A modular joint learning method and system

Abstract

The present disclosure relates to the field of machine learning technology, and provides a modular joint learning method and system. The method includes: providing a training data entry template to the participants according to the joint learning scenario selected by the participants; associating the training data with the joint learning scenario based on the training data fed back by the participants to obtain an associated logical relationship network; using the associated logical relationship network, establishing multiple joint learning tasks according to the joint learning scenario; determining at least one joint learning task applied by the participants as the target joint learning task; according to the target joint learning task, starting the training of the participant joint model corresponding to the joint learning scenario until the preset training end condition is met, and feeding back the final joint model to the participants. The present disclosure solves the problem that users need to spend a lot of time, operate inconveniently, and have high costs to find suitable participants and third parties, and provides users with a one-stop joint learning interactive service with good user experience.

Description

Technical Field

[0001] The present disclosure relates to the technical field of machine learning, and in particular to a modular joint learning method and system. Background Art

[0002] In recent years, machine learning has developed rapidly in the field of artificial intelligence applications, such as computer vision and automatic speech recognition. However, the success of these machine learning technologies, especially deep learning, depends on a large amount of data (also called big data).

[0003] Usually, the amount of data required to train artificial intelligence application models is very large. For example, Facebook's object detection system is trained with 350 million images from Instagram. However, in many application areas, the amount of data owned by a single enterprise is very limited, and it is often difficult to meet such a large amount of training data requirements. Although the problem of insufficient training data can be solved by collecting data from multiple companies and transmitting it to a central point (such as a data center), as artificial intelligence is applied in various industries, people are paying more and more attention to user privacy and data security protection. As a result, federated learning (joint learning) technology, which can ensure user data privacy and higher model performance, came into being and quickly became popular in the field of machine learning.

[0004] However, since federated learning is in an emerging development stage, many users may not understand it. Or some users who have a certain understanding of federated learning want to find suitable participants and third parties for collaborative training, but they often conduct various inquiries and screen participants and third parties online or offline. This process is very time-consuming, troublesome, and costly, so the user experience is very poor. At the same time, this has also become an important factor restricting the promotion, application and development of federated learning. Summary of the invention

[0005] In view of this, the embodiments of the present disclosure provide a modular joint learning service platform and system to solve the problem in the prior art that users need to spend a lot of time, operate inconveniently and at high costs to find suitable participants and third parties, which to a certain extent restricts the promotion, application and development of joint learning.

[0006] A first aspect of an embodiment of the present disclosure provides a modular joint learning method, including:

[0007] Provide participants with training data entry templates based on the joint learning scenarios selected by the participants;

[0008] Based on the training data fed back by the participants, the training data is associated with the joint learning scenario to obtain an associated logical relationship network;

[0009] Using the associated logical relationship network, multiple joint learning tasks are established according to the joint learning scenarios;

[0010] When a participant applies for at least one joint learning task, determining the at least one joint learning task applied for by the participant as a target joint learning task;

[0011] According to the target joint learning task, start training the joint model of the participants corresponding to the joint learning scenario. When the joint model meets the preset training end conditions, the final joint model is fed back to the participants.

[0012] A second aspect of the embodiments of the present disclosure provides a modular joint learning system, including:

[0013] A template providing module, configured to provide a training data entry template to a participant according to a joint learning scenario selected by the participant;

[0014] The association module is configured to associate the training data with the joint learning scenario based on the training data fed back by the participants to obtain an association logic relationship network;

[0015] The task establishment module is configured to establish a plurality of joint learning tasks according to the joint learning scenario by using the associated logical relationship network;

[0016] A target task determination module is configured to, when a participant applies for at least one joint learning task, determine at least one joint learning task applied for by the participant as a target joint learning task;

[0017] The joint training module is configured to start training the joint model of the participants corresponding to the joint learning scenario according to the target joint learning task, and when the joint model meets the preset training end conditions, the final joint model is fed back to the participants.

[0018] Compared with the prior art, the embodiments of the present disclosure have at least the following beneficial effects: providing a training data entry template to a participant according to a joint learning scenario selected by the participant; associating the training data with the joint learning scenario based on the training data fed back by the participant to obtain an associated logical relationship network; using the associated logical relationship network, establishing multiple joint learning tasks according to the joint learning scenario; when a participant applies for at least one joint learning task, determining at least one joint learning task applied for by the participant as a target joint learning task; according to the target joint learning task, starting to train a participant joint model corresponding to the joint learning scenario, and when the joint model meets the preset training end condition, feeding back the final joint model to the participant, and being able to provide users with a one-stop joint learning interactive service, including comprehensive services such as helping users quickly find suitable participants and third parties, which well solves the problem that users need to spend a lot of time, have troublesome operations, and have high costs in finding suitable participants and third parties; and, it can be realized from the perspective of protecting data sovereignty and data privacy, combined with the characteristics of the industrial Internet of Things, from the perspective of ecological users, so that users have an excellent experience of participating in joint learning and sharing results, thereby being able to promote the promotion, application and development of joint learning to a certain extent, and having broad market prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0020] Figure 1 is a schematic diagram of an application scenario of an embodiment of the present disclosure;

[0021] Figure 2 It is a flowchart of a modular joint learning method provided by an embodiment of the present disclosure;

[0022] Figure 3 is a schematic diagram of the structure of a modular joint learning system provided by an embodiment of the present disclosure;

[0023] Figure 4 It is a structural schematic diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0024] In the following description, specific details such as specific system structures and technologies are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present disclosure. However, it should be clear to those skilled in the art that the present disclosure may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to avoid obstructing the description of the present disclosure with unnecessary details.

[0025] Federated learning is the comprehensive use of multiple AI (Artificial Intelligence) technologies, joint cooperation among multiple parties to jointly explore the value of data, and give birth to new intelligent formats and models based on joint modeling. Federated learning has at least the following characteristics:

[0026] (1) In different application scenarios, we use screening and / or combination of AI algorithms and privacy-preserving computing to establish multiple model aggregation optimization strategies to obtain high-level, high-quality models.

[0027] (2) Based on multiple model aggregation optimization strategies, methods to improve the performance of the federated learning engine are obtained. The performance method can be achieved by solving problems including computing architecture parallelism, information interaction under large-scale cross-domain networks, intelligent perception, and exception handling mechanisms to improve the overall performance of the federated learning engine.

[0028] (3) Obtain the needs of multiple users in each scenario, determine a reasonable assessment of the true contribution of each joint participant through a mutual trust mechanism, and distribute incentives.

[0029] Based on the above methods, we can establish an AI technology ecosystem based on federated learning, give full play to the value of industry data, and promote the implementation of scenarios in vertical fields.

[0030] A modular joint learning service platform and system according to an embodiment of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0031] Figure 1 Schematic diagram of a joint learning architecture of an embodiment of the present disclosure. Figure 1 As shown, the architecture of the joint learning may include a modular joint learning service platform 100, and a participant 102, a participant 103, and a participant 104. The modular joint learning service platform 100 may include a server (central node) 101.

[0032] In the joint learning process, the basic model can be established by the server 101, and the server 101 sends the model to the participants 102, 103 and 104 with which the communication connection is established. The basic model can also be uploaded to the server 101 by any participant after establishment, and the server 101 sends the model to other participants with which the communication connection is established. The server 101 starts the participants 102, 103 and 104 to perform model training using the training data, obtains updated model parameters, and aggregates the model parameters sent by the participants 102, 103 and 104 to obtain global model parameters, and transmits the global model parameters back to the participants 102, 103 and 104. The participants 102, 103 and 104 iterate their respective models according to the received global model parameters until the model finally converges, thereby realizing the training of the model. It should be noted that the number of participants is not limited to the above three, but can be set as needed, and the embodiments of the present disclosure are not limited to this.

[0033] Among them, in addition to providing parameter aggregation, model distribution and other services to each participant, the modular joint learning service platform 100 can also provide interactive front-end services to each participant. For example, when a participant needs to seek other participants to jointly train an application model, he can apply for at least one target joint learning task among the multiple joint learning tasks displayed on the front-end interface of the modular joint learning service platform 100 provided by the present disclosure. During this period, the participant will be "united" with other participants of the same target joint learning task in the same joint learning scenario, so that the participant can find a suitable joint learning partner, without the need for the participant to spend energy and time to find other suitable participants, thereby saving most of the expenses of finding suitable participants.

[0034] Figure 2 It is a flowchart of a modular joint learning method provided by an embodiment of the present disclosure. Figure 2 The modular joint learning method can be Figure 1 The modular joint learning service platform 100 is executed.

[0035] like Figure 2 As shown, the modular joint learning method includes:

[0036] Step S201, providing a training data entry template to the participants according to the joint learning scenario selected by the participants.

[0037] Among them, the joint learning scenario usually refers to a business scenario classified according to the business type. Generally, a business scenario corresponds to solving a type (or one) of business problems. For example, the business scenario of gas use in civil buildings involves the business problem of gas load prediction. For another example, the business scenario of commercial electricity use involves the business problem of electricity load prediction. Therefore, according to the above classification criteria, different businesses can be divided into different business scenarios (joint learning scenarios).

[0038] Of course, the division of joint learning scenarios can also be based on other standards. For example, it can be divided according to the type of enterprise. Specifically, it can be divided into joint learning scenarios for Internet enterprises, joint learning scenarios for financial enterprises, joint learning scenarios for medical service enterprises, etc.

[0039] As a preferred embodiment, users (participants) usually only have raw data, and these raw data generally need to be pre-processed before they can be used as training data for joint training. In order to quickly obtain training data that can be directly used for the joint learning training model, this embodiment provides the participants with a training data entry template corresponding to the selected joint learning scenario (for example, power load forecasting). The training data entry template will have relevant entry requirements for the training data of the joint learning scenario, including data format, value range and other requirements. For example, if a participant wants to predict the annual load in 20xx, he must at least provide a description of the load for the entire previous year.

[0040] As an example, after receiving the data entered by the participant (user) based on the above-mentioned training data entry template, the training data required for model training for the joint learning scenario selected by the participant (for example, electricity load forecasting) is obtained.

[0041] Step S202: Based on the training data fed back by the participants, the training data is associated with the joint learning scenario to obtain an associated logical relationship network.

[0042] As an example, the training data is associated with the joint learning scenario to obtain an associated logical relationship network. Specifically, the training data and the joint learning scenario are stored according to the mapping table in Table 1 below to obtain an associated logical relationship network between the two.

[0043] Table 1. Association mapping table between training data and joint learning scenarios

[0044] Training Data Joint learning scenario Use gas load training data Gas load forecast Power load training data Electricity load forecasting … …

[0045] As another example, the training data can be associated with the joint learning scenario to form a tree diagram with the joint learning scenario as the trunk and the training data as the branches, thereby obtaining a logical relationship network of the two.

[0046] Step S203, using the associated logic relationship network, establish multiple joint learning tasks according to the joint learning scenario.

[0047] As an example, further, multiple joint learning tasks corresponding to each joint learning scenario can be established according to the association mapping table between training data and joint learning scenarios constructed in Table 1. For example, a corresponding relationship table of joint learning scenarios-joint learning tasks-training data as shown in Table 2 below can be established.

[0048] Table 2 Correspondence between joint learning scenarios, joint learning tasks and training data

[0049]

[0050] Step S204: when a participant applies for at least one joint learning task, the at least one joint learning task applied for by the participant is determined as a target joint learning task.

[0051] As an example, assuming that the joint learning scenario selected by the user (participant) is gas load forecasting, the joint learning task selection interface can be displayed on the current interface of the modular joint learning service platform 100. The content displayed on the selection interface includes the name of the joint learning task, the initiator, the task scenario (i.e., the joint learning scenario), the task status, my tasks (i.e., the tasks initiated by the initiator), other tasks, and the like. Assuming that 5 joint learning tasks are currently displayed on the selection interface, namely, 01, 02, 03, 04, and 05, the names, initiators, task scenarios (i.e., joint learning scenarios), task status, and the like of these 5 joint learning tasks can be displayed on the interface accordingly, so that users can understand and choose to apply for the joint tasks they need. When the user clicks to apply for joint learning tasks 01 and 02, joint learning tasks 01 and 02 are determined as target joint learning tasks.

[0052] Step S205, according to the target joint learning task, start training the joint model of the participants corresponding to the joint learning scenario, and when the joint model meets the preset training end condition, feed back the final joint model to the participants.

[0053] The preset training end condition may be that the number of training rounds reaches a preset round threshold, or that the model accuracy reaches a preset accuracy.

[0054] The technical solution provided by the embodiment of the present disclosure provides a training data entry template to the participants according to the joint learning scenario selected by the participants; based on the training data fed back by the participants, the training data is associated with the joint learning scenario to obtain an associated logical relationship network; multiple joint learning tasks are established according to the joint learning scenario by using the associated logical relationship network; when the participants apply for at least one joint learning task, the at least one joint learning task applied by the participants is determined as a target joint learning task; according to the target joint learning task, the training of the participant joint model corresponding to the joint learning scenario is started, and when the joint model meets the preset training end condition, the final joint model is fed back to the participants, so as to provide users with a one-stop joint learning interactive service, including comprehensive services such as helping users to quickly find suitable participants and third parties, which well solves the problem that users need to spend a lot of time, operate inconveniently, and have high costs in finding suitable participants and third parties; and it can be realized from the perspective of protecting data sovereignty and data privacy, combined with the characteristics of the industrial Internet of Things, from the perspective of ecological users, so that users have an excellent experience of participating in joint learning and sharing results, thereby promoting the promotion, application and development of joint learning to a certain extent, and has broad market prospects.

[0055] In some embodiments, before providing the training data entry template to the participants according to the joint learning scenario selected by the participants, the following steps are also included:

[0056] Display N application scenarios and their related introduction contents on the current interface;

[0057] When a selection operation of a participant on any one of N application scenarios is received, the application scenario selected by the participant is determined as a joint learning scenario, where N is a positive integer ≥1.

[0058] As an example, after detecting that the user has logged in to an account registered on the modular joint learning service platform, the current interface can jump directly from the login interface to the page for selecting a circle (selecting a joint learning scenario). On this page, a rectangular frame (or a circular frame, etc., the specific display frame shape and form can be flexibly designed according to actual conditions) can be used to display a joint learning scenario, where each joint learning scenario corresponds to a circle, and the number of circles can be expanded. For example, one of the scenarios, such as gas load forecasting, can be displayed on the current page. In addition, in each circle, the business profile of the scenario can be displayed, with pictures and texts, as well as the currently supported joint learning methods (horizontal / vertical), the algorithms used, the encryption method (if any), the number of active people in the circle, and other information can be displayed to facilitate users to understand and select the appropriate circle and join, that is, select the joint learning scenario.

[0059] Exemplarily, users can choose to join one circle or multiple circles, but if they want to join multiple circles at the same time, they must have training data that matches each circle. For example, if a user wants to join both the gas load forecast and electricity load forecast circles, the user must have both gas load data and electricity load data to join both circles at the same time. After the user chooses to join a circle, the uploaded training data will be associated with the circle (joint learning scenario), that is, the electricity load forecast is associated with the electricity load data, and the gas load forecast is associated with the gas load data.

[0060] In some embodiments, the above step S202 includes:

[0061] Receive training data fed back by the participants, perform data verification on the training data, and obtain verification results;

[0062] If the verification result is qualified, the training data is associated with the joint learning scenario to obtain an associated logical relationship network.

[0063] As an example, data verification is performed on the training data entered by the participants, specifically including verifying the data format, data quality, data volume, data range, etc. of the training data entered by the participants, and obtaining corresponding verification results.

[0064] In the disclosed embodiment, a corresponding training data entry template can be set for each type of joint learning scenario, and when setting the training data entry template, the data verification rules (including data format, data range, data quality, etc.) corresponding to each type of joint learning scenario are set at the same time. Among them, the data format usually adopts CSV file (comma separated value file format). The data range includes the data collection time range, the specific value range of the data point, etc. The data volume refers to the number of data items. For example, the gas load data of a certain enterprise for the whole year of 20XX for 12 months (including the gas load data of each month, which can be the monthly gas load average), that is, there are 12 data items. Data quality can be whether there are data anomalies. For example, gas load data is usually non-negative. If a negative value appears in the data uploaded by the user, then it means that the data uploaded by the user is abnormal, that is, there is a data quality problem.

[0065] As an example, the training data entered by the user (participant) can be verified according to preset data verification rules, specifically to check whether the data format is correct, whether the data range is accurate, whether there are problems with the data quality, whether the data volume is too small, etc., to obtain the verification result.

[0066] As an example, suppose that a certain participant (such as enterprise A) wants to predict the annual electricity load in 20XX. The joint learning scenario selected by enterprise A is electricity load prediction. The preset data verification rule corresponding to the electricity load prediction (joint learning scenario) is to provide at least the electricity load data for the whole year before 20XX (that is, the amount of data is not less than 12 (one per month)), the data format is a CSV file, and the data has no negative values. The user can download the corresponding training data entry template by clicking the "Download Data Template" option displayed on the current "Add Data" interface, and enter the data according to the requirements of the template. For example, the user enters the gas load data for the first two years of 20XX based on the template, a total of 24 data (one for each month), the data format is a CSV file, and there is no negative value in the provided data. Then, according to the above data verification rules, the training data entered by the user is verified, and it can be determined that the verification result of the training data entered by enterprise A is qualified. Then, the training data can be associated with the joint learning scenario according to the above steps to obtain an associated logical relationship network.

[0067] Among them, the above-mentioned "Add Data" interface includes a drop-down menu box of "Select Circle" (that is, select the joint learning scenario), "Download Data Template", "Select Training Data", and the inspection results of the training data uploaded by the user. If the inspection result is unqualified, the specific location of the reason why the specific data is unqualified can be displayed on the interface, so that the user can quickly find the location of the abnormal data and re-upload it or modify it before uploading.

[0068] In some embodiments, the above step S204 includes:

[0069] Obtaining task screening conditions input by the participants, and screening out multiple joint learning tasks related to the joint learning scenario according to the task screening conditions;

[0070] When an application operation of a participant for at least one of the multiple joint learning tasks is detected, the at least one joint learning task applied for by the participant is determined as a target joint learning task.

[0071] As an example, when a user (participant) selects a joint learning scenario (such as power load forecasting) and uploads training data (such as power load data), the interface can jump to the joint learning task selection page, on which the user (participant) can see a list of all joint learning tasks related to the joint learning scenario of power load forecasting, where the task list includes my tasks and other tasks. This page also has several filter columns, including "Task Name", "Task Scenario" (Joint Learning Scenario), "Task Status", "Task Initiator", and "Quickly Join Training Task". Users can choose to enter the corresponding filter conditions in the above filter columns according to their actual situation, and click the "Filter" button to view multiple joint learning tasks that meet the filter conditions they entered, saving users a lot of time searching for joint learning tasks, and providing a good user experience.

[0072] Specifically, you can filter by task name, enter the task name, such as "task_2"; you can also filter by task scenario, the drop-down menu includes all preset supported circles (joint learning scenarios); you can also filter by task status, including training, waiting for training, and completed; you can also filter by task initiator, enter the initiator's name, such as "B", and you can also check the quick join box, that is, you can filter out the tasks that currently meet the conditions and can participate in the joint. There are three preset judgment conditions here: ① Whether the joint learning scenario matches (that is, whether the joint learning scenario selected by the user's input filter conditions is consistent with the joint learning scenario selected when uploading training data); ② Whether the selected joint learning task allows other participants to join automatically (this is related to whether the task initiator checked the "Yes" option when setting the "Whether to allow automatic addition of participants" item when creating the task); ③ Whether the task status is not started (tasks that have already started training are not allowed to join midway). If the task is not started, a "Join joint learning task" button is set after the task. If there is no such button, the user can be prompted that there is no task to directly participate in.

[0073] Then, if the user clicks the "Join Joint Learning Task" button, a task waiting pop-up window will pop up on the interface. The page design of the task waiting pop-up window can include the top progress pipeline segments of "task configuration", "pending training", "training", and "training completed". Under this pipeline segment, information such as the number of participants, participants, queue waiting time, and queue waiting tasks can be displayed.

[0074] In some embodiments, after the above step S204, the method further includes:

[0075] Obtain the task creation content entered by the participants;

[0076] A creation task is generated according to the task creation content, and the creation task is determined as the target joint learning task.

[0077] As an example, in combination with the above example, when a user (participant) does not see a suitable joint learning task among the multiple joint learning tasks related to the joint learning scenario selected by the user on the joint learning task selection page, a joint learning task creation request can be initiated to the modular joint learning service platform 100. The joint learning task creation request includes the name of the participant, the currently selected joint learning scenario, and other contents.

[0078] As an example, when the service platform receives a joint learning task creation request initiated by a participant, and the request includes a joint learning scenario (gas load forecasting) selected by the participant, the interface for creating the task may be displayed to the participant.

[0079] In some embodiments, generating a creation task according to the task creation content and determining the creation task as a target joint learning task specifically includes:

[0080] Determine the joint learning scenario selected by the participants, and automatically configure the preset task parameters corresponding to the joint learning scenario in the task parameter column on the task creation interface;

[0081] When a participant's creation task confirmation operation is detected, a creation task is generated according to the task name and preset task parameters corresponding to the joint learning scenario.

[0082] Exemplarily, the progress pipeline segment including the four links of task configuration - waiting for training - training - training completed can be displayed at the top of the interface for creating a task. When the user selects the "task configuration" item, the task configuration page will be displayed on the interface. On this configuration page, the user only needs to fill in the task name and end condition. The other configuration information platform has configured the corresponding parameters by default according to the joint learning scenario selected by the user, and associated the training data prepared by the user before. If certain parameters need to be modified, the user can modify the parameter items that need to be modified. Generally, the task name can be filled in with up to 10 characters to avoid the title being too long and concise.

[0083] Exemplarily, the task configuration items on the task configuration page may include the task configuration content shown in Table 3 below.

[0084] Table 3 Task configuration content

[0085]

[0086]

[0087] In addition, for hyperparameter configuration, when the user selects the "View Details" option on the task configuration page, the interface will jump to the hyperparameter configuration details page. Different joint learning tasks generally have different hyperparameter configuration contents. Users can further manually debug the range of hyperparameters or add restrictions according to actual conditions.

[0088] As an example, when the user currently selects the joint learning task 02 in the gas load prediction scenario, the corresponding hyperparameter configuration details page can display the system default hyperparameter default configuration details document shown in Table 4 below.

[0089] Table 4 Hyperparameter configuration document

[0090]

[0091]

[0092] For the above-mentioned participant selection, two modes are included, specifically mode 1: automatic matching by the system, and mode 2: autonomous selection of participants. Regarding mode 1, in actual application, a certain number of participants can be automatically matched for the user according to the amount of training data entered by the user (participant). In general, the more training data contributed by the participant, the more other participants the system automatically matches for it than the user with less training data contribution. In actual application, the correspondence between the range of training data volume and the number of matching participants can be pre-set. In actual matching, it is also necessary to further automatically match according to the number of other co-participants in the joint learning scenario currently selected by the participant (user). For example, the pre-set amount of training data is more than 1,000, and the corresponding number of matching participants is at least 5. If at this time, there are only 4 other co-participants in the joint learning scenario selected by the user, all participants in the joint learning scenario are automatically matched for the user. For another example, all participants in the same joint learning scenario are sorted according to the amount of training data contribution, and then the user is automatically matched with other participants whose data contribution is equivalent to that of the user according to the user's sorting order.

[0093] Regarding mode 2, in actual application, according to the joint learning scenario selected by the user (participant), the participants that meet the joint learning scenario can be screened out, allowing the user to manually select the number of participants and participants required. This mode is suitable for users who have higher requirements for model performance.

[0094] In addition, a check box for "Whether to allow automatic addition of participants" can be set under Mode 2. If the user checks "Yes" in the check box, other participants can be allowed to automatically choose to join the task after the task is successfully created. Otherwise, other participants are not allowed to automatically choose to join the task.

[0095] As an example, a "Start Joint Learning" button can be set on the task configuration page. When a user clicks the "Start Joint Learning" button on the task configuration page, the task creation is completed, and the interface jumps to the "To be trained" page. On the "To be trained" page, information such as the name of the initiator and its training data name, the name of the participant and its training data name can be displayed. Since the modular joint learning service platform is a comprehensive service platform, there are many tasks to be processed. Usually, a user can only perform one joint learning task in the same joint learning scenario at a time. If the previous task has not yet been completed, information such as "Waiting in queue" and "Queue time XX" can be displayed on the interface to remind the user of the current task status.

[0096] Furthermore, if the service platform can calculate the waiting time of the previous task, the countdown of the last half minute will be displayed on this interface (at this time, the font of the displayed content can be enlarged or highlighted, etc.). In order to prevent too many participants from being displayed on the same interface, thereby affecting the aesthetics of the interface, 6 to 10 participants can be displayed on the same page, and the remaining participants can be represented by "..." (i.e., ellipsis).

[0097] Exemplarily, on this interface, the joint learning engine icon, the task initiator identifier (team leader) can also be displayed. The team leader can be set as the first one on the left, and the remaining number of participants is 8 / 10. The online status of the participants will be displayed in bright color if online and in gray if offline.

[0098] In some embodiments, the above-mentioned participants include a first participant and a second participant, wherein the first participant may be the initiator of the joint learning task, and the second participant is another participant who participates in the same joint learning task with the same goal in the same joint learning scenario with the first participant.

[0099] The above steps start training the joint model of the participants corresponding to the joint learning scenario according to the target joint learning task, and when the joint model meets the preset training end conditions, the final joint model is fed back to the participants, including:

[0100] Using the training data fed back by the first participant, a participant joint model corresponding to the joint learning scenario is trained to obtain first model parameters;

[0101] Using the training data fed back by the second participant, the participant joint model corresponding to the joint learning scenario is trained to obtain second model parameters;

[0102] Aggregate the first model parameter and the second model parameter to obtain an aggregated parameter;

[0103] Using the aggregated parameters, the model parameters of the joint model of the participants are updated to obtain an updated joint model;

[0104] When the updated joint model meets the preset training end condition, the final joint model is fed back to the first participant and the second participant.

[0105] As an example, assume that participant A (assuming it is the task initiator) and other participants B, C, and D jointly participate in the model training of the same target joint learning task "task_2" in the same joint learning scenario (gas load prediction). First, the modular joint learning service platform 100 can provide an independent training space for each participant. Each participant can upload its training data to its corresponding training space. After determining that each participant has selected a target joint learning task, a basic model is distributed to each participant (i.e., participant A and other participants B, C, and D) of the same target joint learning task in the same joint learning scenario. Then, in the modular joint learning service platform 100, it can be simulated that participant A and other participants B, C, and D use their respective uploaded training data to perform the first round of training on the basic model, and obtain their respective first round model parameters after the training is completed. Specifically, participant A obtains the first model parameters after the first round of training and uploads them to the server 101 of the modular joint learning service platform 100. Other participants B, C, and D obtain the second model parameters b, second model parameters c, and second model parameters d respectively after the first round of training, and upload them to the server 101. At this time, after receiving the first model parameters and the second model parameters b, c, and d of the participant A and the other participants B, C, and D, the server 101 aggregates these parameters (for example, using the aggregation method of FedAvg) to obtain aggregated parameters, and returns the aggregated parameters to the participant A and the other participants B, C, and D respectively, so that the participant A and the other participants B, C, and D can use the aggregated parameters to update the model parameters obtained after the first round of training, and use the model with updated model parameters to perform a second round of training on the training data to obtain second-round model parameters, and then upload the second-round model parameters to the server 101 for aggregation to obtain second-round aggregated parameters, and then return them to each participant. Each participant uses the second-round aggregated parameters to iteratively update the model parameters updated in the previous round, and then performs a third round of training after the update. The above training update steps are repeated until the preset training end condition is reached (for example, the number of training rounds reaches the preset training round number threshold of 100 rounds), and the final joint model is fed back to the terminals of the participant A and the other participants B, C, and D.

[0106] In some embodiments, after starting to train a participant joint model corresponding to a joint learning scenario according to a target joint learning task, the method includes:

[0107] Obtain model training information of all participants who are participating in the same joint learning task with the same target joint learning scenario;

[0108] The joint training process results are generated based on the model training information of all participants and displayed to all participants, wherein the model training information includes the model training log information of all participants.

[0109] As an example, assume that the current participant A (the initiator) and other participants B, C, D, E, and F jointly participate in the same joint learning task (e.g., "task_2") of the same joint learning scenario (e.g., gas load prediction). When it is detected that each of the above participants uses their own training data to train the basic model, the model training log information of each of the above participants is obtained, including the time point of joining the training, the training round, etc. Then, according to the model training information of all participants, the joint training process results are generated and displayed to all participants.

[0110] Among them, the results of the above-mentioned model training process include static display of the model training log information of each participant, and dynamic display of the real-time change graph of the performance characterization curve of the model jointly trained by each participant, so that each participant can better experience the process of joint learning.

[0111] When the training is completed, the user can be prompted with a bubble on the page that "Training is Completed", or a "checked" status can be displayed at the "Training Completed" position of the pipeline progress bar at the top of the page. If a serious abnormality occurs during the training process (for example, only one participant's data is normal, and the data of other participants is abnormal, that is, the model parameters of at least two participants cannot be obtained for aggregation, and subsequent joint learning cannot be performed), a prompt text message "Training Task Terminated" can be displayed on the current interface. If the data of at least two participants is normal, the joint learning task can still continue.

[0112] In one embodiment, the method further includes:

[0113] According to the joint learning scenario selected by the participant, provide the participant with a test data entry template corresponding to the joint learning scenario;

[0114] Receive the test data entered by the participant based on the test data entry template, input the test data into the final joint model, and output the test results.

[0115] As an example, after the joint learning task training is completed, when the user clicks on the "Driving Prediction" option displayed on the current interface, the "Driving Prediction" page is entered. On the "Driving Prediction" page, a selection bar (including a drop-down menu) corresponding to "Select Prediction Scenario" can be set, and the user can select the corresponding joint learning scenario according to actual needs. This joint learning scenario is not limited to the joint learning scenarios that the user has participated in before, but also includes some joint learning scenarios that are open online and available for all users to experience. After the user selects the joint learning scenario he needs, the platform will provide a test data entry template corresponding to the joint learning scenario. The test data entry template provided to the user here is similar to the above-mentioned training data entry template, both of which enable users to quickly prepare training data (or test data) for model training. The test data entry template can also refer to the above-mentioned training data entry template to set the corresponding data format, data range and other upload requirements, and set similar data verification rules to improve the reliability and accuracy of subsequent training / test results.

[0116] As an example, a "Start Prediction" button can be set on the "Prediction and Inference" page. After receiving the user's click on the "Start Prediction" button, the prediction results of the test model (one of the final joint models) selected by the user using the test data uploaded by the user can be displayed on the current interface. In order to better evaluate the performance of the model, test data without real result labels and test data with real result labels can also be uploaded at the same time in the prediction stage. The difference between these two sets of test data is only whether they are labeled with real results, and the rest are the same.

[0117] For example, if a user wants to perform a prediction on a gas load prediction model, he can upload gas load data with and without real result labels for 12 points at the same time. After receiving the user's click on the "Start Prediction" button, the 24 gas load data can be tested according to the prediction model selected by the user, and the corresponding test results can be obtained. If the real data of 12 historical moments are input at the same time, in addition to the load prediction values ​​of the above 12 points, the corresponding loss value / accuracy and other information indicating the performance of the model can also be output, so that the user can intuitively understand the performance of the test model.

[0118] After the forecast is completed, the platform will generate a corresponding forecast report and provide a download function for users to download and view. In the forecast report, the forecast results of the most recent model are usually displayed. In combination with the above example, if real data is uploaded, the forecast report will display a comparative bar chart of each group of data and the load forecast value for 12 months; if no real data is uploaded, only the forecast value will be displayed.

[0119] In some embodiments, the above method further includes:

[0120] The final joint model and the joint learning scenario are stored in the model repository.

[0121] As an example, suppose that participant A (the initiator) and other participants B, C, D, E, and F jointly participate in the same target joint learning task (e.g., "task_2") of the same joint learning scenario (e.g., gas load prediction), and the final model obtained after training is the gas load prediction model 01. Then the gas load prediction model 01 can be classified into the model repository of the gas load prediction (joint learning scenario) for storage.

[0122] When a user needs to find a certain model, he or she can enter the corresponding filtering conditions (for example, model name, model type, algorithm type, etc.) in the input area on the preset "Ecological Model" page. At this time, the platform can automatically match models that meet the filtering conditions entered by the user and display the relevant information of these models on the current interface, so that the user can understand and perform the next step (such as downloading or publishing).

[0123] As an example, the technical solution provided by the embodiment of the present disclosure can also edit and manage the data information and joint learning tasks of the participants.

[0124] Specifically, when a user needs to view or modify personal information, he or she can enter the corresponding filtering conditions (for example, data name, data applicable scenarios, etc.) in the input area of ​​the preset "Personal Center" page. At this time, the personal center module can automatically match the personal information that meets the filtering conditions entered by the user and display the personal information to the user, so that the user can view all the data information uploaded by him or her, including data name, data volume, data usage scenarios, number of times used, upload time, etc.

[0125] All the above optional technical solutions can be arbitrarily combined to form optional embodiments of the present application, which will not be described one by one here.

[0126] The following are embodiments of the device disclosed herein, which can be used to execute the method embodiments disclosed herein. For details not disclosed in the device embodiments disclosed herein, please refer to the method embodiments disclosed herein.

[0127] Figure 3 Schematic diagram of a modular joint learning system provided by an embodiment of the present disclosure. Figure 3 As shown, the modular joint learning system includes:

[0128] The template providing module 301 is configured to provide a training data entry template to the participant according to the joint learning scenario selected by the participant;

[0129] The association module 302 is configured to associate the training data with the joint learning scenario based on the training data fed back by the participants to obtain an associated logical relationship network;

[0130] The task establishment module 303 is configured to establish a plurality of joint learning tasks according to the joint learning scenario by using the associated logic relationship network;

[0131] The target task determination module 304 is configured to determine the at least one joint learning task applied by the participant as a target joint learning task when the participant applies for at least one joint learning task;

[0132] The joint training module 305 is configured to start training the joint model of the participants corresponding to the joint learning scenario according to the target joint learning task, and when the joint model meets the preset training end conditions, feed back the final joint model to the participants.

[0133] The technical solution provided by the embodiment of the present disclosure is as follows: the template providing module 301 is configured to provide a training data entry template to the participant according to the joint learning scenario selected by the participant; the association module 302 is configured to associate the training data with the joint learning scenario based on the training data fed back by the participant to obtain an associated logical relationship network; the task establishment module 303 is configured to use the associated logical relationship network to establish multiple joint learning tasks according to the joint learning scenario; the target task determination module 304 is configured to determine at least one joint learning task applied by the participant as a target joint learning task when the participant applies for at least one joint learning task; the joint training module 305 is configured to start training and joint learning according to the target joint learning task. The joint model of the participants corresponding to the learning scenario can provide users with one-stop joint learning interactive services, including comprehensive services such as helping users quickly find suitable participants and third parties, which effectively solves the problem that users need to spend a lot of time, operate inconveniently and have high costs to find suitable participants and third parties. Moreover, it can take the protection of data sovereignty and data privacy as the starting point, combine the characteristics of industrial Internet of Things, and stand from the perspective of ecological users to allow users to have an excellent experience of participating in joint learning and sharing results, thereby promoting the promotion, application and development of joint learning to a certain extent, and has broad market prospects.

[0134] In some embodiments, the above system further includes:

[0135] A scenario display module is configured to display N application scenarios and their related introduction contents on the current interface;

[0136] The scenario determination module is configured to determine the application scenario selected by the participant as the joint learning scenario when receiving a selection operation of any one of N application scenarios by the participant, where N is a positive integer ≥1.

[0137] In some embodiments, the association module 302 includes:

[0138] A verification unit is configured to receive training data fed back by a participant, perform data verification on the training data, and obtain a verification result;

[0139] The association unit is configured to associate the training data with the joint learning scenario to obtain an associated logical relationship network if the verification result is qualified.

[0140] In some embodiments, the target task determination module 304 includes:

[0141] A task screening unit is configured to obtain task screening conditions input by a participant, and screen out a plurality of joint learning tasks related to the joint learning scenario according to the task screening conditions;

[0142] The task determination unit is configured to determine at least one joint learning task applied for by a participant as a target joint learning task when detecting an application operation of the participant for at least one task among the multiple joint learning tasks.

[0143] In some embodiments, the above system further includes:

[0144] A creation content acquisition module is configured to acquire task creation content entered by a participant;

[0145] The task creation module is configured to generate a creation task according to the task creation content, and determine the creation task as a target joint learning task.

[0146] In some embodiments, the task creation content includes a task name. The task creation module includes:

[0147] A parameter configuration unit is configured to determine the joint learning scenario selected by the participant, and automatically configure preset task parameters corresponding to the joint learning scenario in the task parameter column on the task creation interface;

[0148] The task creation unit is configured to generate a creation task according to the task name and preset task parameters corresponding to the joint learning scenario when a task creation confirmation operation of a participant is detected.

[0149] In some embodiments, the above-mentioned participants include a first participant and a second participant. The above-mentioned joint training module 305 includes:

[0150] A first training unit is configured to train a participant joint model corresponding to the joint learning scenario using the training data fed back by the first participant to obtain a first model parameter;

[0151] A second training unit is configured to train a participant joint model corresponding to the joint learning scenario using the training data fed back by the second participant to obtain a second model parameter;

[0152] a parameter aggregation unit, configured to aggregate the first model parameter and the second model parameter to obtain an aggregated parameter;

[0153] A parameter updating unit, configured to update the model parameters of the joint model of the participants using the aggregated parameters to obtain an updated joint model;

[0154] The model feedback unit is configured to feed back the final joint model to the first participant and the second participant when the updated joint model meets the preset training end condition.

[0155] In some embodiments, the system further comprises:

[0156] A model training information acquisition module is configured to acquire model training information of all participants who participate in the same joint learning task of the same target joint learning scenario with the participant;

[0157] The training process display module is configured to generate a joint training process result based on the model training information of all participants and display it to all participants, wherein the model training information includes the model training log information of all participants.

[0158] In some embodiments, the system further comprises:

[0159] A template providing module is configured to provide the participants with a test data entry template corresponding to the joint learning scenario according to the joint learning scenario selected by the participants;

[0160] The test module is configured to receive the test data entered by the participants based on the test data entry template, input the test data into the final joint model, and output the test results.

[0161] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present disclosure.

[0162] Figure 4 4 is a schematic diagram of an electronic device 400 provided in an embodiment of the present disclosure. Figure 4As shown, the electronic device 400 of this embodiment includes: a processor 401, a memory 402, and a computer program 403 stored in the memory 402 and executable on the processor 401. When the processor 401 executes the computer program 403, the steps in the above-mentioned method embodiments are implemented. Alternatively, when the processor 401 executes the computer program 403, the functions of the modules / units in the above-mentioned device embodiments are implemented.

[0163] Exemplarily, the computer program 403 may be divided into one or more modules / units, which are stored in the memory 402 and executed by the processor 401 to complete the present disclosure. The one or more modules / units may be a series of computer program instruction segments capable of completing specific functions, which are used to describe the execution process of the computer program 403 in the electronic device 400.

[0164] The electronic device 400 may be a desktop computer, a notebook, a PDA, a cloud server, or other electronic device. The electronic device 400 may include, but is not limited to, a processor 401 and a memory 402. Those skilled in the art will appreciate that Figure 4 It is only an example of the electronic device 400 and does not constitute a limitation of the electronic device 400. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the electronic device may also include input and output devices, network access devices, buses, etc.

[0165] The processor 401 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or the processor may be any conventional processor, etc.

[0166] The memory 402 may be an internal storage unit of the electronic device 400, for example, a hard disk or memory of the electronic device 400. The memory 402 may also be an external storage device of the electronic device 400, for example, a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (Flash Card), etc. equipped on the electronic device 400. Further, the memory 402 may also include both an internal storage unit of the electronic device 400 and an external storage device. The memory 402 is used to store computer programs and other programs and data required by the electronic device. The memory 402 may also be used to temporarily store data that has been output or is to be output.

[0167] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0168] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0169] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this disclosure.

[0170] In the embodiments provided in the present disclosure, it should be understood that the disclosed devices / electronic devices and methods can be implemented in other ways. For example, the device / electronic device embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. Multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection of devices or units, which may be electrical, mechanical or other forms.

[0171] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0172] In addition, each functional unit in each embodiment of the present disclosure may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0173] If the integrated module / unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the present disclosure implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium, and the computer program can implement the steps of the above-mentioned various method embodiments when executed by the processor. The computer program may include computer program code, and the computer program code may be in source code form, object code form, executable file or some intermediate form. The computer-readable medium may include: any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium. It should be noted that the content contained in the computer-readable medium can be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electric carrier signals and telecommunication signals.

[0174] The above embodiments are only used to illustrate the technical solutions of the present disclosure, rather than to limit them. Although the present disclosure has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present disclosure, and should all be included in the protection scope of the present disclosure.

Claims

1. A modular joint learning method, characterized in that: include: Providing a training data entry template to the participant according to the joint learning scenario selected by the participant; Based on the training data fed back by the participants, the training data is associated with the joint learning scenario to obtain an associated logical relationship network; Using the associated logical relationship network, a plurality of joint learning tasks are established according to the joint learning scenario; When the participant applies for at least one of the joint learning tasks, determining the at least one joint learning task applied for by the participant as a target joint learning task; According to the target joint learning task, start training the joint model of the participants corresponding to the joint learning scenario, and when the joint model meets the preset training end condition, feed back the final joint model to the participants; The step of associating the training data with the joint learning scenario based on the training data fed back by the participant to obtain an associated logical relationship network includes: receiving the training data fed back by the participant, performing data verification on the training data to obtain a verification result; if the verification result is qualified, associating the training data with the joint learning scenario to obtain an associated logical relationship network; The training data is associated with the joint learning scenario to form a tree diagram with the joint learning scenario as the trunk and the training data as branches, so as to obtain the associated logical relationship network.

2. The method according to claim 1, characterized in that Before providing the training data entry template to the participant according to the joint learning scenario selected by the participant, the method further includes: Displaying multiple application scenarios on the current interface, wherein the application scenarios at least include introduction content; When an operation instruction of the participant on any one of the multiple application scenarios is received, the application scenario selected by the participant is determined as the joint learning scenario.

3. The method according to claim 1, characterized in that When the participant applies for at least one of the joint learning tasks, determining the at least one joint learning task applied for by the participant as a target joint learning task includes: Acquiring a task screening condition input by the participant, and screening out a plurality of joint learning tasks related to the joint learning scenario according to the task screening condition; When an application operation of the participant for at least one of the multiple joint learning tasks is detected, the at least one joint learning task applied for by the participant is determined as a target joint learning task.

4. The method according to claim 1, characterized in that: When the participant applies for at least one of the joint learning tasks, after determining the at least one joint learning task applied for by the participant as the target joint learning task, the method further includes: Obtaining the task creation content entered by the participant; A creation task is generated according to the task creation content, and the creation task is determined as a target joint learning task.

5. The method according to claim 4, characterized in that The task creation content includes a task name; The step of generating a creation task according to the task creation content and determining the creation task as a target joint learning task includes: Determine the joint learning scenario selected by the participant, and automatically configure preset task parameters corresponding to the joint learning scenario in the task parameter column on the task creation interface; When the participant's task creation determination operation is detected, a creation task is generated according to the task name and preset task parameters corresponding to the joint learning scenario.

6. The method according to claim 1, characterized in that The participants include a first participant and a second participant; The step of starting training of a participant joint model corresponding to the joint learning scenario according to the target joint learning task, and feeding back a final joint model to the participant when the joint model meets a preset training end condition, includes: Using the training data fed back by the first participant to train a participant joint model corresponding to the joint learning scenario, to obtain first model parameters; Using the training data fed back by the second participant to train the participant joint model corresponding to the joint learning scenario, to obtain second model parameters; Aggregating the first model parameter and the second model parameter to obtain an aggregated parameter; Using the aggregation parameters, the model parameters of the joint model of the participants are updated to obtain an updated joint model; When the updated joint model meets the preset training end condition, the final joint model is fed back to the first participant and the second participant.

7. The method according to claim 1, characterized in that After starting to train the participant joint model corresponding to the joint learning scenario according to the target joint learning task, the method includes: Obtaining model training information of all participants who jointly participate in the same joint learning task of the same target joint learning scenario with the participant; A joint training process result is generated based on the model training information of all the participants and displayed to all the participants, wherein the model training information includes the model training log information of all the participants.

8. The method according to claim 1, characterized in that The method further comprises: According to the joint learning scenario selected by the participant, providing the participant with a test data entry template corresponding to the joint learning scenario; Receive the test data entered by the participant based on the test data entry template, input the test data into the final joint model, and output the test results.

9. A modular joint learning system, characterized in that: include: A template providing module, configured to provide a training data entry template to a participant according to a joint learning scenario selected by the participant; An association module, configured to associate the training data with the joint learning scenario based on the training data fed back by the participant, so as to obtain an association logic relationship network; A task establishment module is configured to establish a plurality of joint learning tasks according to the joint learning scenario by using the associated logical relationship network; a target task determination module, configured to, when the participant applies for at least one of the joint learning tasks, determine at least one joint learning task applied for by the participant as a target joint learning task; A joint training module is configured to start training a joint model of the participants corresponding to the joint learning scenario according to the target joint learning task, and when the joint model meets a preset training end condition, feed back a final joint model to the participants; The association module is further configured to receive the training data fed back by the participant, perform data verification on the training data, and obtain a verification result; If the verification result is qualified, the training data is associated with the joint learning scenario to obtain an associated logical relationship network; The association module is also configured to associate the training data with the joint learning scenario to form a tree diagram with the joint learning scenario as the trunk and the training data as branches, so as to obtain the association logic relationship network.

Images