Multi-party data collaborative decision-making method and device, electronic equipment and readable storage medium
By integrating learning to train multiple decision-makers and adjusting their weights, the problems of data silos and security in decision-making systems of operators in multiple locations are solved, enabling efficient data-driven collaborative decision-making and resource utilization, and ensuring the accuracy and reliability of decision results.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- CHINA MOBILE INFORMATION TECHNOLOGY CO LTD
- Filing Date
- 2021-08-26
- Publication Date
- 2026-06-05
AI Technical Summary
In existing technologies, the decision-making systems of operators distributed in multiple locations suffer from data silos and data security issues. Existing collaborative decision-making methods cannot effectively solve data barriers and privacy protection, resulting in inefficient resource utilization and high computational costs.
By using an ensemble learning approach, multiple decision-makers are trained and their weight coefficients are adjusted. The weights are then updated using the learning error rate, enabling collaborative decision-making based on data from operators in various regions. The approach involves multi-party decision-making followed by integration, and the exchange of training parameters ensures data security and resource utilization.
It enables the breaking down of data barriers, improving resource utilization, saving computing and processing costs, and ensuring the accuracy and reliability of decision-making results without compromising user privacy.
Smart Images
Figure CN115730207B_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of artificial intelligence and data mining, and in particular to a multi-party data collaborative decision-making method and apparatus, electronic device and readable storage medium. Background Technology
[0002] With the development of economic globalization, operators are no longer limited to a single geographical location, but have formed multiple geographically dispersed decision-making centers. Due to data silos between these regional decision-making centers, many large-scale decision-making activities are no longer suitable for centralized processing, and centralized decision-making also makes it difficult to guarantee the security of data transmission. Furthermore, if each decision-making center makes decisions independently, it will lead to fragmented data. Therefore, achieving collaborative decision-making across different regions has become a focus for operators. Collaborative decision-making methods and devices have been extensively researched for many years, resulting in numerous decision-making patents, which can be broadly categorized as follows.
[0003] The first approach involves multi-party data processing for privacy protection. This primarily studies the data encryption process, encrypting and transmitting data from multiple member devices to achieve the original data encryption scheme, followed by centralized decision-making. However, this approach cannot guarantee absolute data security during encrypted transmission of privacy-sensitive data. Furthermore, it requires significant computational resources for the calculations using the first and second functions, hindering efficient resource utilization.
[0004] The second approach utilizes a task priority ranking algorithm to process orders, achieving efficient use of computing resources while saving computational costs. This method employs a discrete particle swarm optimization algorithm to collaboratively determine the average execution time and equipment utilization for each order based on the order category and task priority. However, this approach requires calculating and estimating the average task execution time and equipment utilization, which introduces some prediction errors and incurs computational resource costs. Furthermore, prioritizing tasks results in a loss of the ability to process tasks of equal priority.
[0005] The third method involves accessing the multi-party data processing system through market interfaces, merchant interfaces, and customer interfaces. This allows for convenient and organized data management and processing via functional management items, avoiding the chaos and complexity of data operations and management. This method is suitable for data integration on O2O platforms, but it does not consider situations where data silos exist among multiple parties. Furthermore, it lacks provisions for handling data security issues in multi-party data processing.
[0006] In conclusion, even with existing technologies, the decision-making systems of operators located in multiple locations still face the challenges of data silos and data security. Summary of the Invention
[0007] This invention provides a multi-party data collaborative decision-making method and apparatus, electronic device and readable storage medium to address the technical deficiencies existing in the prior art.
[0008] This invention provides a multi-party data collaborative decision-making method, comprising:
[0009] Based on the first local public dataset of the first local operator and the mean initial weights, a first decision-maker is trained, and the first weight coefficients of the first decision-maker are updated based on the learning error rate performance of the first decision-maker.
[0010] The second to Hth decision-makers are trained sequentially based on the public datasets of the second to Hth slave operators and the weighted weights of the first to H-1 slave operators corresponding to the second to Hth slave operators. The second to Hth weight coefficients of the second to Hth decision-makers are then obtained based on their learning error rate performance. The weighted weights of the second to Hth slave operators are obtained from the weighted weight coefficients of the first to H-1 decision-makers. H is a positive integer greater than two.
[0011] The first to the Hth decision-makers and their corresponding first to Hth weight coefficients are integrated to obtain a multi-party data collaborative decision-maker.
[0012] According to the multi-party data collaborative decision-making method of the present invention, before training a first decision-maker based on a first local public dataset and mean-initialized weights from a first local operator, and updating the first weight coefficients of the first decision-maker based on the learning error rate performance of the first decision-maker, the method includes:
[0013] Obtain data from the first to the Hth local operators;
[0014] Preprocess the data from the first to the Hth slave operators to obtain the public dataset from the first to the Hth slaves.
[0015] According to the multi-party data collaborative decision-making method of the present invention, the preprocessing of the data from the first to the Hth slave operators to obtain the public dataset from the first to the Hth slave operators includes:
[0016] Based on the Boolean values of the queries sent by the first to Hth slave operators to the central server, the public data set is determined, and the public data set is de-identified to obtain the public dataset of the first to Hth slave operators.
[0017] According to the multi-party data collaborative decision-making method of the present invention, the step of training a first decision-maker based on a first local public dataset of a first local operator and mean-initialized weights, and updating the first weight coefficients of the first decision-maker based on the learning error rate performance of the first decision-maker, includes:
[0018] Set the first source operator data {(X1,y1),(X2,y2),...(X N ,y N )}, where X i ∈X, where X is a vector with multidimensional features, y i ∈Y, where Y is the category label data for the decision; the mean-initialized weights m(1)=(w1,w2,w3,...w N ), where the dimension of vector m(1) is the same as that of X. i N is a positive integer greater than two; a first decision-maker is trained, and the learning error rate performance of the first decision-maker is obtained, wherein the calculation formula for the learning error rate performance of the first decision-maker is:
[0019]
[0020] Wherein, I (first decision-maker (X) i )≠y i ) represents the data where the first decision-maker and the label data are not equal, and the dataset where the first decision-maker was trained incorrectly; i represents the index of the sequence list in the formula; w i X represents i The weight.
[0021] Based on the learning error rate performance of the first decision-maker, the formula for updating the first weight coefficient of the first decision-maker is as follows:
[0022]
[0023] Among them, a m a i All represent the first weighting coefficient.
[0024] According to the multi-party data collaborative decision-making method of the present invention, the weighted weights of the second to the Hth slave operators are obtained based on the first to the H-1th weight coefficients of the first to the H-1th decision-makers, including:
[0025] Based on the learning error rate performance of the first decision-maker, the formula for calculating the weighted weight of the first slave operator is as follows:
[0026]
[0027] w m+1,iThe weights w of the first decision-maker are initialized based on its learning error rate performance. m,i The updated weighted weight of the first slave operator, where i represents the index of the sequence list quantity in the formula.
[0028] According to the multi-party data collaborative decision-making method of the present invention, the step of integrating the first to H decision-makers and their corresponding first to H weight coefficients to obtain a multi-party data collaborative decision-maker includes:
[0029] By integrating the first to H decision-makers and their corresponding weight coefficients using an integration strategy function, a multi-party data collaborative decision-maker is obtained; the integration strategy function is:
[0030] Where i represents the index of the sequence list quantity in the formula.
[0031] According to the multi-party data collaborative decision-making method of the present invention, the step of integrating the first to H decision-makers and their corresponding first to H weight coefficients to obtain a multi-party data collaborative decision-maker includes:
[0032] The decision results obtained by the multi-party data collaborative decision-making unit are output externally via HTTP or file protocols.
[0033] The present invention also provides a multi-party data collaborative decision-making device, comprising:
[0034] The first decision-maker training module is used to train the first decision-maker based on the first local public dataset of the first local operator and the mean initial weights, and to update the first weight coefficients of the first decision-maker based on the learning error rate performance of the first decision-maker.
[0035] The collaborative decision-maker training module is used to train the second to Hth decision-makers sequentially based on the second to Hth slave public datasets of the second to Hth slave operators and the weighted weights of the first to H-1 slave operators corresponding to the second to Hth slave operators, and to obtain the second to Hth weight coefficients of the second to Hth decision-makers based on the learning error rate performance of the second to Hth decision-makers; wherein, the weighted weights of the second to Hth slave operators are obtained based on the first to H-1 weight coefficients of the first to H-1 decision-makers; and H is a positive integer greater than two;
[0036] The collaborative integration module is used to integrate the first to the Hth decision-makers and their corresponding first to Hth weight coefficients to obtain a multi-party data collaborative decision-maker.
[0037] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of any of the multi-party data collaborative decision-making methods described above.
[0038] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the multi-party data collaborative decision-making methods described above.
[0039] This invention achieves weight updates for each subsequent operator by increasing the weights of data with high learning error rates in the front-end decision-makers and then updating the weights of the training data using the learning error rates. It also achieves multi-location data collaboration by serially training and updating the initial training weights of the decision-makers from different operators, followed by a weighted summation to represent the final collaborative decision-making result. Furthermore, this invention utilizes a multi-party decision-making integration approach to achieve rational resource utilization and ensures data security by exchanging only training parameters. It leverages the servers of different operators to perform some decision-making, thereby improving resource utilization and maximizing data-driven decision-making for more accurate and reliable results. This approach improves resource utilization and saves on computational costs. Attached Figure Description
[0040] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.
[0041] Figure 1 This is a flowchart illustrating the multi-party data collaborative decision-making method provided by the present invention;
[0042] Figure 2 This is a schematic diagram of the structure of the multi-party data collaborative decision-making device provided by the present invention;
[0043] Figure 3 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation
[0044] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.
[0045] The following is combined Figure 1 This invention describes a multi-party data collaborative decision-making method, the method comprising:
[0046] S1. Based on the first local public dataset and the mean initial weights of the first local operator, train the first decision-maker, and update the first weight coefficient of the first decision-maker based on the learning error rate performance of the first decision-maker.
[0047] A first decision-maker is trained using the first local public dataset and the mean initial weights; simultaneously, a first weight coefficient is obtained based on the learning error rate performance of the first decision-maker; this first weight coefficient is also used to form the weighted weights of the second local operator used to train the subsequent local decision-maker.
[0048] S2. Based on the public datasets of the second to Hth slave operators and the weighted weights of the first to H-1 slave operators respectively, the second to Hth decision-makers are trained sequentially. The second to Hth weight coefficients of the second to Hth decision-makers are obtained based on their learning error rate performance. The weighted weights of the second to Hth slave operators are obtained based on the first to H-1 weight coefficients of the first to H-1 decision-makers. H is a positive integer greater than two.
[0049] The working mechanism of this collaborative system algorithm is as follows: First, it utilizes public datasets from various regions, where local operators train their own decision-makers. The updated weight data is then exchanged centrally by the decision-maker, ensuring data security while resolving the data silo problem. A second decision-maker is then trained based on a second public dataset with adjusted weights, and this process is repeated until public datasets from multiple regions have been used for training.
[0050] S3. Integrate the first to H decision-makers and their corresponding first to H weight coefficients to obtain a multi-party data collaborative decision-maker.
[0051] By leveraging shared knowledge across regions without compromising user privacy and ensuring data security, the problem of data silos in data-driven decision-making can be solved. Joint decision-making by operators in different regions can effectively avoid conflicts in operating methods and achieve optimal operational decisions, maximizing the use of data to make the results more accurate and reliable.
[0052] The basic idea of this invention is based on ensemble learning. An ensemble method is added between decision models trained by operators in different regions to exchange the error rates and weight coefficients of decision models in different regions. This avoids the exchange of private data between operators, ensures data security, and linearly combines these decision-makers into a more powerful decision-maker to make the final decision.
[0053] This method offers a reliable solution to the shortcomings of existing technologies, taking into account the need to avoid customer privacy leaks, ensure data security, and break down data barriers between different regions of the operator. Because each decision-maker performs calculations in a different location, it achieves efficient utilization of computing resources and savings in computing processing costs.
[0054] This invention achieves weight updates for each downstream operator by increasing the weights of data with high learning error rates in the front-end decision-makers and then using the learning error rates to update the weights of the training data. By serially training and updating the initial training weights of the decision-makers of various operators and then summing them in a weighted manner, the final collaborative decision-making result is represented, thus realizing multi-location data collaboration of computing devices.
[0055] This invention employs a multi-party decision-making and integration approach to achieve rational resource utilization and a method of exchanging only training parameters to ensure data security. It utilizes servers from various operators to compute part of the decision-making process, thereby improving resource utilization. At the same time, it maximizes the use of data to achieve data-driven decision-making, making the results more accurate and reliable. This improves resource utilization and saves on computational processing costs.
[0056] According to the multi-party data collaborative decision-making method of the present invention, before training a first decision-maker based on a first local public dataset and mean-initialized weights from a first local operator, and updating the first weight coefficients of the first decision-maker based on the learning error rate performance of the first decision-maker, the method includes:
[0057] Obtain data from the first to the Hth local operators;
[0058] Preprocess the data from the first to the Hth slave operators to obtain the public dataset from the first to the Hth slaves.
[0059] The data from the first to the Hth local operators includes information such as ID card numbers and mobile phone numbers.
[0060] According to the multi-party data collaborative decision-making method of the present invention, the preprocessing of the data from the first to the Hth slave operators to obtain the public dataset from the first to the Hth slave operators includes:
[0061] Based on the Boolean values of the queries sent by the first to Hth slave operators to the central server, the public data set is determined, and the public data set is de-identified to obtain the public dataset of the first to Hth slave operators.
[0062] The identification and judgment of public datasets are initiated by a central server issuing queries, with local operators' decision centers simply answering "yes" or "no" to determine the public data set. Anonymization primarily involves irreversibly anonymizing sensitive information, such as ID card numbers and mobile phone numbers. This data is accumulated after the central server queries local decision centers.
[0063] According to the multi-party data collaborative decision-making method of the present invention, the step of training a first decision-maker based on a first local public dataset of a first local operator and mean-initialized weights, and updating the first weight coefficients of the first decision-maker based on the learning error rate performance of the first decision-maker, includes:
[0064] Set the first source operator data {(X1,y1),(X2,y2),...(X N ,y N )}, where X i ∈X, where X is a vector with multidimensional features, y i ∈Y, where Y is the category label data for the decision; the mean-initialized weights m(1)=(w1,w2,w3,...w N ), to W N End, here (w1,w2,w3,...w N ) sequence number and {(X1,y1),(X2,y2),...(X N ,y N The X index in )} corresponds to the X index; where the dimension of vector m(1) is the same as that of X. i N is a positive integer greater than two. A first decision-maker is trained, and its learning error rate is obtained. The weighted weights of the training data are updated based on this learning error rate, increasing the weight of training sample points with high learning error rates. This ensures these high-error-rate points receive more attention in the subsequent second decision-maker. In other words, the weights of each training sample point are adjusted based on the learning error rate. A high error rate indicates that the first decision-maker has not learned the information of that sample well; giving it a higher weight strengthens the second decision-maker's focus on that sample point. The formula for calculating the learning error rate of the first decision-maker is:
[0065]
[0066] Wherein, I (first decision-maker (X) i )≠y i ) represents the data where the first decision-maker and the label data are not equal, and the dataset where the first decision-maker was trained incorrectly; i represents the index of the sequence list in the formula; w i X representsi The weights are adjusted. A second decision-maker is trained based on a second public dataset.
[0067] Based on the learning error rate performance of the first decision-maker, the formula for updating the first weight coefficient of the first decision-maker is as follows:
[0068]
[0069] Among them, a m a i All represent the first weighting coefficient.
[0070] According to the multi-party data collaborative decision-making method of the present invention, the weighted weights of the second to the Hth slave operators are obtained based on the first to the H-1th weight coefficients of the first to the H-1th decision-makers, including:
[0071] Based on the learning error rate performance of the first decision-maker, the formula for calculating the weighted weight of the first slave operator is as follows:
[0072]
[0073] w m+1,i The weights w of the first decision-maker are initialized based on its learning error rate performance. m,i The updated weighted weight of the first slave operator, where i represents the index of the sequence list quantity in the formula.
[0074] According to the multi-party data collaborative decision-making method of the present invention, the step of integrating the first to H decision-makers and their corresponding first to H weight coefficients to obtain a multi-party data collaborative decision-maker includes:
[0075] By using an integration strategy function, the first to H decision-makers and their corresponding weight coefficients are integrated to obtain a multi-party data collaborative decision-maker; the integration strategy function is:
[0076] Here, i represents the index of the sequence list quantity in the formula. That is, by serially training and updating the initial training weight coefficients of decision-makers from various operators, and then weighting and summing the decision-makers from each region, the final collaborative decision-making result is represented, resulting in a multi-party data collaborative decision-maker. This ensures data security while solving the data silo problem.
[0077] According to the multi-party data collaborative decision-making method of the present invention, the step of integrating the first to H decision-makers and their corresponding first to H weight coefficients to obtain a multi-party data collaborative decision-maker includes:
[0078] The decision results obtained by the multi-party data collaborative decision-making unit are output externally via HTTP or file protocols.
[0079] Finally, the decision results obtained by the multi-party data collaborative decision-making system are output, thus realizing the entire process of multi-party data collaborative decision-making.
[0080] See Figure 2 The following describes the multi-party data collaborative decision-making device provided by the present invention. The multi-party data collaborative decision-making device described below can be referred to in correspondence with the multi-party data collaborative decision-making method described above. The multi-party data collaborative decision-making device includes:
[0081] The first decision-maker training module 10 is used to train the first decision-maker based on the first local public dataset of the first local operator and the mean initial weights, and to update the first weight coefficients of the first decision-maker based on the learning error rate performance of the first decision-maker.
[0082] A first decision-maker is trained using the first local public dataset and the mean initial weights; simultaneously, a first weight coefficient is obtained based on the learning error rate performance of the first decision-maker; this first weight coefficient is also used to form the weighted weights of the second local operator used to train the subsequent local decision-maker.
[0083] The collaborative decision-maker training module 20 is used to train the second to H decision-makers sequentially based on the second to Hth slave public datasets of the second to Hth slave operators and the weighted weights of the first to H-1 slave operators corresponding to the second to Hth slave operators, and to obtain the second to Hth weight coefficients of the second to Hth decision-makers based on the learning error rate performance of the second to Hth decision-makers; wherein, the weighted weights of the second to Hth slave operators are obtained based on the first to H-1 weight coefficients of the first to H-1 decision-makers; and H is a positive integer greater than two;
[0084] For example, the second slave operator corresponds to the first slave operator before it; the first slave operator corresponds to the second slave operator after it. The working mechanism of this collaborative system algorithm is as follows: First, it utilizes public datasets from various regions, and then each local operator trains its own decision-maker. The updated weight data is exchanged centrally by the decision-maker, thus ensuring data security while solving the data silo problem. The second decision-maker is then trained based on the second slave public dataset with adjusted weights, and this process is repeated until public datasets from multiple regions have been used for training.
[0085] The collaborative integration module 30 is used to integrate the first to the H decision-makers and their corresponding first to H weight coefficients to obtain a multi-party data collaborative decision-maker.
[0086] By leveraging shared knowledge across regions without compromising user privacy and ensuring data security, the problem of data silos in data-driven decision-making can be solved. Joint decision-making by operators in different regions can effectively avoid conflicts in operating methods and achieve optimal operational decisions, maximizing the use of data to make the results more accurate and reliable.
[0087] This invention achieves weight updates for each downstream operator by increasing the weights of data with high learning error rates in the front-end decision-makers and then using the learning error rates to update the weights of the training data. By serially training and updating the initial training weights of the decision-makers of various operators and then summing them in a weighted manner, the final collaborative decision-making result is represented, thus realizing multi-location data collaboration of computing devices.
[0088] According to the multi-party data collaborative decision-making device of the present invention, the device further includes a data preprocessing module for:
[0089] Obtain data from the first to the Hth local operators;
[0090] Preprocess the data from the first to the Hth slave operators to obtain the public dataset from the first to the Hth slaves.
[0091] The data from the first to the Hth local operators includes information such as ID card numbers and mobile phone numbers.
[0092] According to the multi-party data collaborative decision-making apparatus of the present invention, the data preprocessing module is further configured to:
[0093] Based on the Boolean values of the queries sent by the first to Hth slave operators to the central server, the public data set is determined, and the public data set is de-identified to obtain the public dataset of the first to Hth slave operators.
[0094] The identification and judgment of public datasets are initiated by a central server issuing queries. Local operators' decision centers simply answer "yes" or "no" to determine the public data set. If the answer is yes, data from that operator's data center will be retrieved. Anonymization primarily involves irreversibly anonymizing sensitive information, such as ID card numbers and mobile phone numbers. This data is accumulated after the central server queries local decision centers.
[0095] According to the multi-party data collaborative decision-making apparatus of the present invention, the first decision-maker training module 10 is specifically used for:
[0096] Set the first source operator data {(X1,y1),(X2,y2),...(X N ,y N )}, where X i ∈X, where X is a vector with multidimensional features, y i∈Y, where Y is the category label data for the decision; the mean-initialized weights m(1)=(w1,w2,w3,...w N ), where the dimension of vector m(1) is the same as that of X. i N is a positive integer greater than two. A first decision-maker is trained, and its learning error rate is obtained. The weighted weights of the training data are updated based on this learning error rate, increasing the weight of training sample points with high learning error rates. This ensures these high-error-rate points receive more attention in the subsequent second decision-maker. In other words, the weights of each training sample point are adjusted based on the learning error rate. A high error rate indicates that the first decision-maker has not learned the information of that sample well; giving it a higher weight strengthens the second decision-maker's focus on that sample point. The formula for calculating the learning error rate of the first decision-maker is:
[0097]
[0098] Wherein, I (first decision-maker (X) i )≠y i ) represents the data where the first decision-maker and the label data are not equal, and the dataset where the first decision-maker was trained incorrectly; i represents the index of the sequence list in the formula; w i X represents i The weights are adjusted. A second decision-maker is trained based on a second public dataset.
[0099] Based on the learning error rate performance of the first decision-maker, the formula for updating the first weight coefficient of the first decision-maker is as follows:
[0100]
[0101] Among them, a m a i All represent the first weighting coefficient.
[0102] According to the multi-party data collaborative decision-making apparatus of the present invention, the collaborative decision-maker training module 20 is specifically used for:
[0103] Based on the learning error rate performance of the first decision-maker, the formula for calculating the weighted weight of the first slave operator is as follows:
[0104]
[0105] w m+1,i The weights w of the first decision-maker are initialized based on its learning error rate performance. m,iThe updated weighted weights of the first slave operator, where i represents the index of the sequence list quantity in this formula. w m,i It refers to the weights of the sample data used by the first (pre-decision) decision maker. The error rate is calculated based on the decision results of the first (pre-decision) decision maker, and this error rate is used to measure w. m,i Based on the updated weights, we obtain w m+1,i .
[0106] It should be noted that the formula for updating the weighted weights of the i-th slave operator based on the learning error rate performance of the i-th decision-maker is logically consistent with the formula for updating the weighted weights of the first slave operator based on the learning error rate performance of the first decision-maker; only the corresponding parameters need to be changed.
[0107] According to the multi-party data collaborative decision-making device of the present invention, the collaborative integration module 30 is specifically used for:
[0108] By using an integration strategy function, the first to H decision-makers and their corresponding weight coefficients are integrated to obtain a multi-party data collaborative decision-maker; the integration strategy function is:
[0109] Here, i represents the index of the sequence list quantity in the formula. That is, by serially training and updating the initial training weight coefficients of decision-makers from various operators, and then weighting and summing the decision-makers from each region, the final collaborative decision-making result is represented, resulting in a multi-party data collaborative decision-maker. This ensures data security while solving the data silo problem.
[0110] According to the multi-party data collaborative decision-making apparatus of the present invention, the apparatus further includes a decision result output module, the decision result output module being used for:
[0111] The decision results obtained by the multi-party data collaborative decision-making unit are output externally via HTTP or file protocols.
[0112] Finally, the decision results obtained by the multi-party data collaborative decision-making system are output, thus realizing the entire process of multi-party data collaborative decision-making.
[0113] Figure 3 A schematic diagram of the physical structure of an electronic device is provided. This electronic device may include: a processor 310, a communication interface 320, a memory 330, and a communication bus 340. The processor 310, communication interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can invoke logical instructions from the memory 330 to execute a multi-party data collaborative decision-making method, which includes:
[0114] S1. Based on the first local public dataset and the mean initial weights of the first local operator, train the first decision-maker, and update the first weight coefficient of the first decision-maker based on the learning error rate performance of the first decision-maker.
[0115] S2. Based on the public datasets of the second to Hth slave operators and the weighted weights of the first to H-1 slave operators respectively, the second to Hth decision-makers are trained sequentially. The second to Hth weight coefficients of the second to Hth decision-makers are obtained based on their learning error rate performance. The weighted weights of the second to Hth slave operators are obtained based on the first to H-1 weight coefficients of the first to H-1 decision-makers. H is a positive integer greater than two.
[0116] S3. Integrate the first to H decision-makers and their corresponding first to H weight coefficients to obtain a multi-party data collaborative decision-maker.
[0117] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0118] On the other hand, the present invention also provides a computer program product, the computer program product comprising a computer program stored on a non-transitory computer-readable storage medium, the computer program comprising program instructions, wherein when the program instructions are executed by a computer, the computer is able to execute the multi-party data collaborative decision-making method provided by the above methods, the method comprising:
[0119] S1. Based on the first local public dataset and the mean initial weights of the first local operator, train the first decision-maker, and update the first weight coefficient of the first decision-maker based on the learning error rate performance of the first decision-maker.
[0120] S2. Based on the public datasets of the second to Hth slave operators and the weighted weights of the first to H-1 slave operators respectively, the second to Hth decision-makers are trained sequentially. The second to Hth weight coefficients of the second to Hth decision-makers are obtained based on their learning error rate performance. The weighted weights of the second to Hth slave operators are obtained based on the first to H-1 weight coefficients of the first to H-1 decision-makers. H is a positive integer greater than two.
[0121] S3. Integrate the first to H decision-makers and their corresponding first to H weight coefficients to obtain a multi-party data collaborative decision-maker.
[0122] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the aforementioned multi-party data collaborative decision-making methods, the method comprising:
[0123] S1. Based on the first local public dataset and the mean initial weights of the first local operator, train the first decision-maker, and update the first weight coefficient of the first decision-maker based on the learning error rate performance of the first decision-maker.
[0124] S2. Based on the public datasets of the second to Hth slave operators and the weighted weights of the first to H-1 slave operators respectively, the second to Hth decision-makers are trained sequentially. The second to Hth weight coefficients of the second to Hth decision-makers are obtained based on their learning error rate performance. The weighted weights of the second to Hth slave operators are obtained based on the first to H-1 weight coefficients of the first to H-1 decision-makers. H is a positive integer greater than two.
[0125] S3. Integrate the first to H decision-makers and their corresponding first to H weight coefficients to obtain a multi-party data collaborative decision-maker.
[0126] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.
[0127] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.
[0128] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A multi-party data collaborative decision-making method, characterized in that, include: Obtain data from the first to the Hth local operators; Based on the Boolean values of the queries sent by the first to the Hth slave operators to the central server, the public data set is determined, and the public data set is de-identified to obtain the public dataset of the first to the Hth slave operators. Based on the first local public dataset of the first local operator and the mean initial weights, a first decision-maker is trained, and the first weight coefficients of the first decision-maker are updated based on the learning error rate performance of the first decision-maker. The second to Hth decision-makers are trained sequentially based on the public datasets of the second to Hth slave operators and the weighted weights of the first to H-1 slave operators corresponding to the second to Hth slave operators. The second to Hth weight coefficients of the second to Hth decision-makers are then obtained based on their learning error rate performance. The weighted weights of the second to Hth slave operators are obtained based on the first to H-1 weight coefficients of the first to H-1 decision-makers, and these weighted weights are the weights of the training sample points. H is a positive integer greater than two. The first to the Hth decision-makers and their corresponding first to Hth weight coefficients are integrated to obtain a multi-party data collaborative decision-maker.
2. The multi-party data collaborative decision-making method according to claim 1, characterized in that, The process of training a first decision-maker based on the first local public dataset of the first local operator and the mean-initialized weights, and updating the first weight coefficients of the first decision-maker based on the learning error rate performance of the first decision-maker, includes: Set the first source operator data ,in X, where X is a vector with multidimensional features. Y represents the category label data for the decision; weights are initialized with the mean. , where the dimension of vector m(1) is the same as N is a positive integer greater than two; a first decision-maker is trained, and the learning error rate performance of the first decision-maker is obtained, wherein the calculation formula for the learning error rate performance of the first decision-maker is: ； in, This represents the data where the first decision-maker and the label data are not equal, and the dataset where the first decision-maker was trained incorrectly. i represents the index of the sequence list in the formula. Based on the learning error rate performance of the first decision-maker, the formula for updating the first weight coefficient of the first decision-maker is as follows: ； 。 3. The multi-party data collaborative decision-making method according to claim 2, characterized in that, The weighted weights of the second to Hth slave operators are obtained based on the weight coefficients of the first to H-1th decision-makers, including: Based on the learning error rate performance of the first decision-maker, the formula for calculating the weighted weight of the first slave operator is as follows: ； The weights of the first decision-maker are initialized based on its learning error rate. , where i represents the index of the sequence list quantity in the formula.
4. The multi-party data collaborative decision-making method according to claim 3, characterized in that, The process of integrating the first to H decision-makers and their corresponding first to H weight coefficients to obtain a multi-party data collaborative decision-maker includes: By integrating the first to H decision-makers and their corresponding weight coefficients using an integration strategy function, a multi-party data collaborative decision-maker is obtained; the integration strategy function is: Decision Maker (X) = , where i represents the index of the sequence list quantity in the formula.
5. The multi-party data collaborative decision-making method according to claim 1, characterized in that, The process of integrating the first to H decision-makers and their corresponding first to H weight coefficients to obtain a multi-party data collaborative decision-maker includes: The decision results obtained by the multi-party data collaborative decision-making unit are output externally via HTTP or file protocols.
6. A multi-party data collaborative decision-making device, characterized in that, include: Obtain data from the first to the Hth local operators. Based on the Boolean values of the queries sent by the first to the Hth slave operators to the central server, the public data set is determined, and the public data set is de-identified to obtain the public dataset of the first to the Hth slave operators. The first decision-maker training module is used to train the first decision-maker based on the first local public dataset of the first local operator and the mean initial weights, and to update the first weight coefficients of the first decision-maker based on the learning error rate performance of the first decision-maker. The collaborative decision-maker training module is used to train the second to Hth decision-makers sequentially based on the second to Hth public datasets of the second to Hth slave operators and the weighted weights of the first to H-1 slave operators corresponding to the second to Hth slave operators, and to obtain the second to Hth weight coefficients of the second to Hth decision-makers based on the learning error rate performance of the second to Hth decision-makers; wherein, the weighted weights of the second to Hth slave operators are obtained based on the first to H-1 weight coefficients of the first to H-1 decision-makers, and the weighted weights are the weights of the training sample points; H is a positive integer greater than two; The collaborative integration module is used to integrate the first to the Hth decision-makers and their corresponding first to Hth weight coefficients to obtain a multi-party data collaborative decision-maker.
7. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the multi-party data collaborative decision-making method as described in any one of claims 1 to 5.
8. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the multi-party data collaborative decision-making method as described in any one of claims 1 to 5.