Calibration Parameter Calculation Method and Device, User Classification Method, Medium, and Equipment

Abstract

The present disclosure relates to a calibration parameter calculation method and apparatus, a user classification method, a medium, and a device, and relates to the technical field of machine learning. The method includes: receiving an encrypted user label and a user personal identification code of a historical user sent by a second data holder, and obtaining a first user feature of the historical user according to the user personal identification code; calculating a prediction score of the historical user according to a federated model and the first user feature, and determining a bin to which the historical user belongs according to the prediction score; performing a summation operation on the encrypted user labels assigned to the bin to obtain a first summation result, and sending the first summation result to the second data holder; receiving an actual probability sent by the second data holder after decrypting the first summation result, and calculating a calibration parameter of the federated model according to the actual probability. The present disclosure realizes collaborative calibration.

Description

Background Art

[0002] Model calibration is usually an important step before model deployment. Especially in business scenarios where conversion rates and default rates are concerned, the model not only needs to perform a simple binary classification function, but the values between 0 and 1 output by the model often also need to correspond to probabilities with business attributes. Model calibration can more precisely match the output values of the model to these probabilities with specific meanings for subsequent applications.

[0003] Currently, model calibration is only applied after single - party model training, and there is no federated version. It cannot meet the model calibration requirements across data silos, and cannot enable the model - owning party and the label - owning party to calculate the calibration parameters of the federated model without exposing their respective data.

[0004] Therefore, there is a need to provide a new method and device for calculating calibration parameters of a federated model.

[0005] It should be noted that the information disclosed in the background art above is only used to enhance the understanding of the background of the present disclosure. Therefore, it may include information that does not constitute the prior art known to those of ordinary skill in the art. Summary of the Invention

[0006] The purpose of the present disclosure is to provide a method for calculating calibration parameters of a federated model, a method for classifying users based on a federated model, a federated model calibration device, a computer - readable storage medium, and an electronic device, so as to at least to some extent overcome the problem of being unable to calculate the calibration parameters of the federated model due to the limitations and defects of related technologies.

[0007] According to one aspect of the present disclosure, there is provided a method for calculating calibration parameters of a federated model, configured in a first data holder that provides first user features in federated learning. The method for calculating calibration parameters of the federated model includes:

[0008] Receiving the encrypted user labels and user personal identification codes of historical users sent by a second data holder, and obtaining the first user features of the historical users according to the user personal identification codes;

[0009] Calculating the prediction scores of the historical users according to the federated model and the first user features, and determining the bins to which the historical users belong according to the prediction scores;

[0010] Performing a summation operation on the encrypted user labels assigned to the bins to obtain a first summation result, and sending the first summation result to the second data holder;

[0011] Receiving the actual probabilities sent by the second data holder after decrypting the first summation result, and calculating the calibration parameters of the federated model according to the actual probabilities.

[0012] In an exemplary embodiment of the present disclosure, the method for calculating the calibration parameters of the federated model further includes:

[0013] Receiving a binning request sent by a second data holder, and initializing binning thresholds according to the binning rules included in the binning request to obtain multiple bins;

[0014] Wherein, the starting endpoint of the binning threshold is 0 and the ending endpoint is 1.

[0015] In an exemplary embodiment of the present disclosure, calculating the calibration parameters of the federated model according to the actual probability includes:

[0016] Sorting the actual probabilities according to the encoding of the bin to which the actual probabilities belong, and determining whether the sorted actual probabilities increase monotonically;

[0017] If so, using the sorted actual probabilities as the calibration parameters of the federated model;

[0018] If not, performing monotonicity calibration on the sorted actual probabilities based on a preset isotonic regression algorithm to obtain target probabilities, and using the target probabilities as the calibration parameters of the federated model.

[0019] In an exemplary embodiment of the present disclosure, performing monotonicity calibration on the sorted actual probabilities based on a preset isotonic regression algorithm to obtain target probabilities includes:

[0020] Extracting the non-monotonically increasing actual probabilities from the sorted actual probabilities, and when it is determined that the positions of the non-monotonically increasing actual probabilities in the sorted actual probabilities are adjacent positions, merging the bins corresponding to the non-monotonically increasing actual probabilities;

[0021] Calculating the bin probabilities of the merged bins, and updating the sorted actual probabilities according to the bin probabilities of the merged bins to obtain the target probabilities.

[0022] In an exemplary embodiment of the present disclosure, calculating the bin probabilities of the merged bins includes:

[0023] Calculating the bin probabilities of the merged bins based on the non-monotonically increasing actual probabilities and the bin boundary values of the bins corresponding to the non-monotonically increasing actual probabilities.

[0024] According to one aspect of the present disclosure, there is provided a user classification method based on a federated model, configured in a first data holder that provides first user features in federated learning. The user classification method based on the federated model includes:

[0025] Receive the user personal identification code of the user to be classified sent by the second data holder, and obtain the second user feature of the user to be classified according to the user personal identification code;

[0026] Input the second user feature into the federated model to obtain the initial prediction probability of the user to be classified, and determine the bin to which the user to be classified belongs according to the initial prediction probability;

[0027] Determine the target prediction probability of the user to be classified according to the initial prediction probability and the calibration parameter of the bin to which the user to be classified belongs; wherein, the calibration parameter is calculated by the calibration parameter calculation method of the federated model described in any one of the foregoing;

[0028] Send the target prediction probability to the second data holder, so that the second data holder determines the user category to which the user to be classified belongs according to the target prediction probability.

[0029] In an exemplary embodiment of the present disclosure, determining the target prediction probability of the user to be classified according to the initial prediction probability and the calibration parameter of the bin to which the user to be classified belongs includes:

[0030] Calculate a calibration coefficient according to the initial prediction probability and the bin boundary value of the bin to which the user to be classified belongs;

[0031] Calculate the target prediction probability of the user to be classified according to the calibration coefficient and the calibration parameter of the bin to which the user to be classified belongs.

[0032] According to one aspect of the present disclosure, there is provided a calibration parameter calculation device for a federated model, configured in a first data holder that provides a first user feature in federated learning. The calibration device for the federated model includes:

[0033] A first user feature acquisition module, configured to receive the encrypted user label and user personal identification code of a historical user sent by a second data holder, and obtain the first user feature of the historical user according to the user personal identification code;

[0034] A first bin determination module, configured to calculate a prediction score of the historical user according to the federated model and the first user feature, and determine the bin to which the historical user belongs according to the prediction score;

[0035] A first calculation module, configured to perform a summation operation on the encrypted user labels assigned to the bin to obtain a first summation result, and send the first summation result to the second data holder;

[0036] A calibration parameter calculation module, configured to receive the actual probability sent by the second data holder after decrypting the first summation result, and calculate the calibration parameter of the federated model according to the actual probability.

[0037] According to one aspect of the present disclosure, there is provided a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the calibration parameter calculation method of the federated model described in any one of the above, and the user classification method based on the federated model described in any one of the above.

[0038] According to one aspect of the present disclosure, there is provided an electronic device, including:

[0039] a processor; and

[0040] a memory, configured to store executable instructions of the processor;

[0041] wherein, the processor is configured to execute the calibration parameter calculation method of the federated model described in any one of the above, and the user classification method based on the federated model described in any one of the above by executing the executable instructions.

[0042] For the calibration parameter calculation method of the federated model provided by the embodiments of the present disclosure, on the one hand, since the bin to which the historical user belongs can be determined according to the prediction score; then, the encrypted user labels assigned within the bin are summed to obtain a first summation result, and the first summation result is sent to the second data holder; finally, the actual probability sent by the second data holder after decrypting the first summation result is received, and the calibration parameter of the federated model is calculated according to the actual probability, which realizes the calculation of the calibration parameter of the federated model by the first data holder with the model party and the second data holder with the label party without exposing their respective data, and thus realizes cooperative calibration; on the other hand, it realizes the calibration of the federated model, and solves the problem in the prior art that due to the lack of a federated version, the calculation requirement of the calibration parameter of the model across data islands cannot be met; on the further hand, since cooperative calculation can be performed without exposing their respective data, the security of the first user features and user labels is improved.

[0043] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] The accompanying drawings here are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present disclosure, and are used together with the specification to explain the principles of the present disclosure. Obviously, the accompanying drawings in the following description are only some embodiments of the present disclosure. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0045] Figure 1 A flowchart schematically showing a method for calculating calibration parameters of a federated model according to an exemplary embodiment of the present disclosure.

[0046] Figure 2 A flowchart schematically showing a method for calculating the calibration parameters of the federated model according to the actual probability according to an exemplary embodiment of the present disclosure.

[0047] Figure 3 A flowchart schematically showing a method for classifying users based on a federated model according to an exemplary embodiment of the present disclosure.

[0048] Figure 4 An exemplary diagram schematically showing linear interpolation according to an exemplary embodiment of the present disclosure.

[0049] Figure 5 A flowchart schematically showing another method for calculating calibration parameters of a federated model according to an exemplary embodiment of the present disclosure.

[0050] Figure 6 A flowchart schematically showing another method for classifying users based on a federated model according to an exemplary embodiment of the present disclosure.

[0051] Figure 7 A block diagram schematically showing a device for calculating calibration parameters of a federated model according to an exemplary embodiment of the present disclosure.

[0052] Figure 8 A block diagram schematically showing a device for classifying users based on a federated model according to an exemplary embodiment of the present disclosure.

[0053] Figure 9 A schematic diagram showing an electronic device for implementing the above method for calculating calibration parameters of a federated model and the method for classifying users based on a federated model according to an exemplary embodiment of the present disclosure.

[0054] Figure 10 A schematic diagram showing a computer-readable storage medium for storing the above method for calculating calibration parameters of a federated model and the method for classifying users based on a federated model according to an exemplary embodiment of the present disclosure. Detailed implementation manners

[0055] Example embodiments will now be described more fully with reference to the accompanying drawings. However, the example embodiments can be implemented in various forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of the embodiments of the present disclosure. However, those skilled in the art will recognize that the technical solutions of the present disclosure may be practiced without one or more of the specific details, or may be implemented using other methods, components, devices, steps, etc. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring aspects of the present disclosure.

[0056] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and thus repeated descriptions thereof will be omitted. Some of the block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, or in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0057] Currently, data and models have great value, and at the same time, people are paying more and more attention to data security and privacy protection. With the tightening of regulations, a lot of cross-industry and cross-enterprise data and models cannot be shared, forming data islands; in this context, federated learning has emerged and is used to solve the problem of joint data modeling on the premise that data cannot flow. In actual business applications, there are cases of cooperation between multiple enterprises or units. Among the two parties in cooperation, they respectively hold different features of the same user. For example, e-commerce and banks jointly build models, and this method is called vertical federated learning. In the federated scenario, homomorphic encryption is often used for communication to protect the data privacy of each participating party.

[0058] In this example embodiment, a method for calculating calibration parameters of a federated model is first provided. This method can run on a server, a server cluster, or a cloud server where the first data holder is located, etc.; of course, those skilled in the art can also run the method of the present disclosure on other platforms according to requirements, and no special limitation is made in this exemplary embodiment. Refer to Figure 1 As shown, the method for calculating calibration parameters of the federated model may include the following steps:

[0059] Step S110. Receive the encrypted user label and the user personal identification code of the historical user sent by the second data holder, and obtain the first user feature of the historical user according to the user personal identification code;

[0060] Step S120. Calculate the prediction score of the historical user according to the federated model and the first user feature, and determine the bin to which the historical user belongs according to the prediction score;

[0061] Step S130. Perform a summation operation on the encrypted user labels assigned to the bin to obtain a first summation result, and send the first summation result to the second data holder;

[0062] Step S140. Receive the actual probability sent by the second data holder after decrypting the first summation result, and calculate the calibration parameter of the federated model according to the actual probability.

[0063] In the above method for calculating the calibration parameter of the federated model, on the one hand, since the bin to which the historical user belongs can be determined according to the prediction score; then perform a summation operation on the encrypted user labels assigned to the bin to obtain a first summation result, and send the first summation result to the second data holder; finally, receive the actual probability sent by the second data holder after decrypting the first summation result, and calculate the calibration parameter of the federated model according to the actual probability, it realizes that the first data holder with the model and the second data holder with the labels can calculate the calibration parameter of the federated model without exposing their respective data, thereby realizing cooperative calibration; on the other hand, it realizes the calibration of the federated model, solving the problem in the prior art that due to the lack of a federated version, the calculation requirement for the calibration parameter of the model across data islands cannot be met; on the further hand, since cooperative calculation can be performed without exposing their respective data, the security of the first user feature and the user label is improved.

[0064] Hereinafter, the steps included in the method for calculating the calibration parameter of the federated model in the exemplary embodiment of the present disclosure will be explained and described in detail with reference to the accompanying drawings.

[0065] First, the nouns involved in the exemplary embodiment of the present disclosure will be explained and described.

[0066] Model calibration, a commonly used tool in scenarios concerned with conversion rates or default rates. In a typical binary classification problem, the output results are discrete 0 or 1, corresponding to "not converted or converted" and "fulfilled or not fulfilled" respectively in the fields of advertising marketing and financial risk control. However, in advertising marketing, different marketing costs may be incurred for populations with "different conversion rates", so simply outputting 0 or 1 is not sufficient. People hope to obtain a probability value between 0 and 1. Similarly, in the field of financial risk control, simply predicting whether a user will fulfill the contract is not enough. The probability of the user fulfilling the contract may be required, and further corresponding to a "scorecard" or "credit score" to make a more detailed risk assessment of the user. Generally, the numbers in the [0,1] output by the binary classification model itself do not strictly correspond to such probabilities and need to be calibrated by the model to make them as close as possible to the true probabilities while not decreasing the sorting level. Specifically, in the exemplary embodiments provided in this disclosure, the model is not necessarily accurate as j increases. Instead, when binning (bucketing) in this solution, it is already sorted. The process of model calibration can be intuitively understood as not changing the order but only adjusting the absolute values. It is necessary to perform PAVA (Pool Adjacent Violators Algorithm, isotonic regression algorithm) correction for monotonicity in advance, and the non-federated version of the entire process is called isotonic regression.

[0067] Among them, the specific process of the isotonic regression algorithm may include:

[0068] First, perform binning so that different ranges of predicted values are the dimensions for calibration. Calculate the conversion rate or default rate within the bins. Among them:

[0069] Second, perform isotonic regression calculation: utilize the model's sorting ability to reduce the impact of the sparsity problem. That is, if r i >r i+1 (violating the size order between bins), then merge the bins.

[0070] Second, explain and illustrate the application scenarios and invention purposes of the exemplary embodiments of this disclosure.

[0071] In actual application scenarios, the scales of data assets held by the two parties in data cooperation may not be equal. For example, Party A (the first data holder), which focuses on data accumulation and model development, may have more user data and richer user characteristics. Moreover, due to Party A's long-term model development in this industry, it already has relatively mature machine learning or deep learning models established based on its own rich data. For the cooperating Party B (the second data holder), except for its own user tags, the dimension and breadth of data accumulation are limited. It hopes to use Party A's model to predict its own customers and perform targeted model calibration before deployment. In the cooperation, both Party A and Party B hope that the model output can meet the specific scenario business requirements, and at the same time, they hope that this process does not expose their respective data and models. Under this premise, through the calibration parameter calculation method of the federated model described in the exemplary embodiments of the present disclosure, Party A and Party B can complete a calibration model specialized for Party B's users without disclosing the data and models, and Party B can use the calibrated model to predict future unlabeled data. Of course, if Party B needs, it can also initiate multiple calibration applications using the labels collected later to continuously optimize.

[0072] Secondly, in a calibration parameter calculation method of a federated model in the exemplary embodiments of the present disclosure:

[0073] In step S110, receive the encrypted user tags of historical users and the user personal identification code sent by the second data holder, and obtain the first user characteristics of the historical users according to the user personal identification code.

[0074] In this exemplary embodiment, first, the above-mentioned first data holder and second data holder are explained and described. Specifically, the first data holder may be, for example, an e-commerce platform, which has a relatively mature model in this field, a relatively wide user coverage rate, and the relevant behavior and portrait characteristics of these users. The second data holder has some of its own users and the user tags of its own users. The user tags may be, for example, in a marketing scenario, whether the user has successfully converted, or in a financial risk control scenario, whether the user has defaulted, etc. The second data holder may be a bank or other institutions that need to classify users. This exemplary embodiment does not make special restrictions on this.

[0075] Secondly, when the second data holder needs to calibrate the federal model, it can send the encrypted user tags of historical users and the user personal identification number to the first data holder; among them, the user personal identification number (PIN, Personal Identification Number) is a number that identifies the unique identity of the historical user, which can be the ciphertext of the historical user's mobile phone number (specifically, it can be obtained by encrypting the user's mobile phone number through MD5 encryption or ACES encryption, etc.), the device number, etc.; but in specific cases, it may also be the encryption of the ID card, and this example does not make special restrictions on this; at the same time, the encrypted user tag can be obtained by encrypting according to the actual user category of the historical user, and the actual user category can include the actual conversion rate or actual default rate of the user, etc.

[0076] Furthermore, for supervised machine learning or deep learning modeling (i.e., the above-mentioned federal model), the user tag, that is, the so-called y, is the target of model learning. Specifically in the example embodiment, in the financial risk control scenario, the own tag (user tag) of Party B can be whether there is a default; while in the advertising and marketing scenario, the own tag (user tag) of Party B can be whether there is a conversion. At the same time, in order to ensure the data security of Party B, Party B needs to homomorphically encrypt the user tag y i to obtain the ciphertext [[y i(i.e., encrypted user tags). The reason for encrypting user tags using homomorphic encryption is as follows: For confidentiality, since the real tags belong to Party B and Party A is not expected to know. For example, the tags of users user1, user2, user3, user4, user5 for Party B are 0, 0, 0, 1, 0 respectively. Assuming these users are assigned to bin j, Party A sums up the ciphertexts [[0]], [[0]], [[0]], [[1]], [[0]] (note that in the case of homomorphic encryption, the ciphertext [[0]]s are not all the same, so Party A has no way of knowing that the tag of user4 is different from others). The sum value is sent to Party B. Party B decrypts the sum of the ciphertexts to obtain the sum of the plaintexts = 1 and then passes it to Party A to ensure that Party A does not know who exactly defaulted or converted; this is also why the binning rules need to be determined by Party B in advance according to its own data scale. Because once the binning scheme proposed by Party A is too detailed and the number of users assigned to each bin is small, there is a risk of leakage. It should be noted here that, as needed, it can also be replaced with encryption algorithms that satisfy additive homomorphism such as Iterative Affine Homomorphic Encryption; even any fully homomorphic encryption algorithm (fully homomorphic encryption algorithms necessarily also satisfy the requirements of additive homomorphism, but may sacrifice some efficiency). Each algorithm has corresponding encryption and decryption schemes based on its own principles and can be selected according to actual needs. This example does not make special restrictions on this.

[0077] Finally, when the first data holder receives the encrypted user tags and the user personal identification code, it can obtain the above-mentioned first user characteristics according to the user identification code; among them, the first user characteristics can be the user portrait characteristics of the historical user and the relevant behavior characteristics of the historical user; specifically, the user portrait characteristics are some characteristics abstracted from information such as the demographic characteristics, network browsing content, consumption content, and social activities of the user, including but not limited to: gender, age, education level, price sensitivity, geographical location, occupation, family structure; the relevant behavior characteristics include but not limited to: browsing, adding to cart, clicking, purchasing, etc. under the target brand or category (such as brand: Apple, category: mobile phone) (which may be a sequence or the number of statistics), and so on.

[0078] In step S120, according to the federated model and the first user characteristics, calculate the prediction score of the historical user, and determine the bin to which the historical user belongs according to the prediction score.

[0079] In this example embodiment, first, input the first user characteristics into the federated model to obtain the prediction score of the historical user. Among them, the federated model can be a logistic regression model, a deep neural network learning model, or a supervised machine learning model, which can be selected according to actual needs, and this example does not make special restrictions on this; secondly, after obtaining the prediction score, the bin to which the historical user belongs can be determined according to the prediction score.

[0080] Meanwhile, in order to determine the bin to which the historical user belongs, first, the bin thresholds need to be binned according to the binning request sent by the second data holder. The specific binning method is as follows: receive the binning request sent by the second data holder, and initialize the bin thresholds according to the binning rules included in the binning request to obtain multiple bins; among them, the starting endpoint of the bin threshold is 0, and the ending endpoint is 1. For example, Party B (the second data holder) can propose a binning request for binning calculation, and the binning request includes specific binning rules, such as equal-width binning, equal-frequency binning, and how many bins need to be allocated specifically, etc.; when Party A (the first data holder) receives the binning request, it can initialize the bin thresholds according to the binning rules, and the boundary values of each bin obtained can specifically include: 0 ≤ v1 < v2 <... < v k <v k+1 ≤ 1; then, based on the boundary values of each bin, [v1, v2), [v2, v3),..., [v k , v k+1 ) a total of k bins can be generated; it should be supplemented here that the specific binning steps can also be completely determined by Party A according to the needs of Party B (that is, Party A can determine the binning rules), but considering that if the binning is too fine, the labels of Party B may have a risk of leakage, so even if the specific binning scheme is determined by Party A, at least Party B needs to know.

[0081] Finally, after obtaining the bins, the bin to which the historical user belongs can be determined according to the value range interval where the prediction score is located.

[0082] In step S130, perform a summation operation on the encrypted user labels assigned to the bin to obtain a first summation result, and send the first summation result to the second data holder.

[0083] In this example embodiment, calculate the first summation result of the encrypted user labels in each bin j Meanwhile, according to the properties of homomorphic encryption Then send the first summation result to Party B, and Party B decrypts it to obtain the actual probability of the plaintext And through the actual probability of the plaintext Dividing by the number of samples included in the bin, the actual conversion rate or default rate of each bin can be obtained, denoted as r j, and then transmit it back to Party A; among them, bin j is the binning of the sample of

[0084] In step S140, receive the actual probability sent by the second data holder after decrypting the first summation result, and calculate the calibration parameter of the federated model according to the actual probability.

[0085] In this exemplary embodiment, after receiving the actual probability, the calibration parameter of the federated model can be calculated according to the actual probability. Specifically, referring to Figure 2 shown, the following steps can be included:

[0086] Step S210, sort the actual probabilities according to the encoding of the bin to which the actual probabilities belong, and determine whether the sorted actual probabilities increase monotonically;

[0087] Step S220, if so, use the sorted actual probabilities as the calibration parameters of the federated model;

[0088] Step S230, if not, perform monotonicity calibration on the sorted actual probabilities based on a preset isotonic regression algorithm to obtain the target probability, and use the target probability as the calibration parameter of the federated model.

[0089] Specifically, performing monotonicity calibration on the sorted actual probabilities based on a preset isotonic regression algorithm to obtain the target probability may include: First, extract the non-monotonically increasing actual probabilities from the sorted actual probabilities, and when determining that the positions of the non-monotonically increasing actual probabilities in the sorted actual probabilities are adjacent positions, merge the bins corresponding to the non-monotonically increasing actual probabilities; Secondly, calculate the bin probabilities of the merged bins, and update the sorted actual probabilities according to the bin probabilities of the merged bins to obtain the target probability. Among them, calculating the bin probabilities of the merged bins may include: calculating the bin probabilities of the merged bins based on the non-monotonically increasing actual probabilities and the bin boundary values of the bins corresponding to the non-monotonically increasing actual probabilities.

[0090] Next, steps S210 - S230 will be explained and described. First, the actual probabilities are sorted in ascending order according to the size of the encoding of the bin to which the actual probability belongs, and it is determined whether the sorted actual probabilities increase monotonically; if so, the sorted actual probabilities are used as the calibration parameters of the federated model; if not, the non-monotonically increasing actual probabilities are extracted from the sorted actual probabilities, and then the pairwise adjacent non-monotonically increasing actual probabilities are combined, and the bins corresponding to the combined non-monotonically increasing actual probabilities are merged; then the bin probabilities of the merged bins are calculated. Among them, the specific calculation method can be shown in the following formula (1):

[0091]

[0092] Among them, r new is the bin probability of the merged bin, r i and r i+1 are the actual probabilities respectively, v i and v i+1 are the bin boundary values of the bin.

[0093] Furthermore, when the bin probabilities of the merged bins are obtained, the sorted actual probabilities can be updated according to the bin probabilities of the merged bins to obtain the target probabilities, that is, using the calibrated r1, r2,..., r k to replace the original values, and finally satisfying the monotonicity requirement of r1 ≤ r2 ≤... ≤ r k . Thus, Party A can store the parameters r1 ≤ r2 ≤... ≤ r k , and send a calibration completion message to Party B. Thus, the federated model calibration process ends.

[0094] This exemplary embodiment of the present disclosure also provides a user classification method based on a federated model, configured for a first data holder that provides first user features in federated learning. Referring to Figure 3 as shown, the user classification method based on the federated model may include the following steps:

[0095] Step S310, receiving the user personal identification code of the user to be classified sent by the second data holder, and obtaining the second user features of the user to be classified according to the user personal identification code;

[0096] Step S320, inputting the second user features into the federated model to obtain the initial prediction probability of the user to be classified, and determining the bin to which the user to be classified belongs according to the initial prediction probability;

[0097] Step S330: Determine the target prediction probability of the user to be classified according to the initial prediction probability and the calibration parameter of the bin to which the user to be classified belongs; wherein, the calibration parameter is calculated by the calibration parameter calculation method of the federated model described in any of the foregoing items.

[0098] Among them, determining the target prediction probability of the user to be classified according to the initial prediction probability and the calibration parameter of the bin to which the user to be classified belongs includes: First, calculate the calibration coefficient according to the initial prediction probability and the bin boundary values of the bin to which the user to be classified belongs; Second, calculate the target prediction probability of the user to be classified according to the calibration coefficient and the calibration parameter of the bin to which the user to be classified belongs.

[0099] Step S340: Send the target prediction probability to the second data holder so that the second data holder can determine the user category to which the user to be classified belongs according to the target prediction probability.

[0100] Next, steps S310 - S340 will be explained and described. Specifically, after the calibration of the federated model is completed, when Party B has new unlabeled users to be predicted (that is, needs to classify new users to be classified), the specific classification method can include: First, Party B sends the PIN (Personal Identification Number) of the user to be classified to Party A. Party A receives the PIN, extracts the first user feature of the user to be classified after colliding the PIN, and then inputs the first user feature into the federated model to obtain the initial prediction probability. And determine which bin this prediction probability belongs to. For example, it falls into bin j , and the bin boundary values of this bin are (v i , v i+1 ; Further, Party A calculates the calibrated target prediction probability according to the following formula (2):

[0101]

[0102] Wherein,

[0103] Specifically, it can be as Figure 4 shown; r i and r i+1 are the calibration parameters of the bin.

[0104] Finally, send the obtained target prediction probability to Party B, so that Party B can determine the corresponding solution for the user to be classified according to this target prediction probability.

[0105] Figure 3In the user classification method based on the federated model shown, since the first user feature of the user to be classified can be obtained according to the user personal identification code, and then the first user feature is input into the federated model to obtain the initial prediction probability of the user to be classified, and the bin to which the user to be classified belongs is determined according to the initial prediction probability; finally, according to the initial prediction probability and the calibration parameter of the bin to which the user to be classified belongs, the target prediction probability of the user to be classified is determined, which improves the accuracy of the target prediction probability and further improves the accuracy of the user classification result.

[0106] Next, the calibration parameter calculation method of the federated model of the exemplary embodiments of the present disclosure will be further explained and described in conjunction with Figure 5 This is further explained and illustrated. Referring to Figure 5 As shown, the calibration parameter calculation method of the federated model may specifically include the following steps:

[0107] Step S510, the first data holder receives the binning request sent by the second data holder and performs binning according to the binning rules;

[0108] Step S520, the second data holder encrypts the user labels of historical users to obtain encrypted user labels, and sends the encrypted user labels and the user personal identification code to the first data holder;

[0109] Step S530, the first data holder obtains the first user feature of the historical user according to the user personal identification code;

[0110] Step S530, the first data holder inputs the first user feature into the federated model to obtain a prediction score, and determines the bin to which the historical user belongs according to the prediction score;

[0111] Step S540, the first data holder performs a summation operation on the encrypted user labels assigned in the bin to obtain a first summation result, and sends the first summation result to the second data holder;

[0112] Step S550, the second data holder decrypts the first summation result to obtain the actual probability, and sends the actual probability to the first data holder;

[0113] Step S560, the first data holder determines whether the actual probability is monotonically increasing; if so, the actual probability is used as the calibration parameter; if not, the actual probability is corrected for monotonicity through the PAVA algorithm to obtain the target probability, and the target probability is used as the calibration parameter;

[0114] Step S570, the first data holder sends a message indicating that the calibration is completed to the second data holder.

[0115] In Figure 5In the method for calculating the calibration parameters of the federated model shown, the second data holder uses the homomorphic encryption method, enabling Party A with the model and feature data to cooperate with Party B with labels to complete the calculation of the calibration parameters for the business requirements of Party B; at the same time, it can also be based on the actual conversion rate or default rate of the samples with existing labels of Party B, without exposing the specific labels of the samples of Party B.

[0116] Next, in combination with Figure 6 This will further explain and illustrate the user classification method based on the federated model of the exemplary embodiments of the present disclosure. Refer to Figure 6 As shown, the user classification method based on the federated model may specifically include the following steps:

[0117] Step S610, the first data holder receives the user personal identification code of the user to be classified sent by the second data holder, and obtains the second user features of the user to be classified according to the user personal identification code;

[0118] Step S620, the first user holder inputs the second user features into the federated model to obtain the initial prediction probability of the user to be classified, and determines the bin to which the user to be classified belongs according to the initial prediction probability;

[0119] Step S630, the first user holder calculates the calibration coefficient according to the initial prediction probability and the bin boundary values of the bin to which the user to be classified belongs;

[0120] Step S640, the first user holder calculates the target prediction probability of the user to be classified according to the calibration coefficient and the calibration parameters of the bin to which the user to be classified belongs;

[0121] Step S650, the first user holder sends the target prediction probability to the second data holder;

[0122] Step S660, the second data holder determines the user category to which the user to be classified belongs according to the target prediction probability.

[0123] Figure 6 In the user classification method based on the federated model shown, when Party B has a new user to predict, first an initial prediction probability is output by the initial model of Party A, and the bin is determined according to this initial prediction probability. Then, the calibrated target prediction probability is calculated through the stored calibration parameters and returned to Party B; this target prediction probability can correspond to the probability with specific business meanings of Party B. For example, in the marketing scenario, the predicted conversion probability of this user, or in the financial risk control scenario, the predicted default probability of this user, so that the user can adopt corresponding solutions according to the specific probability to reduce resource waste and increase the conversion rate and reduce the default rate.

[0124] The exemplary embodiments of the present disclosure also provide a calibration parameter calculation device for a federated model, which is configured in a first data holder that provides first user features in federated learning. Refer to Figure 7 As shown, the calibration device for the federated model may include a first user feature acquisition module 710, a first bin determination module 720, a first calculation module 730, and a calibration parameter calculation module 740. Among them:

[0125] The first user feature acquisition module 710 may be configured to receive the encrypted user tags and user personal identification codes of historical users sent by a second data holder, and obtain the first user features of the historical users according to the user personal identification codes;

[0126] The first bin determination module 720 may be configured to calculate the prediction scores of the historical users according to the federated model and the first user features, and determine the bins to which the historical users belong according to the prediction scores;

[0127] The first calculation module 730 may be configured to perform a summation operation on the encrypted user tags assigned to the bins to obtain a first summation result, and send the first summation result to the second data holder;

[0128] The calibration parameter calculation module 740 may be configured to receive the actual probabilities sent by the second data holder after decrypting the first summation result, and calculate the calibration parameters of the federated model according to the actual probabilities.

[0129] In an exemplary embodiment of the present disclosure, the calibration parameter calculation device for the federated model further includes:

[0130] A binning module, which may be configured to receive a binning request sent by a second data holder, and initialize the binning thresholds according to the binning rules included in the binning request to obtain multiple bins;

[0131] Among them, the starting endpoint of the binning threshold is 0, and the ending endpoint is 1.

[0132] In an exemplary embodiment of the present disclosure, calculating the calibration parameters of the federated model according to the actual probabilities includes:

[0133] Sorting the actual probabilities according to the encoding of the bins to which the actual probabilities belong, and determining whether the sorted actual probabilities increase monotonically;

[0134] If so, using the sorted actual probabilities as the calibration parameters of the federated model;

[0135] Otherwise, perform monotonicity calibration on the sorted actual probabilities based on a preset isotonic regression algorithm to obtain target probabilities, and use the target probabilities as the calibration parameters of the federated model.

[0136] In an exemplary embodiment of the present disclosure, performing monotonicity calibration on the sorted actual probabilities based on a preset isotonic regression algorithm to obtain target probabilities includes:

[0137] Extract the non-monotonically increasing actual probabilities from the sorted actual probabilities, and when it is determined that the positions of the non-monotonically increasing actual probabilities in the sorted actual probabilities are adjacent positions, merge the bins corresponding to the non-monotonically increasing actual probabilities;

[0138] Calculate the bin probabilities of the merged bins, and update the sorted actual probabilities according to the bin probabilities of the merged bins to obtain the target probabilities.

[0139] In an exemplary embodiment of the present disclosure, calculating the bin probabilities of the merged bins includes:

[0140] Calculate the bin probabilities of the merged bins based on the non-monotonically increasing actual probabilities and the bin boundary values of the bins corresponding to the non-monotonically increasing actual probabilities.

[0141] The present disclosure also provides a user classification device based on a federated model, configured in a first data holder that provides first user features in federated learning. Refer to Figure 8 As shown, the user classification device based on the federated model may include a second user feature acquisition module 810, a second bin determination module 820, a target prediction probability determination module 830, and a user classification module 840. Among them:

[0142] The second user feature acquisition module 810 may be configured to receive the user personal identification code of the user to be classified sent by the second data holder, and obtain the second user features of the user to be classified according to the user personal identification code;

[0143] The second bin determination module 820 may be configured to input the second user features into the federated model to obtain the initial prediction probability of the user to be classified, and determine the bin to which the user to be classified belongs according to the initial prediction probability;

[0144] The target prediction probability determination module 830 may be configured to determine the target prediction probability of the user to be classified according to the initial prediction probability and the calibration parameters of the bin to which the user to be classified belongs; wherein, the calibration parameters are calculated by the calibration parameter calculation method of the federated model described in any one of the foregoing;

[0145] The user classification module 840 can be used to send the target prediction probability to the second data holder, so that the second data holder determines the user category to which the user to be classified belongs according to the target prediction probability.

[0146] In an exemplary embodiment of the present disclosure, determining the target prediction probability of the user to be classified according to the initial prediction probability and the calibration parameter of the bin to which the user to be classified belongs includes:

[0147] Calculating a calibration coefficient according to the initial prediction probability and the bin boundary value of the bin to which the user to be classified belongs;

[0148] Calculating the target prediction probability of the user to be classified according to the calibration coefficient and the calibration parameter of the bin to which the user to be classified belongs.

[0149] The specific details of each module in the above-mentioned calibration parameter calculation device of the federated model and the user classification device based on the federated model have been described in detail in the corresponding calibration parameter calculation method of the federated model and the user classification method based on the federated model, so they will not be elaborated here.

[0150] It should be noted that although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of the two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0151] In addition, although the steps of the methods in the present disclosure are described in a specific order in the drawings, this does not require or imply that these steps must be executed in this specific order, or that all the steps shown must be executed to achieve the desired result. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step for execution, and / or one step may be decomposed into multiple steps for execution, etc.

[0152] In an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

[0153] Those skilled in the art to which the present disclosure pertains can understand that various aspects of the present disclosure can be implemented as a system, method, or program product. Therefore, various aspects of the present disclosure can be specifically implemented in the following forms, namely: a complete hardware implementation, a complete software implementation (including firmware, microcode, etc.), or an implementation combining hardware and software aspects, which can be collectively referred to herein as "circuit", "module", or "system".

[0154] Next, refer toFigure 9 Describe the electronic device 900 according to this embodiment of the present disclosure. Figure 9 The shown electronic device 900 is merely an example and should not impose any limitations on the functions and usage scope of the embodiments of the present disclosure.

[0155] As Figure 9 shown, the electronic device 900 is presented in the form of a general computing device. The components of the electronic device 900 may include but are not limited to: at least one of the above-mentioned processing units 910, at least one of the above-mentioned storage units 920, a bus 930 connecting different system components (including the storage unit 920 and the processing unit 910), and a display unit 940.

[0156] Among them, the storage unit stores program code, and the program code can be executed by the processing unit 910, so that the processing unit 910 executes the steps according to various exemplary embodiments of the present disclosure described in the above "Exemplary Method" section of this specification. For example, the processing unit 910 may execute steps S110 as shown in Figure 1 : Receive the encrypted user tag and the user personal identification code of the historical user sent by the second data holder, and obtain the first user feature of the historical user according to the user personal identification code; step S120: Calculate the prediction score of the historical user according to the federated model and the first user feature, and determine the bin to which the historical user belongs according to the prediction score; step S130: Perform a summation operation on the encrypted user tags assigned to the bin to obtain a first summation result, and send the first summation result to the second data holder; step S140: Receive the actual probability sent by the second data holder after decrypting the first summation result, and calculate the calibration parameter of the federated model according to the actual probability.

[0157] The processing unit 910 may execute steps S130 as shown in Figure 3 : Receive the user personal identification code of the user to be classified sent by the second data holder, and obtain the second user feature of the user to be classified according to the user personal identification code; step S320: Input the second user feature into the federated model to obtain the initial prediction probability of the user to be classified, and determine the bin to which the user to be classified belongs according to the initial prediction probability; step S330: Determine the target prediction probability of the user to be classified according to the initial prediction probability and the calibration parameter of the bin to which the user to be classified belongs; wherein, the calibration parameter is calculated by the calibration parameter calculation method of the federated model described in any one of the foregoing; step S340: Send the target prediction probability to the second data holder so that the second data holder determines the user category to which the user to be classified belongs according to the target prediction probability.

[0158] The storage unit 920 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 9201 and / or a cache storage unit 9202, and may further include a read-only storage unit (ROM) 9203.

[0159] The storage unit 920 may also include a program / utilities 9204 having a set (at least one) of program modules 9205. Such program modules 9205 include, but are not limited to: an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment.

[0160] The bus 930 may represent one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus structures.

[0161] The electronic device 900 may also communicate with one or more external devices 1000 (such as a keyboard, a pointing device, a Bluetooth device, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 900, and / or may communicate with any device that enables the electronic device 900 to communicate with one or more other computing devices (such as a router, a modem, etc.). Such communication may be carried out through an input / output (I / O) interface 950. Moreover, the electronic device 900 may also communicate with one or more networks (such as a local area network (LAN), a wide area network (WAN), and / or a public network, such as the Internet) through a network adapter 960. As shown in the figure, the network adapter 960 communicates with other modules of the electronic device 900 through the bus 930. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 900, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems, etc.

[0162] Through the description of the above embodiments, those skilled in the art can easily understand that the example embodiments described herein can be implemented by software, or can be implemented by the way of software in combination with necessary hardware. Therefore, the technical solutions according to the embodiments of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

[0163] In an exemplary embodiment of the present disclosure, there is also provided a computer-readable storage medium, on which a program product capable of implementing the above-described method of this specification is stored. In some possible implementation manners, various aspects of the present disclosure may also be implemented in the form of a program product, which includes program code. When the program product runs on a terminal device, the program code is used to cause the terminal device to execute the steps according to various exemplary embodiments of the present disclosure described in the above "Exemplary Method" section of this specification.

[0164] Referring Figure 10 As shown, a program product 1010 for implementing the above method according to an embodiment of the present disclosure is described. It may be in the form of a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, a readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0165] The program product may adopt any combination of one or more readable media. The readable media may be a readable signal medium or a readable storage medium. The readable storage medium may, for example, but not be limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples (non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.

[0166] The computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries the readable program code. Such a propagated data signal may take various forms, including but not limited to an electromagnetic signal, an optical signal, or any suitable combination of the above. The readable signal medium may also be any readable medium other than the readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0167] The program code contained on the readable medium may be transmitted by any appropriate medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination of the above.

[0168] Program code for performing the operations of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages such as Java, C++, etc., and also including conventional procedural programming languages such as the "C" language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user's computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider). In addition, the above drawings are only schematic illustrations of the processes included in the method according to the exemplary embodiments of the present disclosure, and are not for limiting purposes. It is easy to understand that the processes shown in the above drawings do not indicate or limit the chronological order of these processes. Additionally, it is also easy to understand that these processes may be executed, for example, synchronously or asynchronously in multiple modules.

[0169] Other embodiments of the present disclosure will be readily apparent to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include known common knowledge or conventional technical means in the art that are not invented by the present disclosure. The specification and embodiments are to be considered exemplary only, and the true scope and spirit of the present disclosure are pointed out by the claims.

Claims

1. A method for calculating calibration parameters of a federated model, characterized in that, The first data holder configured to provide the first user features in federated learning, the method for calculating the calibration parameters of the federated model includes: Receiving the encrypted user tags and user personal identification codes of historical users sent by the second data holder, and obtaining the first user features of the historical users according to the user personal identification codes; Calculating the prediction scores of the historical users according to the federated model and the first user features, and determining the bin to which the historical users belong according to the prediction scores; Performing a summation operation on the encrypted user tags assigned to the bin to obtain a first summation result, and sending the first summation result to the second data holder; Receiving the actual probability sent by the second data holder after decrypting the first summation result, and calculating the calibration parameters of the federated model according to the actual probability.

2. The method for calculating calibration parameters of a federated model according to claim 1, characterized in that, The method for calculating the calibration parameters of the federated model further includes: Receiving a binning request sent by the second data holder, and initializing the binning thresholds according to the binning rules included in the binning request to obtain multiple bins; Wherein, the starting endpoint of the binning threshold is 0 and the ending endpoint is 1.

3. The method for calculating calibration parameters of a federated model according to claim 1, characterized in that, Calculating the calibration parameters of the federated model according to the actual probability includes: Sorting the actual probability according to the encoding of the bin to which the actual probability belongs, and determining whether the sorted actual probability increases monotonically; If so, using the sorted actual probability as the calibration parameters of the federated model; If not, performing monotonicity calibration on the sorted actual probability based on a preset isotonic regression algorithm to obtain a target probability, and using the target probability as the calibration parameters of the federated model.

4. The method for calculating calibration parameters of a federated model according to claim 3, characterized in that, Performing monotonicity calibration on the sorted actual probability based on a preset isotonic regression algorithm to obtain a target probability, including: Extracting the non-monotonically increasing actual probability from the sorted actual probability, and when it is determined that the position of the non-monotonically increasing actual probability in the sorted actual probability is an adjacent position, merging the bins corresponding to the non-monotonically increasing actual probability; Calculating the bin probabilities of the merged bins, and updating the sorted actual probability according to the bin probabilities of the merged bins to obtain the target probability.

5. The method for calculating calibration parameters of a federated model according to claim 4, characterized in that, Calculating the bin probabilities of the merged bins includes: Calculating the bin probabilities of the merged bins according to the non-monotonically increasing actual probability and the bin boundary values of the bins corresponding to the non-monotonically increasing actual probability.

6. A method for user classification based on a federated model, characterized in that, The first data holder configured to provide the first user features in federated learning, the user classification method based on the federated model includes: Receiving the user personal identification code of the user to be classified sent by the second data holder, and obtaining the second user features of the user to be classified according to the user personal identification code; Inputting the second user features into the federated model to obtain the initial prediction probability of the user to be classified, and determining the bin to which the user to be classified belongs according to the initial prediction probability; Determine the target prediction probability of the user to be classified according to the initial prediction probability and the calibration parameter of the bin to which the user to be classified belongs; wherein, the calibration parameter is calculated by the calibration parameter calculation method of the federated model according to any one of claims 1-5; Send the target prediction probability to the second data holder so that the second data holder determines the user category to which the user to be classified belongs according to the target prediction probability.

7. The method for user classification based on a federated model according to claim 6, characterized in that, Determining the target prediction probability of the user to be classified according to the initial prediction probability and the calibration parameter of the bin to which the user to be classified belongs includes: Calculate a calibration coefficient according to the initial prediction probability and the bin boundary values of the bin to which the user to be classified belongs; Calculate the target prediction probability of the user to be classified according to the calibration coefficient and the calibration parameter of the bin to which the user to be classified belongs.

8. A calibration device for a federated model, characterized in that, A first data holder configured to provide first user features in federated learning, the calibration device of the federated model includes: A first user feature acquisition module, configured to receive the encrypted user label and the user personal identification code of the historical user sent by the second data holder, and obtain the first user feature of the historical user according to the user personal identification code; A first bin determination module, configured to calculate the prediction score of the historical user according to the federated model and the first user feature, and determine the bin to which the historical user belongs according to the prediction score; A first calculation module, configured to perform a summation operation on the encrypted user labels assigned to the bin to obtain a first summation result, and send the first summation result to the second data holder; A calibration parameter calculation module, configured to receive the actual probability sent by the second data holder after decrypting the first summation result, and calculate the calibration parameter of the federated model according to the actual probability.

9. A computer-readable storage medium, on which a computer program is stored, characterized in that, When the computer program is executed by a processor, it implements the calibration parameter calculation method of the federated model according to any one of claims 1-5, and the user classification method based on the federated model according to any one of claims 6-7.

10. An electronic device, characterized in that, Comprising: A processor; And A memory for storing executable instructions of the processor; Wherein, the processor is configured to execute the calibration parameter calculation method of the federated model according to any one of claims 1-5 and the user classification method based on the federated model according to any one of claims 6-7 by executing the executable instructions.

Images