Machine learning method and system based on machine quantitative diligence

Abstract

The present invention discloses a machine learning method and system based on machine quantified diligence, wherein the machine learning method based on machine quantified diligence involves: S1. Constructing a set of responsibility lists related to artificial intelligence application roles; S2. Constructing a responsibility trust fulfillment evaluation function based on responsibility time and space; S3. Summarizing and evaluating all constructed responsibility list sets based on the fulfillment evaluation function to obtain an evaluation result; S4. Based on the evaluation result and the weights and rules of different responsibility lists, obtaining the fulfillment result of the responsibility list. The present invention establishes a target equation for quantitative fulfillment to achieve precise learning, avoiding the problems of relying on excessively high dimensions of neural network parameters leading to serious waste of computing power, incomplete marginal knowledge, and difficulty in algorithm interpretability; the fulfillment gap evaluates the learning effect, uniformly optimizes various machine learning algorithms for fulfillment optimization or knowledge gap filling, which can be referred to as fulfillment learning. By customizing the responsibility list set, an engineerable general artificial intelligence is achieved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a machine learning method and system based on machine quantification and diligence. Background Art

[0002] The core of artificial intelligence is machine learning, and the core of machine learning is deep learning. Currently, there is a lack of a complete mathematical foundation, so artificial intelligence is more of a hodgepodge rather than a real discipline.

[0003] Various machine learning methods have been very popular in recent years both in theory and in application. For example, different machine learning methods such as supervised learning, reinforcement learning, self-supervised learning, unsupervised learning, and federated learning have been undergoing algorithm improvement, theoretical expansion, and application testing to varying degrees.

[0004] However, the overall effect is not very ideal, mainly manifested in: complex engineering and industrialization cannot be popularized, the cost-effectiveness is low, and the effect is not as expected. It is currently impossible to replace people with AI. Unmanned factories are more of a replacement for standardized operating procedures, which is difficult to meet the dynamic value realization needs required by market competition.

[0005] The main reason for the above is that machine learning cannot effectively assume corresponding responsibilities, and AI itself is not clear about the management significance of its algorithms.

[0006] For example, the explainability requirement for machine learning algorithms is also a technical requirement that is difficult for ordinary people to understand and operate. In addition, there are loopholes in explainability, making it difficult to conduct a comprehensive and quantitative evaluation.

[0007] Just like human learning, whether the learning outcomes of machine learning have engineering and industrial value depends fundamentally on whether the results of learning can achieve better due diligence. Therefore, truly meaningful machine learning must be a due diligence process, and the machine learning algorithm can be called a quantitative due diligence verification process.

[0008] In response to the above technical problems, the present invention provides a machine learning method and system based on machine quantification and diligence. Summary of the invention

[0009] The purpose of the present invention is to address the deficiencies of the prior art and provide a machine learning method and system based on machine quantification and diligence.

[0010] In order to achieve the above objectives, the present invention adopts the following technical solutions:

[0011] Machine learning methods based on machine quantitative diligence include:

[0012] S1. Construct a set of responsibility checklists related to AI application roles;

[0013] S2. Construct a responsibility trust fulfillment evaluation function based on responsibility time and space;

[0014] S3. Based on the responsibility evaluation function, all responsibility list sets constructed are summarized and evaluated to obtain the evaluation results;

[0015] S4. Based on the evaluation results and the weights and rules of different responsibility lists, the due diligence results of the responsibility list are obtained.

[0016] Furthermore, the responsibility list set in step S1 includes a job responsibility set, a legal responsibility set, and an auxiliary responsibility set.

[0017] Furthermore, the responsibility space-time in step S2 includes a time dimension, a space dimension, an information dimension or a relationship dimension.

[0018] Furthermore, the responsibility performance evaluation function in step S3 includes a risk function and a learning function; the risk function is used to detect parameters or indicator items that are not fulfilled properly; the learning function is used to combine empirical data, various learning algorithms and risk functions to achieve fulfillment of the inadequate parameters.

[0019] Furthermore, the learning function is also used to organize the responsibility fulfillment process data into a standard responsibility fulfillment library.

[0020] Furthermore, the step S4 also includes obtaining a responsibility list of irresponsible persons based on the evaluation results and the weights of different responsibility lists.

[0021] Furthermore, after step S4, the following steps are further included:

[0022] S5. Optimize the fitting of the responsibility performance evaluation function according to the risk factors to obtain the final responsibility performance evaluation function.

[0023] Accordingly, a machine learning system based on machine quantitative diligence is also provided, including:

[0024] The first building block is used to build a set of responsibility checklists related to AI application roles;

[0025] The second building block is used to construct a responsibility trust fulfillment evaluation function based on responsibility time and space;

[0026] The evaluation module is used to summarize and evaluate all responsibility list sets constructed based on the responsibility evaluation function to obtain the evaluation results;

[0027] The due diligence module is used to obtain the due diligence results of the responsibility list based on the evaluation results and the weights and rules of different responsibility lists.

[0028] Furthermore, the responsibility performance evaluation function in the evaluation module includes a risk function and a learning function; the risk function is used to detect parameters or indicator items that are not fulfilled properly; the learning function is used to combine empirical data, various learning algorithms and risk functions to achieve fulfillment of the inadequate parameters.

[0029] Furthermore, it also includes:

[0030] The optimization module is used to optimize the fitting of the responsibility performance evaluation function according to risk factors to obtain the final responsibility performance evaluation function.

[0031] Compared with the prior art, the present invention achieves precise learning through the mechanism of quantitative and responsible target equation, avoiding the serious waste of computing power, poor edge knowledge system, and difficulty in algorithm interpretability caused by the current reliance on the excessively high dimension of neural network parameters; through the mechanism of targeted evaluation of learning effects based on responsible gaps, the current various machine learning algorithms are unified for responsible optimization or knowledge gap filling, which can be referred to as responsible learning; the responsible evaluation mechanism solves the problem of difficult implementation of algorithm interpretability. At the same time, by configuring the responsibility list set of artificial intelligence, an engineering technical framework is also established for general artificial intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 It is a flow chart of a machine learning method based on machine quantification and diligence provided in Example 1;

[0033] Figure 2 It is a schematic diagram of various current and learning methods provided by the first embodiment;

[0034] Figure 3 is a schematic diagram of a quantitative responsibility performance evaluation function provided in Example 1;

[0035] Figure 4 is a schematic diagram of risk, cognition, and responsibility functions provided in Example 1;

[0036] Figure 5 It is a schematic diagram of the formal specification, theoretical basis and evaluation method provided in Example 1. DETAILED DESCRIPTION

[0037] The following describes the embodiments of the present invention by specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict.

[0038] The purpose of the present invention is to provide a machine learning method and system based on machine quantification and diligence in order to address the deficiencies of the prior art.

[0039] Embodiment 1

[0040] This embodiment provides a machine learning method based on machine quantification and diligence, such as Figure 1 As shown, including:

[0041] S1. Construct a set of responsibility checklists related to AI application roles;

[0042] S2. Construct a responsibility trust fulfillment evaluation function based on responsibility time and space;

[0043] S3. Based on the responsibility evaluation function, all responsibility list sets constructed are summarized and evaluated to obtain the evaluation results;

[0044] S4. Based on the evaluation results and the weights and rules of different responsibility lists, the due diligence results of the responsibility list are obtained.

[0045] There are many kinds of learning at present, but there is no systematic one, such as Figure 2 As shown, all learning is about how to fulfill responsibilities, but the responsibility levels, complexity and difficulty are different. Therefore, whether the learning results are reliable or not, and whether machine learning can be engineered, just like human learning, needs to be verified through due diligence. Therefore, high-quality machine learning is definitely a process of fulfilling responsibilities in a responsible manner, and the machine learning algorithm can be called a quantitative due diligence verification process.

[0046] The machine learning quantitative responsibility of this embodiment requires the following processes: 1. Quantified responsibilities; 2. Quantified awareness of responsibilities; 3. Quantified fulfillment of responsibilities; 4. Quantified evaluation of responsibilities; 5. Quantified optimization of responsibilities.

[0047] In step S1, a responsibility list set related to the artificial intelligence AI application role is constructed.

[0048] Constructing a responsibility list set is a form of quantitative responsibility clarification. Quantitative responsibility clarification is constructed through a responsibility list set related to AI roles. Fulfilling responsibilities especially requires fulfilling the legal responsibility set. Therefore, the responsibility list set can include a job responsibility set, a legal responsibility set, and an auxiliary responsibility set.

[0049] In step S2, a responsibility trust fulfillment evaluation function based on responsibility time and space is constructed.

[0050] Constructing a responsibility trust and fulfillment evaluation function is to quantify responsibility. Quantitative responsibility needs to be constructed from the responsibility time and space to achieve learning risk management (responsibility risk); responsibility time and space include time dimension, space dimension, information dimension or relationship dimension. Based on the responsibility time and space, various parameters and parameter compliance rule sets of the four dimensions of time and space of each responsibility list are obtained.

[0051] In step S3, all constructed responsibility list sets are summarized and evaluated based on the responsibility performance evaluation function to obtain an evaluation result.

[0052] The evaluation result is quantitative responsibility fulfillment, which is the management of the learning process (responsibility fulfillment experience, derivative risk judgment, quantitative responsibility fulfillment). The learning algorithm is optimized through the responsibility trust gap in the responsibility space and time.

[0053] In this embodiment, the quantitative responsibility model includes a risk function and a learning function.

[0054] like Figure 3 As shown in the figure, given a set of responsibility lists, the trust in fulfilling responsibilities is decomposed according to the time and space of responsibilities, and after the decomposition, a set of responsibility parameters in various dimensions is generated.

[0055] The risk function is responsible for dynamic detection of possible deficiencies in the performance of responsibilities, and detects parameter content or indicator items that are not in place; in this embodiment, the risk function is also called risk management. It finds the risks or reasons for non-performance, and can find the reasons in the time, space, information or relationship dimensions in the responsibility space-time, and performs boundary calculations on various parameters in the four dimensions of space-time, and lists the parameters that are not met.

[0056] The learning function is selected from the existing knowledge base of fulfillment of responsibilities, or from the fulfillment of responsibilities reasoning records obtained from the neural network model, or extracted from the probability statistics of machine learning (such as the Bayesian equation), structured reasoning in supervised learning, density estimation in unsupervised learning, semantic models of reinforcement learning, etc., and the preferred choice is determined by comparing with the risk function. This process can also be called the gap filling of knowledge points, and the relative amount of calculation can be greatly reduced. In this embodiment, the learning function is also used to organize the fulfillment of responsibilities process data into a standard fulfillment of responsibilities library; in this embodiment, the learning function is also called cognitive management, which is to establish a knowledge base model to quickly optimize the algorithm and conduct in-depth testing on the list of parameters that are not satisfied, including:

[0057] 1) Matching in the responsibility knowledge base (especially the responsibility knowledge base);

[0058] 2) Neural network memory (reasoning), responsible parameter attention retrieval;

[0059] 3) Matching in probability statistics;

[0060] 4) Matching of supervised learning, unsupervised learning, reinforcement learning, etc., is verified to be optimal through the risk function. If it passes, it is stored in the due diligence knowledge base, otherwise it is stored in the neural network memory.

[0061] The responsibility evaluation function evaluates the performance results, and uses the responsibility gap function to calculate the trust gap in time and space. It is a one-by-one calculation and comprehensive evaluation of the entire responsibility list set. Those who are responsible can be included in the existing responsibility knowledge base (part of the responsibility knowledge base), and the list set that needs to be optimized can be obtained; then those that need to be greatly optimized should be listed from the original responsibility list as the list set that cannot be fulfilled, and these will be standardized. In this embodiment, the responsibility evaluation function is also called responsibility management, which is to find the responsibility gap. Responsibility may be a single responsibility list or a set of multiple responsibility lists, so there are many gaps.

[0062] In addition, in the process of fulfilling responsibilities, we will encounter some non-standard content, which also needs to be optimized from the list that cannot be fulfilled.

[0063] like Figure 4 As shown, the risk function dynamically identifies the possibility of liability accidents caused by the current liability risk or other risks, matches different risks as shown in Table 2, and can also be combined as shown in Table 2 to form a new responsibility list decomposition of the prevention level, or the corresponding algorithm solution capability, Risk (id, risk, type) triples, the categories are the categories in Table 2, if there are enough data, the risk is relatively small, and it can also be considered as a supervised learning risk; if the data is not clear whether it is complete enough but the data is enough, the risk is more complex, and it can be considered as a semi-supervised learning or self-supervised learning risk; if there are some potentially useful data, it is a fuzzy risk, which can be a deep learning risk; if there is no useful data that is easy to judge, it is an unknown risk, and it is more necessary to combine various learning mechanisms to achieve risk control. The risk function alg (risk) can be found from Table 2 according to different categories, so that the risks of the responsibility list can be graded and controlled.

[0064] Cognitive management, that is, the learning function includes empirical sub-functions (cognitive functions), such as feedforward networks, graph networks, memory networks, etc. of neural networks. The learning function is the evaluation and optimization of historical data of fulfillment of responsibilities, combined with the responsibility and rights quantification model for optimization. The empirical risk set, the vector set related to fulfillment of responsibilities data, the boundary of the fulfillment optimization combination, and the algorithm are also a combination of Table 2 in terms of the algorithm;

[0065] Gap function (responsibility assessment function), such as learning objectives, learning methods, and learning outcome verification of machine learning; responsibility target functions include single target gap (responsibility trust), multiple target gaps (quantified irresponsibility and rules, quantitative fulfillment and rules), responsibility trust of multiple responsibility lists, dynamic evaluation, and quantitative accountability.

[0066] Table 1 shows the functional principles of learning responsibility goals:

[0067] Table 1 Principles of various functions of learning responsibility goals

[0068]

[0069] Machines perform their duties differently from humans. There are no risks in two of the three aspects of responsibility quality: responsibility awareness and responsible behavior. Therefore, the above is a mechanism of responsibility capacity, that is, a quantified responsibility capacity mechanism.

[0070] The risk calculation method can also adopt the mechanism of Table 2.

[0071] Table 2 Different types of risk functions

[0072]

[0073] Cognitive mechanisms, various learning mechanisms: supervised learning, self-supervised learning, enhanced learning, unsupervised learning, federated learning, etc.

[0074] like Figure 5 Shown are schematic diagrams of the formal specification, theoretical basis, and evaluation methods.

[0075] It can be expressed in the following equation: RespSet is the job responsibility set, RespList is the responsibility list of each item, and each responsibility list is decomposed to sort out the parameters and rules. In the time dimension, it is {tpObj}-{tpRuleList}, in the space dimension, it is {spObj}-{spRuleList}, in the information dimension, it is {ipObj}-{ipRuleList}, and in the relationship dimension, it is {rpObj}-{rpRuleList}.

[0076] The risk function RiskFun generates a parameter list ObjList from various parameters (which may be a list, a matrix, or an object) that match various rules that do not meet the standards. StudyFun uses knowledge supplements from the historical fulfillment library HistKB, the neural network rule library NN_RuleLib, and various machine learning methods to fulfill the standards for each parameter that does not meet the standards, that is, to find the minimum RiskFun value, preferably passing. If it passes, it will be placed in the historical fulfillment library or the neural network rule library.

[0077] DistFun is a gap function that evaluates the job responsibility set and selects the job responsibility set that meets the requirements:

[0078] In this embodiment, the specific formulas for quantifying responsibilities, quantifying fulfillment of responsibilities, and quantifying evaluation of responsibilities are as follows:

[0079] Quantify and clarify responsibilities, expressed as:

[0080]

[0081] Quantify the fulfillment of responsibility, expressed as:

[0082]

[0083] Note: RiskFun calculations can be classified according to Table 2, and ML is a combination of Table 2

[0084] Quantitative assessment, expressed as:

[0085]

[0086] The responsibility list can be configured according to different needs, which forms a general learning framework that can intelligently and effectively decompose the responsibility list in time and space, thus forming a general artificial intelligence algorithm.

[0087] In step S4, based on the evaluation results and the weights and rules of different responsibility lists, the due diligence results of the responsibility list are obtained.

[0088] The trust evaluation of responsibility is quantitative assessment of responsibility, which is learning gap management (quantitative assessment of responsibility). In this way, the mathematical basis of the entire learning can be reconstructed.

[0089] Each responsibility list result is evaluated, and the corresponding degree of diligence for each responsibility list is obtained based on the weights of different responsibility lists; a list of irresponsible people will also be obtained.

[0090] After step S4, the method further includes:

[0091] S5. Optimize and fit the optimized responsibility fulfillment model according to risk factors to obtain the final responsibility fulfillment model.

[0092] Optimizing fitting means quantifying responsibility. Quantifying responsibility is achieved through later algorithm fitting and optimization and upgrading to achieve higher responsibility requirements, realize the growth of machine learning, and then achieve efficient algorithm management.

[0093] This embodiment can be applied to image recognition intelligent algorithm (single responsibility), BERT Chinese semantic recognition (multi-responsibility), selection of responsible persons, adaptation of different learning mechanisms (supervised learning, self-supervised learning, enhanced learning, unsupervised learning, federated learning), etc.

[0094] Compared with the existing technology, the present invention achieves precise learning through a quantified and responsible target equation mechanism, avoiding the current problems of serious waste of computing power, poor edge knowledge system, and difficulty in algorithm interpretability, which are mainly caused by the excessively high dimension of neural network parameters. Through the mechanism of targeted evaluation of learning effects based on the responsible gap, the current various machine learning algorithms are unified for responsible optimization or knowledge gap filling, which can be referred to as responsible learning. The responsible evaluation mechanism solves the problem of difficult implementation of algorithm interpretability. At the same time, by configuring the responsibility list set of artificial intelligence, an engineering technical framework is also established for general artificial intelligence.

[0095] Accordingly, this embodiment also provides a machine learning system based on machine quantification and diligence, including:

[0096] The first building block is used to build a set of responsibility checklists related to AI application roles;

[0097] The second building block is used to construct a responsibility trust fulfillment evaluation function based on responsibility time and space;

[0098] The evaluation module is used to summarize and evaluate all responsibility list sets constructed based on the responsibility evaluation function to obtain the evaluation results;

[0099] The due diligence module is used to obtain the due diligence results of the responsibility list based on the evaluation results and the weights and rules of different responsibility lists.

[0100] Furthermore, the responsibility performance evaluation function in the evaluation module includes a risk function and a learning function; the risk function is used to detect parameters or indicator items that are not fulfilled properly; the learning function is used to combine empirical data, various learning algorithms and risk functions to achieve fulfillment of the inadequate parameters.

[0101] Furthermore, it also includes:

[0102] The optimization module is used to optimize the fitting of the responsibility performance evaluation function according to risk factors to obtain the final responsibility performance evaluation function.

[0103] Embodiment 2

[0104] The machine learning method based on machine quantitative diligence provided in this embodiment is different from that in the first embodiment in that:

[0105] This embodiment takes the scene recognition intelligent algorithm as an example to illustrate:

[0106] Quantify and clarify responsibilities:

[0107] 1) Responsibility list set: It is necessary to identify the scene in the photo and the objects in the scene; and identify the relationship between the objects.

[0108] 2) Expanded legal liability: It is necessary to clarify the responsibilities of users, product quality responsibilities, infringement of personal privacy, etc.

[0109] 3) Auxiliary responsibility: Some difficulties require intervention from other people or things, and the division of responsibilities requires other people or things

[0110] Construct the responsibility space-time (RespListId, Resp4dSpace, mandatory, rate, rt) of each responsibility list, which respectively represent: id, responsibility space-time (refined), whether it is mandatory, weight (if it is mandatory, it can be not set), responsibility trust).

[0111] Quantified responsibility awareness (responsibility setting): mandatory rt must be 1; non-mandatory rt pass rate can be included. Semantic understanding of responsibility list, including semantic understanding of responsibility time and space.

[0112] Quantified responsibility fulfillment (responsibility risk assessment): dynamically calculate rt in the process, improve the responsibility fulfillment degree in the corresponding responsibility list: experience data and function, algorithm evaluation;

[0113] Empirical data and functions will use neural network mechanisms;

[0114] Optimize algorithm (semantic verification);

[0115] Evaluation of various algorithms (learning methods) in the information dimension, such as: analysis of the risk and possibility of being held accountable by users for only being able to identify common objects; legal liability risk; (judgment of responsibility trust, and improvement of corresponding factors);

[0116] Identification of learning risks: Model problems and algorithm problems can lead to failure to fulfill responsibilities.

[0117] Identification results and comparative analysis.

[0118] Quantitative accountability assessment (responsibility outcome assessment): accountability trust assessment, which requires trust assessment for important responsibilities and outcome assessment for unimportant responsibility sets;

[0119] (Responsibility review can be done manually or by comparing with standard algorithms. According to the above time and space of responsibility, for example, whether the problems analyzed are acceptable)

[0120] Quantified accountability (improvements based on risk): Improve algorithms or support exceptions to further improve accountability.

[0121] Embodiment 3

[0122] The machine learning method based on machine quantitative diligence provided in this embodiment is different from that in the first embodiment in that:

[0123] This embodiment is described by taking the selection of a responsible person as an example:

[0124] Quantify and clarify responsibilities:

[0125] 1) Responsibility list set: From the existing list of personnel and related information, quickly select the corresponding responsible person according to the hidden danger situation; for those who cannot be selected, you can create a list of responsible persons with appropriate matching degrees of responsibility in time and space, list the top three, and recommend the first one.

[0126] 2) Extended legal responsibilities: Labor protection clauses in the Labor Law; personal privacy protection in the Personal Information Protection Law; various statutory rights and obligations in the Work Safety Law

[0127] 3) Auxiliary responsibilities: When all employees leave, human handling is required.

[0128] Construct the responsibility space and time (RespListId, Resp4dSpace, mandatory, rate, rt) of each responsibility list above, which respectively represent: id, responsibility space and time of the responsibility list (detailed), whether it is mandatory, weight (if it is mandatory, it can be not set, such as legal terms), responsibility trust).

[0129] Quantified responsibility awareness (responsibility setting): mandatory rt must be 1; non-mandatory rt pass rate can be included. Semantic understanding of responsibility list, including semantic understanding of responsibility time and space.

[0130] Quantify responsibility fulfillment (due diligence risk assessment): dynamically calculate rt in the process, improve the degree of responsibility fulfillment in the corresponding responsibility list: experience data and function, algorithm evaluation.

[0131] The experience data and functions will use the neural network mechanism: for existing regions, organizational structures, job responsibilities, historical data on fulfillment of responsibilities, equipment affiliation, etc., the focus is on using attention BERT to build;

[0132] Optimization algorithm (semantic verification): verify and optimize the responsible person’s current tasks, the responsible person’s current performance, the responsible person’s vacation arrangements, the responsible person’s recent incentives, etc.

[0133] Evaluation of various algorithms (learning methods) in the information dimension, such as whether time, space (territory), and daily information collection are sufficient, etc., and analysis of the risks and possibilities of being held accountable by the selected personnel; legal liability risks; (responsibility trust judgment, and improvement of corresponding factors).

[0134] Identification of learning risks: Model problems and algorithm problems can lead to failure to fulfill responsibilities.

[0135] Identification results, comparative analysis;

[0136] Quantitative accountability assessment (responsibility results assessment): Responsibility trust assessment requires trust assessment for important responsibilities and outcome assessment for unimportant responsibility sets.

[0137] (Responsibility review can be done manually or by comparing with standard algorithms. According to the above time and space of responsibility, for example, whether the problems analyzed are acceptable)

[0138] Quantified accountability (improvements based on risk): Improve algorithms or support exceptions to further improve accountability.

[0139] This embodiment can optimize the above responsibilities by dividing them into different responsibility levels according to the scope of rights, responsibilities, and interests, and gradually perform extended optimization and improvement from the perspectives of spatial responsibility, information responsibility, and relationship responsibility:

[0140] 1) Level of personal responsibility.

[0141] Mainly the expansion of personal interests, the protection of personal rights, and the continuation of personal responsibility

[0142] Risk: not daring to take responsibility, not wanting to take responsibility; greed;

[0143] Cognition: Seeking benefits and avoiding harm

[0144] 2) Departmental responsibility levels.

[0145] Mainly, what kind of responsibilities have you assumed in the department, how well have you achieved your goals, and whether you have made a profit for the department.

[0146] Responsibility coordination is needed, and it is best to achieve responsibility trust between departments

[0147] The competitiveness of the department lies in its ability to iterate quickly and transform into competitiveness such as business innovation. This requires department personnel to continuously improve and break existing cognitions.

[0148] Form responsible positions and responsible departments.

[0149] 3) Unit (enterprise) responsibility level.

[0150] Similar to a department, but involving collaboration between departments, it requires multi-level decomposition. Of course, in essence, from the perspective of responsibilities, rights and interests, it can be completely flat, which is more conducive to cognitive innovation and business innovation.

[0151] Better form responsible units to serve innovation needs.

[0152] 4) Level of social responsibility.

[0153] Needed by a huge customer base.

[0154] 5) Level of moral responsibility.

[0155] 6) Levels of human ethical responsibilities.

[0156] The specific embodiments described herein are merely examples of the spirit of the present invention. Those skilled in the art may make various modifications or additions to the specific embodiments described or replace them in similar ways, but they will not deviate from the spirit of the present invention or exceed the scope defined by the appended claims.

Claims

1. A machine learning method based on machine quantification and conscientiousness, characterized in that: include: S1. Construct a responsibility list set related to the AI ​​application role; the responsibility list set includes a job responsibility set, a legal responsibility set, and an auxiliary responsibility set; S2. Decompose each responsibility list, sort out the parameters and rules, and construct a responsibility trust and responsibility evaluation function based on responsibility time and space; among them, various parameters and parameter compliance rule sets of the four-dimensional space and time of each responsibility list are obtained based on responsibility time and space; S3. Based on the responsibility evaluation function, all responsibility list sets constructed are summarized and evaluated to obtain the evaluation results; The responsibility performance evaluation function includes a risk function and a learning function; the risk function is used to detect parameters or indicator items that are not fulfilled; the learning function is used to combine empirical data and various learning algorithms with the risk function to achieve fulfillment of the parameters that are not fulfilled; the risk function matches each parameter with a list of parameters that do not meet the standards, and the learning function is to supplement each parameter that does not meet the standards with knowledge from the historical responsibility performance library, neural network rule library, and various machine learning methods to meet the standards and find the minimum risk function value; The responsibility evaluation function evaluates the responsibility performance results, and uses the responsibility gap function to calculate the trust gap in time and space. Those who fulfill their responsibilities are included in the existing responsibility knowledge base, and a list set that needs to be optimized is obtained; those who fail to fulfill their responsibilities are listed from the original responsibility list as a list set that cannot be fulfilled; The risk function dynamically identifies the liability accidents caused by the current liability risk or other risks, matches different risks, and forms a new liability list decomposition. It also finds risk functions according to different categories and manages the risks of the liability list in a graded manner. The different risks include simple risks, complex risks, fuzzy risks, and unpredictable risks. If it is a simple risk, it is the risk of supervised learning; if it is a complex risk, it is the risk of semi-supervised learning or self-supervised learning; if it is a fuzzy risk, it is the risk of deep learning; if it is an unpredictable risk, various learning mechanisms are combined to achieve risk management. S4. Based on the evaluation results and the weights and rules of different responsibility lists, the due diligence results of the responsibility list are obtained; the trustworthiness of the responsibility is evaluated, that is, the responsibility is quantitatively evaluated, and the due diligence level and irresponsible list corresponding to each responsibility list are obtained according to the weights of different responsibility lists; S5. Optimize the fitting of the responsibility performance evaluation function according to the risk factors to obtain the final responsibility performance evaluation function.

2. The machine learning method based on machine quantification and diligence according to claim 1, characterized in that: The learning function is also used to organize the fulfillment process data into a standard fulfillment library.

3. A machine learning system based on machine quantification and responsibility, characterized by: include: The first construction module is used to construct a responsibility list set related to the artificial intelligence application role; the responsibility list set includes a job responsibility set, a legal responsibility set, and an auxiliary responsibility set; The second construction module is used to decompose each responsibility list, sort out the parameters and rules, and construct a responsibility trust and responsibility evaluation function based on responsibility time and space; among them, various parameters and parameter compliance rule sets of the four-dimensional space and time of each responsibility list are obtained based on responsibility time and space; The evaluation module is used to summarize and evaluate all responsibility list sets constructed based on the responsibility evaluation function to obtain the evaluation results; The performance evaluation function in the evaluation module includes a risk function and a learning function; the risk function is used to detect parameters or indicator items that are not fully fulfilled; the learning function is used to combine empirical data and various learning algorithms with the risk function to achieve full fulfillment of the parameters that are not in place; the risk function matches each parameter with a list of parameters that do not meet the standards, and the learning function is to supplement each parameter that does not meet the standards with knowledge from the historical performance library, neural network rule library, and various machine learning methods to achieve performance and find the minimum risk function value; The responsibility evaluation function evaluates the responsibility performance results, and uses the responsibility gap function to calculate the trust gap in time and space. Those who fulfill their responsibilities are included in the existing responsibility knowledge base, and a list set that needs to be optimized is obtained; those who fail to fulfill their responsibilities are listed from the original responsibility list as a list set that cannot be fulfilled; The risk function dynamically identifies the liability accidents caused by the current liability risk or other risks, matches different risks, and forms a new liability list decomposition. It also finds risk functions according to different categories and manages the risks of the liability list in a graded manner. The different risks include simple risks, complex risks, fuzzy risks, and unpredictable risks. If it is a simple risk, it is the risk of supervised learning; if it is a complex risk, it is the risk of semi-supervised learning or self-supervised learning; if it is a fuzzy risk, it is the risk of deep learning; if it is an unpredictable risk, various learning mechanisms are combined to achieve risk management. The due diligence module is used to obtain the due diligence results of the responsibility list based on the evaluation results and the weights and rules of different responsibility lists; to conduct a trust evaluation on the responsibility, that is, to quantify the responsibility, and to obtain the due diligence level corresponding to each responsibility list and a list of non-due diligence according to the weights of different responsibility lists; The optimization module is used to optimize the fitting of the responsibility performance evaluation function according to risk factors to obtain the final responsibility performance evaluation function.

Images