Automatic machine learning implementation method, platform and device for scientific research applications
Through an automatic machine learning platform for scientific research applications, the use of relationship maps, intelligent analysis and visual programming, the shortcomings of existing platforms in data sharing, solution retrieval and user customization are solved, and the utilization rate and user experience of scientific research data are improved.
Patent Information
- Application Number
- CN202210023585.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-01-10
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2042-01-10
AI Technical Summary
The existing machine learning platform has shortcomings in data sharing, solution retrieval, user customization and programming support, which makes it difficult for scientific researchers to effectively use machine learning algorithms to solve professional problems.
设计一种面向科研应用的自动机器学习平台,通过获取用户输入的科研数据,生成关系图谱,提供智能分析、个性化数据推荐和可视化编程,结合拖拽式编程与代码编程构建算法流程图,降低使用门槛。
It breaks down data barriers, improves the utilization rate of scientific research data, reduces the communication costs between scientific researchers, and meets the diverse needs of users.
Smart Images

Figure CN114528477B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of machine learning platforms, and in particular to an automatic machine learning implementation method, platform and device for scientific research applications. Background Art
[0002] Machine learning technology has a wide range of application prospects, and related machine learning algorithms are also being developed at a high speed. However, due to the limitation of professional knowledge, many researchers who are not machine learning professionals cannot apply cutting-edge machine learning algorithms to actual scientific research. The main solution to this problem is to design an automatic machine learning platform to lower the threshold for using machine learning algorithms and help researchers focus on problems in their professional fields.
[0003] However, the machine learning platform products on the market currently generally have the following problems. Take the current mainstream machine learning platforms as an example. In terms of data, the platforms Baidu Machine Learning and AI Platform do not support the sharing of basic data. Although AI Platform provides the function of sharing solutions, it does not have a complete solution retrieval function. At the same time, all current machine learning platforms lack the construction of relationship maps between data, making it difficult for users to further use related data. In terms of the scope of machine learning tasks that can be completed, the platforms Baidu Machine Learning and ModelArts currently rely on preset tasks. Users cannot take over the construction of machine learning solutions throughout the entire process, making it difficult to customize solutions according to their own needs. At the same time, the platform also introduces complex development and maintenance work. In terms of programming, the platform Amazon Sagemaker develops algorithms through code based on jupyter notebooks, and users who lack background knowledge in machine learning find it difficult to get started. The platform based on preset tasks lacks support for the programming needs of scientific researchers, and it is difficult for users to fine-tune the code according to existing algorithms. Traditional machine learning platforms can no longer meet the usage needs of existing users, and machine learning platforms are in urgent need of new architecture technologies. Summary of the invention
[0004] In order to solve at least one of the technical problems existing in the prior art to a certain extent, the purpose of the present invention is to provide an automatic machine learning implementation method, platform and device for scientific research applications.
[0005] The technical solution adopted by the present invention is:
[0006] A method for implementing automatic machine learning for scientific research applications, comprising the following steps:
[0007] Obtain scientific research data input by users, and generate solutions based on the scientific research data and the relationship graph constructed by the presets;
[0008] According to the solution, use deep learning methods to intelligently analyze user needs and generate standardized search forms;
[0009] Provide users with personalized data recommendation services based on their operational information on various scientific research data;
[0010] Combine drag-and-drop programming with code programming to build visual operators, and use visual operators to build algorithm flowcharts for experiments to achieve visual programming and display;
[0011] Among them, the relationship map is obtained by constructing the label system and the association system.
[0012] Furthermore, the categories of scientific research data include: data sets, models, algorithms, papers, experiments, and blog documents.
[0013] Furthermore, the solution includes data sets, models, algorithms, papers and experiments that construct relationship maps in scientific research data; the solution is a description of the associations between different types of scientific research data, and points out which specific scientific research tasks are accomplished by subsets of these scientific research data.
[0014] Furthermore, the method further comprises the following steps:
[0015] When the relationship between scientific research data and the data in the platform database is unclear, users use the annotation system to annotate the data; the annotation content includes one or more of the following: application field, application task, data sample format, etc. of the scientific research data;
[0016] The platform database analyzes the label information of scientific research data and associates scientific research data with the same label content;
[0017] When clarifying the association relationship between scientific research data and data in the platform database, users use the association system to directly create association relationships between multiple scientific research data.
[0018] Furthermore, the operation information includes: creation, collection, retrieval and use of scientific research data;
[0019] The personalized data recommendation service provided to users includes:
[0020] After the user determines the search scope, he / she builds and submits a standardized search form;
[0021] Record the user's search behavior, analyze the search form and return the query results;
[0022] By analyzing the user's operation information on scientific research data, we generate user portraits and provide personalized data recommendation services based on the user portraits.
[0023] Furthermore, the use of deep learning methods to intelligently analyze user needs and generate standardized search forms includes:
[0024] Through deep learning methods, data entity information in descriptions or images is extracted, and information such as data sets and model sets that solve or contain such data are analyzed and added to the search form;
[0025] Through semantic understanding, we analyze the tasks that need to be completed in the requirements, and thus build a complete standardized search form;
[0026] Among them, the standardized search form includes the following contents: data set set, model set, algorithm set, paper set, experiment set and solved tasks; the solutions existing in the platform database are searched through this standardized search form.
[0027] Furthermore, the combination of drag-and-drop programming and code programming to construct a visualization operator, and the algorithm flow chart of the experiment is constructed through the visualization operator, including:
[0028] Combine code programming with drag-and-drop programming to create visualization operators, and reuse visualization operators to build algorithm flowcharts for experiments;
[0029] When running an experiment, parse the algorithm flow chart of the experiment and generate executable code;
[0030] After the experiment is over, the experimental results are displayed to users in the form of charts.
[0031] Furthermore, code programming analyzes the parameters contained in the code by parsing the code with a preset specific structure input by the user, and uses the parameter information to construct a specific visualization operator;
[0032] Drag-and-drop programming allows users to select one or more preset visualization operators, and then build new visualization operators by changing operator parameters and operator connection logic.
[0033] Another technical solution adopted by the present invention is:
[0034] An automated machine learning implementation platform for scientific research applications, including:
[0035] The data input module is used to obtain the scientific research data input by the user and generate solutions based on the scientific research data and the preset relationship map;
[0036] Intelligent analysis module, which is used to intelligently analyze user needs based on solutions using deep learning methods and generate standardized search forms;
[0037] User operation module, used to provide users with personalized data recommendation services based on their operation information on various scientific research data;
[0038] Visual display module, which is used to combine drag-and-drop programming with code programming to build visual operators, and to build the algorithm flow chart of the experiment through visual operators to achieve visual programming and display;
[0039] Among them, the relationship map is obtained by constructing the label system and the association system.
[0040] Another technical solution adopted by the present invention is:
[0041] An automatic machine learning implementation device for scientific research applications, comprising:
[0042] at least one processor;
[0043] at least one memory for storing at least one program;
[0044] When the at least one program is executed by the at least one processor, the at least one processor implements the above method.
[0045] The beneficial effects of the present invention are: the present invention breaks the data barriers existing in the existing machine learning platform, allowing scientific researchers in different professional fields to make full use of the existing machine learning solutions to solve problems in their own professional fields, effectively improving the utilization rate of scientific research data and reducing the communication costs between scientific researchers. BRIEF DESCRIPTION OF THE DRAWINGS
[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the embodiments of the present invention or the drawings of related technical solutions in the prior art are introduced below. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative work.
[0047] Figure 1 is a data lake business diagram in an embodiment of the present invention;
[0048] Figure 2 is a flow chart of intelligent demand analysis in an embodiment of the present invention;
[0049] Figure 3 is a data retrieval flow chart in an embodiment of the present invention;
[0050] Figure 4 is a data recommendation flow chart in an embodiment of the present invention;
[0051] Figure 5It is a visual programming and display flow chart in an embodiment of the present invention;
[0052] Figure 6 It is an auxiliary scheme and design flow chart in the embodiment of the present invention;
[0053] Figure 7 It is a platform business process diagram in an embodiment of the present invention. DETAILED DESCRIPTION
[0054] The embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and are not to be construed as limitations of the present invention. For the step numbers in the following embodiments, they are only provided for the convenience of explanation, and the order between the steps is not limited in any way, and the execution order of each step in the embodiment can be adaptively adjusted according to the understanding of those skilled in the art.
[0055] In the description of the present invention, it should be understood that descriptions involving orientations, such as up, down, front, back, left, right, etc., and orientations or positional relationships indicated are based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as a limitation on the present invention.
[0056] In the description of the present invention, "several" means one or more, "more" means more than two, "greater than", "less than", "exceed" etc. are understood as not including the number itself, and "above", "below", "within" etc. are understood as including the number itself. If there is a description of "first" or "second", it is only used for the purpose of distinguishing the technical features, and cannot be understood as indicating or implying the relative importance or implicitly indicating the number of the indicated technical features or implicitly indicating the order of the indicated technical features.
[0057] In the description of the present invention, unless otherwise clearly defined, terms such as setting, installing, connecting, etc. should be understood in a broad sense, and technicians in the relevant technical field can reasonably determine the specific meanings of the above terms in the present invention based on the specific content of the technical solution.
[0058] The present invention is an automatic machine learning platform for scientific research applications, in which the main users are divided into non-professional users without background knowledge of machine learning and researchers with certain background knowledge, such as Figure 7 In order to meet the different needs of the two platforms, the platform has Figures 2 to 6The four major functional modules shown are efficient data retrieval and recommendation, intelligent demand analysis, visual programming and display, and auxiliary solution design. The data storage in the platform gives priority to the data lake implementation plan, the server uses Nginx as a reverse proxy, and users call functions through the RESTFUL interface.
[0059] The data lake is the preferred implementation scheme. Its advantages are that it can store massive amounts of data of various types and support querying and processing of data. After entering the lake, the data retains its original format and content, waiting to be further processed and used. Figure 1 As shown, users can customize the processing of specific data by uploading a processing method for certain data, and other users can effectively avoid the occurrence of data swamps by simply retrieving and reusing the processing method.
[0060] The specific implementation plans of the four major functional modules are as follows:
[0061] S1. In the data retrieval and recommendation module, data represents scientific research data, and its types include data sets, models, algorithms, papers, experiments, blog documents, and solutions. In order to achieve efficient data retrieval, first, the platform constructs a relationship map between scientific research data to reveal the association between data. Users can construct a relationship map through the annotation system and the association system. The steps are as follows:
[0062] Use the annotation system. When the user is not sure which data in the platform should be associated with the scientific research data, the user uses the annotation system to annotate the data. The annotation content includes one or more of the application field, application task, data sample format, etc. of the scientific research data. The platform generates a label entity for each annotation content without duplication, and associates the scientific research data with the label entity. The label entity and the association relationship are stored in the data lake. Afterwards, the platform analyzes the label information of the scientific research data, and the scientific research data with the same label content will be associated with the same label entity, realizing the indirect association of the scientific research data.
[0063] Use the association system. When the user has a clear and specific association goal, the user uses the association system to create direct associations between multiple scientific research data. By combining the annotation system with the association system, the user can flexibly associate the scientific research data with known data and unknown data. The relationship map can be used to efficiently use the association relationship for data retrieval and recommendation, such as Figure 3 shown.
[0064] The tag system and the association system are two methods used in this embodiment to construct a data relationship map. The data of the existing machine learning platform lacks association, and the data can only be retrieved by keyword matching. Due to the large amount of data and complex data types of the machine learning platform, it is difficult for users to retrieve the required data through simple keyword matching, resulting in low platform data utilization. To solve this problem, this embodiment introduces these two systems to construct a data relationship map to achieve intelligent retrieval and recommendation of data.
[0065] The tag system is a more traditional data association method. When creating data, users add specific tags to each data, such as "NLP model", "high-performance model", etc. In this way, data with the same type of tags indicate that they have certain similarities. Users can discover more similar data through tags, and the system can also recommend similar data to users. In the association system, users explicitly point out that the two data are related. The difference from the tag system is that the tag system points out that the data belongs to a certain tag. There is a relationship between the data under the same tag, but it cannot guarantee that there is a strong relationship between them. The main application scenario of the association system is that this data uses or contains another data, or is in a relationship of being used or contained. At this time, there is a strong relationship connection between the two data. These two systems have similar products in other products, such as e-commerce platforms, but they are still in a blank stage on the machine learning platform. We combine the two systems with the machine learning platform to solve the problem that data is difficult to be discovered and the reuse rate is low.
[0066] In order to realize personalized data recommendation services, such as Figure 4 As shown in the figure, the platform records the user's operation records on each scientific research data, including the creation, collection, retrieval and use of data. According to the user's operation records, the user portrait is analyzed using a machine learning algorithm. At the same time, for each scientific research data, its label information and description information are also analyzed by machine learning methods to obtain a data portrait. The personalized data recommendation function is realized by using a recommendation algorithm to analyze the user portrait and data portrait.
[0067] For example, based on the user's operation information on various scientific research data, personalized data recommendation services are provided to users. The operation information here refers to the user's creation, collection, retrieval and use of data. Every time a user operates on the data, it will be automatically recorded by the system. Then the system uses a traditional recommendation algorithm based on deep learning. First, the content of each data itself is analyzed for its association with other data to generate a data portrait. The data portrait is an abstract representation of the data, which is a combination of the data's own information and related information; then, by analyzing what operations the user has performed on the existing data, the corresponding user portrait is generated through the recommendation model algorithm. The user portrait indicates which types of data the user has a preference for, and further predicts which other data the user will be interested in and further recommends. This part is the final output of the system. Since each user has different operation information on the data, the recommended data obtained by each user will be different, realizing personalized recommendation services. These contents are the relatively lacking parts of the existing machine learning platforms, and are also the reason why the existing machine learning platforms have a narrow audience.
[0068] S2. In the intelligent demand analysis module, natural language processing and object recognition technologies in deep learning methods are used. For data-related images input by users, object recognition technology in image processing is used to analyze the data entities in the images by extracting, understanding and integrating features. For the requirements described in natural language in user input, natural language processing technology is used to segment words and label each segmentation, and the data entities in the description are extracted by integrating the label information. The system compares the extracted data entity information with each data portrait to analyze the data set set, model set and other information that contains or solves this type of data entity. At the same time, the semantic understanding technology in natural language processing is used to analyze the tasks that need to be completed in the requirements.
[0069] like Figure 2 As shown in the figure, by integrating the information of each part, the system automatically constructs a standardized search form. The form contains the dataset set, model set, algorithm set, paper set, experiment set and solved task information. Through this standardized search form, the solutions existing in the platform can be efficiently searched.
[0070] The main purpose of introducing deep learning technology to analyze user needs is to allow users to efficiently retrieve existing data while lowering the user's usage threshold. In the process, the user first enters a combination of text and images. The trained object detection model can detect what types of objects appear in the text and images entered by the user, query the database to find out which data sets contain these types of objects, and use natural language processing technology to determine what specific tasks the user wants to achieve, such as Figure 2 Shown is a classification task.
[0071] Finally, by combining the data set with the specific task category, the relationship graph is used to mine suitable models and other information, which are then added to the search form to complete the construction of the overall structured form.
[0072] The solution is just a subset of the existing data and a description of the subset. After the user selects data such as experiments and data sets, the platform will first automatically add other data that is strongly associated with these data, such as used data or data with dependencies. Secondly, the relationship graph of the existing data is analyzed to recommend data that is weakly associated with the existing data to the user, allowing the user to add it selectively. After the user has selected all the data and provided the corresponding description, the platform generates a solution entity and establishes a usage relationship between the entity and the included data through the association system. At the same time, the labels existing in the parsed data are added to the label set of the entity. The user fine-tunes the label set to determine the final label set, and the label system is used to establish a weak association between the entity and other data with the same label.
[0073] In some optional embodiments, a specific example of generating a solution is that experiments A, B, and C jointly solve a problem in the field of computer vision. Since there is a relationship map between the data, when creating the solution, we can just choose to add experiments A, B, and C. The system automatically adds data sets, algorithms, and papers that are strongly related to these experiments to the solution, avoiding errors and omissions caused by manual addition by users, and recommends other weakly related data to users for them to selectively add. Therefore, the use of relationship maps can further help generate solutions.
[0074] In some optional embodiments, not only solutions are generated, but also corresponding data sets, models and other information are selected when generating experiments. The platform records the user's previous preferences for data usage, that is, statistics the distribution of label data used by users, calculates which data users are more likely to use, and then recommends this part of data to users for selection. At the same time, through the relationship map, the relationship between user preference data and other data is analyzed, and new data that users may be willing to use is determined and recommended to users. This avoids users having to repeatedly search for usage data every time they create new data, and is also conducive to allowing users to access and use new data.
[0075] In the S3. Visual Programming and Display module, programming is divided into code programming and drag-and-drop programming. Users create visual operators through drag-and-drop or code programming, and reuse operators to build the algorithm flowchart of the experiment. When running the experiment, the platform parses the algorithm flowchart of the experiment and generates executable code. After the experiment is over, the experimental results are displayed to the user in combination with charts. The specific implementation methods of the two programming methods are as follows:
[0076] Operators are created through code programming. Users input codes with specific structures, and the platform uses regular expressions to parse the initialization parameters, input parameters, and output parameters contained in the code and generate basic visual operators. The parameters include the name of the parameter, the default value of the parameter, and the type of the parameter. Type matching checks will be performed in subsequent operator connections.
[0077] Operators are created through drag-and-drop programming. Users reuse existing visualization operators by dragging and dropping. Operators are connected with certain logic and initialized by providing corresponding initialization parameters. Unused input interfaces and output interfaces are visualized as input parameters and output parameters of new operators. Initialization parameters provided when creating operators are used as default initialization parameters of new operators.
[0078] The specific implementation of the construction and translation of the algorithm flowchart is as follows. Figure 5 As shown, the user drags existing operators to connect and fills in the necessary initialization parameters in the operator to build the algorithm flowchart. The platform performs a validity check on the algorithm flowchart, including whether the input and output types of the operator match, whether the necessary initialization parameters are correctly initialized, and whether the necessary input interface is connected to the output of a certain operator. After the flowchart passes the validity check, recursive translation is performed. The platform determines the running order of the operators through the topological sorting of the operators and creates intermediate variables to store the output of each operator. If the operator currently being translated is created by code programming, the code entered by the user when creating the operator is simply used. If the current operator is created by drag-and-drop programming, the operator is expanded as an algorithm flow subgraph for recursive translation.
[0079] S4. The auxiliary scheme design module mainly focuses on the design of the algorithm flow chart in the auxiliary scheme. Figure 6 As shown in the figure, users can quickly understand the implementation process and logic of other solutions by viewing the algorithm flowcharts of other solutions, and modify the connection logic and configuration parameters of the operators according to their own needs to achieve efficient customization of the algorithm flowchart.
[0080] Among them, data retrieval and recommendation, intelligent demand analysis, visual programming and display, and auxiliary solution design have the following correlation: the reuse of operators in visual programming and display and the search for solutions in auxiliary solution design rely on the data retrieval and recommendation function; secondly, auxiliary solution design realizes solution customization by modifying the visualized algorithm flow chart in the solution, which relies on the visual programming and display function; thirdly, in the data retrieval module, users can use the intelligent demand analysis function to construct standardized retrieval forms based on the requirements described in natural language.
[0081] In the above description of this specification, the description with reference to the terms "one embodiment / example", "another embodiment / example" or "certain embodiments / examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0082] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.
[0083] The above is a specific description of the preferred implementation of the present invention, but the present invention is not limited to the above embodiments. Those skilled in the art may make various equivalent modifications or substitutions without violating the spirit of the present invention. These equivalent modifications or substitutions are all included in the scope defined by the claims of this application.
Claims
1. A method for implementing automatic machine learning for scientific research applications, characterized in that: The following steps are involved: Obtain scientific research data input by users, and generate solutions based on the scientific research data and the relationship graph constructed by the presets; According to the solution, use deep learning methods to intelligently analyze user needs and generate standardized search forms; Provide users with personalized data recommendation services based on their operational information on various scientific research data; Combine drag-and-drop programming with code programming to build visual operators, and use visual operators to build algorithm flowcharts for experiments to achieve visual programming and display; Among them, the relationship map is obtained by constructing the tag system and the association system; The following steps are also included: When the relationship between scientific research data and the data in the platform database is unclear, users use the annotation system to annotate the data; the annotation content includes: application field, application task, data sample format, or more; The platform database analyzes the label information of scientific research data and associates scientific research data with the same label content; When clarifying the association relationship between scientific research data and data in the platform database, users use the association system to directly create association relationships between multiple scientific research data; The method of using deep learning to intelligently analyze user needs and generate a standardized search form includes: Through deep learning methods, data entity information in descriptions or images is extracted, and data sets and model sets containing data entity information are analyzed and added to the search form; Through semantic understanding, we analyze the tasks that need to be completed in the requirements, thus building a complete standardized search form; Among them, the standardized search form includes the following contents: data set set, model set, algorithm set, paper set, experiment set and solved tasks; the solutions existing in the platform database are searched through this standardized search form.
2. The method for implementing automatic machine learning for scientific research applications according to claim 1, characterized in that: The categories of scientific research data include: data sets, models, algorithms, papers, experiments, and blog documents.
3. The method for implementing automatic machine learning for scientific research applications according to claim 1, characterized in that: The solution includes data sets, models, algorithms, papers and experiments that construct relationship graphs in scientific research data; the solution is a description of the associations between different types of scientific research data, and points out which specific scientific research tasks are accomplished by subsets of these scientific research data.
4. The method for implementing automatic machine learning for scientific research applications according to claim 1, characterized in that: The operation information includes: creation, collection, retrieval and use of scientific research data; The personalized data recommendation service provided to users includes: After the user determines the search scope, he / she builds and submits a standardized search form; Record the user's search behavior, analyze the search form and return the query results; By analyzing the user's operation information on scientific research data, we generate user portraits and provide personalized data recommendation services based on the user portraits.
5. The method for implementing automatic machine learning for scientific research applications according to claim 1, characterized in that: The combination of drag-and-drop programming and code programming to construct a visualization operator, and the algorithm flow chart of the experiment is constructed through the visualization operator, including: Combine code programming with drag-and-drop programming to create visualization operators, and reuse visualization operators to build algorithm flowcharts for experiments; When running an experiment, parse the algorithm flow chart of the experiment and generate executable code; After the experiment is over, the experimental results are displayed to users in the form of charts.
6. The method for implementing automatic machine learning for scientific research applications according to claim 5, characterized in that: Code programming parses the code with a preset specific structure input by the user, analyzes the parameters contained in the code, and uses the parameter information to build a specific visualization operator; Drag-and-drop programming allows users to select one or more preset visualization operators, and then build new visualization operators by changing operator parameters and operator connection logic.
7. An automatic machine learning implementation platform for scientific research applications, applied to an automatic machine learning implementation method for scientific research applications as described in any one of claims 1 to 6, characterized in that: include: The data input module is used to obtain the scientific research data input by the user and generate solutions based on the scientific research data and the preset relationship map; Intelligent analysis module, which is used to intelligently analyze user needs based on solutions using deep learning methods and generate standardized search forms; User operation module, used to provide users with personalized data recommendation services based on their operation information on various scientific research data; Visual display module, which is used to combine drag-and-drop programming with code programming to build visual operators, and to build the algorithm flow chart of the experiment through visual operators to achieve visual programming and display; Among them, the relationship map is obtained by constructing the label system and the association system.
8. An automatic machine learning implementation device for scientific research applications, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements the method according to any one of claims 1 to 6.
