empower the ecosystem. In addition, as a service, there are also some functional component abstractions of core algorithms, as well as service encapsulation and management.

[0057] See also Figure 2-Figure 4 , the model library management includes the following steps:

[0058] S10: Collecting the joint learning model through the model collection unit, and transmitting the collected model to the model detection unit;

[0059] S20: The model detection unit detects the model by clustering, conversion, statistical description and causal analysis, and transmits the detection result to the model execution unit;

[0060] S30: If the collected model has been stored in the model database, the model execution unit transmits an instruction to the model cleaning unit, and the collected and stored model is cleared by the model cleaning unit;

[0061] S40: If the collected model is not stored in the model database, the model execution unit transmits instructions to the model management unit, and the collected model that has not been stored is stored through the model management unit. The collected model data is sorted in a clustering manner, and the sorted model data is compressed and fused and stored in the model database.

[0062] See also Figure 5-Figure 7 , the S40 comprises the following steps:

[0063] S401: Detecting the model database through the database detection unit, and transmitting the detection result to the database execution unit;

[0064] S402: If the memory of the model database is full, the database execution unit transmits an instruction to the database cleaning unit, and the database cleaning unit clears the models stored in the database, and clears the models stored in the model database according to the priority algorithm;

[0065] S403: If the memory of the model database is not full, the database execution unit transmits an instruction to the database management unit, and the model is stored in the model database through the database management unit.

[0066] To sum up, in the overall architecture system of the Internet of Things-based federated learning engine of the present invention, the cloud infrastructure provides guarantee for the cloud service, and the cloud service provides a stable and reliable foundation for the federated learning engine service, ensuring that the federated learning engine service can better systematically mobilize and manage multiple joint participants, and finally deposit the jointly trained model in the model library to empower the ecology. The federated learning engine service provides intelligent federated learning for federated learning applications through standard interfaces, and the framework has strong fusion capabilities. Federated learning applications are used in various fields to improve the efficiency of federated learning in various fields. The various architectures cooperate and coordinate with each other to effectively use data and machine learning modeling, which can improve the efficiency of federated learning.

[0067] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. The overall architecture system of the joint learning engine based on the Internet of Things includes a cloud service end and a joint learning engine, which is characterized by: The cloud server is used to mobilize multiple federation participants to the federated learning engine; The federated learning engine is used to respond to applications based on the federated learning architecture and respond to the needs of target clients through standard interfaces; The joint learning engine includes model library management, and the model library management includes the following steps: S10: Collecting the joint learning model through the model collection unit, and transmitting the collected model to the model detection unit; S20: The model detection unit detects the model by clustering, conversion, statistical description and causal analysis, and transmits the detection result to the model execution unit; S30: If the collected model has been stored in the model database, the model execution unit transmits an instruction to the model cleaning unit, and the collected and stored model is cleared by the model cleaning unit; S40: If the collected model is not stored in the model database, the model execution unit transmits an instruction to the model management unit, and the model management unit stores the collected model that has not been stored.

2. The overall architecture system of the joint learning engine based on the Internet of Things as claimed in claim 1, characterized in that: The cloud service end includes container service, API gateway and message queue, among which: container service is used to provide high-performance and scalable container application management service, supports application lifecycle management using Docker containers, provides multiple application publishing methods and continuous delivery capabilities and supports microservice architecture. Container service simplifies the construction of container management clusters.

3. The overall architecture system of the joint learning engine based on the Internet of Things as claimed in claim 2, characterized in that: The cloud service end also includes a database service and a service pipeline. The database service is used to provide database resources to one or more tenants in the form of standard services. The database service provides users with a unified database access service.

4. The overall architecture system of the joint learning engine based on the Internet of Things as claimed in claim 3, characterized in that: The federated learning engine includes federated learning basic algorithms, federated learning key technologies, privacy protection key technologies, core algorithm functional components and service encapsulation and model management, wherein: the federated learning basic algorithm provides the corresponding basic algorithm for the federated learning engine service, the federated learning key technology provides the corresponding key technology for the federated learning engine service, the privacy protection key technology provides the corresponding privacy protection for the federated learning engine service, the core algorithm functional component provides the corresponding core algorithm for the federated learning engine service, and the service encapsulation model management provides the corresponding service encapsulation model for the federated learning engine service.

5. The overall architecture system of the joint learning engine based on the Internet of Things as claimed in claim 4, characterized in that: The joint learning engine also includes joint strategies, aggregation strategies, security strategies, distributed exception strategies, engine control, coordination of training tasks and communication protocols. The joint strategies are used to provide corresponding policy support for joint learning. The aggregation strategies are used to attach one or more aggregation nodes to the message flow to control the timeout value or event storage at runtime. The security strategy is a set of rules for all security-related activities in a certain security area. These rules are established by a security authority established in this security area and are described, implemented or realized by the security control agency. The distributed exception strategy is used to detect anomalies in joint learning. The engine control is used to control the operation of the entire engine system. The coordination of training tasks is used to coordinate the joint learning training tasks to make them fully cooperate. The communication protocol is used to agree on the data transmission control between the communicating parties. The agreement includes unified regulations on data format, synchronization method, transmission speed, transmission steps, error detection and correction method and control character definition, which must be complied with by both communicating parties. Model library management is used to manage the database.

6. The overall architecture system of the joint learning engine based on the Internet of Things as claimed in claim 5, characterized in that: The basic algorithms of federated learning include federated machine learning and federated deep learning. The key technologies of federated learning include support for distributed remote parallel computing, independent deployment prediction of end-to-end models, and unified end-to-cloud models. The key technologies of privacy protection include secure multi-party computing MPC, homomorphic encryption HKE, and differential privacy DP. The core algorithm functional components include data loading, general model training, model evaluation components, and application model components. Service encapsulation and model management include various interface APIs, model publishing, model monitoring, and task log management.

7. The overall architecture system of the joint learning engine based on the Internet of Things as claimed in claim 6, characterized in that: The model data collected in S40 are sorted in a clustering manner, and the sorted model data are compressed and fused and then stored in the model database.

8. The overall architecture system of the joint learning engine based on the Internet of Things as claimed in claim 7, characterized in that: The S40 includes the following steps: S401: Detecting the model database through the database detection unit, and transmitting the detection result to the database execution unit; S402: If the memory of the model database is full, the database execution unit transmits an instruction to the database cleaning unit, and the database cleaning unit clears the model stored in the database; S403: If the memory of the model database is not full, the database execution unit transmits an instruction to the database management unit, and the model is stored in the model database through the database management unit.

9. The overall architecture system of the joint learning engine based on the Internet of Things as claimed in claim 8, characterized in that: In S402, the models stored in the model database are cleared according to the priority algorithm.

Images