WO2020233137A1 - 损失函数取值的确定方法、装置和电子设备 - Google Patents

损失函数取值的确定方法、装置和电子设备 Download PDF

Info

Publication number
WO2020233137A1
WO2020233137A1 PCT/CN2020/070939 CN2020070939W WO2020233137A1 WO 2020233137 A1 WO2020233137 A1 WO 2020233137A1 CN 2020070939 W CN2020070939 W CN 2020070939W WO 2020233137 A1 WO2020233137 A1 WO 2020233137A1
Authority
WO
WIPO (PCT)
Prior art keywords
share
data
value
item
term
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Ceased
Application number
PCT/CN2020/070939
Other languages
English (en)
French (fr)
Inventor
周亚顺
李漓春
殷山
王华忠
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Advanced New Technologies Co Ltd
Original Assignee
Advanced New Technologies Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Advanced New Technologies Co Ltd filed Critical Advanced New Technologies Co Ltd
Priority to US16/786,337 priority Critical patent/US10956597B2/en
Publication of WO2020233137A1 publication Critical patent/WO2020233137A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING OR CALCULATING; COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10Complex mathematical operations
    • G06F17/15Correlation function computation including computation of convolution operations

Definitions

  • the embodiments of this specification relate to the field of computer technology, and in particular to a method, device and electronic equipment for determining the value of a loss function.
  • the value of the loss function of the data processing model needs to be calculated; the value of the loss function can measure the training effect of the data processing model (such as over-fitting, under-fitting, etc.), and then decide whether to terminate the training . Since the data used to train the data processing model is scattered among the parties involved in the cooperative modeling, in related technologies, the data of the parties involved in the cooperative modeling is usually collected in an independent third party. To calculate the value of the loss function. Since the data of the parties involved in the cooperative modeling are summarized, it is easy to cause the leakage of enterprise data.
  • the purpose of the embodiments of this specification is to provide a method, device and electronic device for determining the value of the loss function, so that the value of the loss function can be calculated by the modeled data parties under the premise of protecting data privacy.
  • a method for determining the value of a loss function including a first term and a second term; the method includes: according to the share of the first parameter Secretly share the first data with the partner to obtain the share of the first data; determine the share of the first item based on the share of the first data; share the second data secretly with the partner according to the share of the second parameter and the random number, Obtain the share of the second data; determine the coefficient of the second item according to the random number, and the coefficient of the second item and the second data are used to jointly determine the value of the second item; send the share of the value of the first item, the first item to the partner Second, the share of data and the coefficient of the second term so that the partner can determine the value of the loss function.
  • a device for determining the value of a loss function where the loss function includes a first term and a second term; the device includes: a first secret sharing unit, Used to secretly share the first data with the partner according to the share of the first parameter to obtain the share of the first data; the first determining unit is used to determine the share of the value of the first item according to the share of the first data; the second secret The sharing unit is used to secretly share the second data with the partner according to the share of the second parameter and the random number to obtain the share of the second data; the determining unit is used to determine the coefficient of the second term and the coefficient of the second term according to the random number And the second data are used to jointly determine the value of the second item; the sending unit is used to send the share of the value of the first item, the share of the second data and the coefficient of the second item to the partner, so that the partner can determine the The value of the loss function.
  • a first secret sharing unit Used to secretly share the first data with the partner according to the share of the first parameter to obtain the share
  • an electronic device including: a memory, configured to store computer instructions; a processor, configured to execute the computer instructions to implement the computer instructions described in the first aspect Method steps.
  • a method for determining the value of a loss function includes a first term and a second term; the method includes: according to the share of the first parameter Secretly share the first data with the tag value and the partner to obtain the share of the first data; determine the first share of the value of the first item according to the share of the first data; share the second share secretly with the partner according to the share of the second parameter Data, get the first share of the second data; receive the second share of the first item value, the second share of the second data and the coefficient of the second item, the coefficient of the second item and the second data from the partner Used to jointly determine the value of the second item; according to the first share of the first item, the second share of the first item, the first share of the second data, the second share of the second data, and the second The coefficient of the term determines the value of the loss function.
  • a device for determining the value of a loss function includes a first term and a second term; the device includes: a first secret sharing unit, Used to secretly share the first data with the partner according to the share and tag value of the first parameter to obtain the share of the first data; the first determining unit is used to determine the first value of the first item according to the share of the first data Share; the second secret sharing unit is used to secretly share the second data with the partner according to the share of the second parameter to obtain the first share of the second data; the receiving unit is used to receive the value of the first item sent by the partner The second share of the second data, the second share of the second data and the coefficient of the second term, the coefficient of the second term and the second data are used to jointly determine the value of the second term; the determining unit is used to obtain the value according to the first term The value of the loss function is determined by the first share of, the second share of the value of the first item, the first share of
  • an electronic device including: a memory, configured to store computer instructions; and a processor, configured to execute the computer instructions to implement the computer instructions described in the fourth aspect Method steps.
  • the first data party and the second data party can use the secret sharing algorithm to jointly calculate the loss function without revealing the data they own. Therefore, it is convenient to measure the training effect of the data processing model according to the value of the loss function, and then decide whether to terminate the training.
  • Fig. 1 is a flowchart of a method for determining the value of a loss function according to an embodiment of the specification
  • FIG. 2 is a flowchart of a method for determining the value of a loss function according to an embodiment of the specification
  • FIG. 3 is a flowchart of a method for determining the value of a loss function according to an embodiment of the specification
  • FIG. 4 is a functional structure diagram of a device for determining the value of a loss function in an embodiment of the specification
  • Fig. 5 is a functional structure diagram of a device for determining the value of a loss function according to an embodiment of the specification
  • Fig. 6 is a functional structure diagram of an electronic device according to an embodiment of the specification.
  • Secret Sharing is an algorithm to protect data privacy. Multiple data parties can use secret sharing algorithms to perform collaborative calculations to obtain secret information without leaking their own data. Each data party can obtain a share of the secret information. A single data party cannot recover the secret information. Only multiple data parties can work together to recover the secret information.
  • the data party P 1 owns the data x 1
  • the data party P 2 owns the data x 2 .
  • the data party P 1 can obtain the share y 1 of the secret information y after the calculation
  • the data party P 2 can obtain the share y 2 of the secret information y after the calculation.
  • Loss function can be used to measure the degree of inconsistency between the predicted value of the data processing model and the true value. The smaller the value of the loss function, the better the robustness of the data processing model. In the process of training the data processing model, the value of the loss function can be calculated to measure the training effect of the data processing model (such as over-fitting, under-fitting, etc.), and then decide whether to terminate the training.
  • the data processing model includes but is not limited to logistic regression model, linear regression model, neural network model, etc. Different data processing models can be measured by different loss functions. For example, a logistic regression model can be measured by a logarithmic loss function, and a linear regression model can be measured by a square loss function (Square Loss).
  • the number of data parties performing cooperative security modeling is two.
  • One of the data parties can have the complete sample data, and the other data party can own the label value of the sample data; or, one of the data parties can own part of the data items in the sample data, and the other data party can own the other part of the sample data
  • the label value of the data item and sample data includes the user's savings amount and loan amount.
  • One of the data parties can have the user's savings amount, and the other data party can have the user's loan amount and the label value of the sample data.
  • the multiple data parties need to cooperate to calculate the value of the loss function to decide whether to terminate the training.
  • the data used to train the data processing model is scattered among the data parties of the cooperative modeling, if the secret sharing algorithm is adopted, the data parties of the cooperative modeling can be based on the premise of not leaking their own data. The data owned by oneself cooperate to calculate the value of the loss function.
  • This specification provides an embodiment of a method for determining the value of the loss function.
  • This embodiment may include the following steps.
  • Step S101 The first data party secretly shares the first data according to the first share of the first parameter, and the second data party secretly shares the first data according to the second share of the first parameter and the tag value.
  • the first data party obtains the first share of the first data
  • the second data party obtains the second share of the first data.
  • Step S103 The first data party determines the first share of the value of the first item according to the first share of the first data.
  • Step S105 The second data party determines the second share of the value of the first item according to the second share of the first data.
  • Step S107 The first data party secretly shares the second data according to the first share of the second parameter and the random number, and the second data party secretly shares the second data according to the second share of the second parameter.
  • the first data party obtains the first share of the second data
  • the second data party obtains the second share of the second data.
  • Step S109 The first data party determines the coefficient of the second term according to the random number.
  • Step S111 The first data sends the first share of the value of the first item, the first share of the second data, and the coefficient of the second item to the second data party.
  • Step S113 The second data party receives the first share of the value of the first item, the first share of the second data, and the coefficient of the second item.
  • Step S115 The second data party uses the first share of the first item value, the second share of the first item value, the first share of the second data, the second share of the second data, and the coefficient of the second item, Determine the value of the loss function.
  • the first term and the second term are respectively function terms in the loss function.
  • the loss function may be a log loss function m represents the number of sample data; x i represents the i-th sample data; y i represents the label value of the sample data x i ; ⁇ represents the model parameter of the data processing model; h ⁇ (x i ) represents the activation function of the data processing model Value,
  • the first item can be
  • the second item can be
  • the first data party and the second data party are respectively two parties of cooperative security modeling.
  • the first data party may be a data party that does not have a tag value
  • the second data party may be a data party that has a tag value.
  • the first data party may have complete sample data
  • the second data party may have the label value of the sample data.
  • the first data party may own part of the data items of the sample data
  • the second data party may own another part of the data items and label values of the sample data.
  • the tag value can be used to distinguish different types of sample data, and the specific value can be taken from 0 and 1, for example.
  • the data party may be an electronic device.
  • the electronic equipment may include a personal computer, a server, a handheld device, a portable device, a tablet device, a multi-processor device; or, it may also include a cluster formed by any of the above devices or devices.
  • the first parameter and the second parameter are respectively intermediate results obtained by the first data party and the second data party in the cooperative security modeling process.
  • the first parameter and the second parameter are different.
  • the first parameter may be the product of the sample data and the model parameter of the data processing model
  • the second parameter may be the value of the activation function of the data processing model.
  • the first data party and the second data party respectively obtain a share of the first parameter.
  • the share obtained by the first data party may be used as the first share of the first parameter
  • the share obtained by the second data party may be used as the second share of the first parameter.
  • the sum of the first share of the first parameter and the second share of the first parameter is the first parameter.
  • the number of the first parameter may be multiple. In this way, the first data party may have first shares of multiple first parameters, and the second data party may have second shares of multiple first parameters.
  • the first parameter can be expressed as ⁇ x i
  • the first share of the first parameter can be expressed as ⁇ x i > 0
  • the second share of the first parameter can be expressed as ⁇ x i > 1 .
  • ⁇ x i > 0 + ⁇ x i > 1 ⁇ x i .
  • the first data party and the second data party respectively obtain a share of the second parameter.
  • the share obtained by the first data party may be used as the first share of the second parameter
  • the share obtained by the second data party may be used as the second share of the second parameter.
  • the sum of the first share of the second parameter and the second share of the second parameter is the second parameter.
  • the number of the second parameter may be multiple. In this way, the first data party may have first shares of multiple second parameters, and the second data party may have second shares of multiple second parameters.
  • the second parameter can be expressed as The first share of the second parameter can be expressed as ⁇ h ⁇ (x i )> 0 , and the second share of the second parameter can be expressed as ⁇ h ⁇ (x i )> 1 .
  • ⁇ h ⁇ (x i )> 0 + ⁇ h ⁇ (x i )> 1 h ⁇ (x i ).
  • the first data and the second data are respectively calculated by the first data party and the second data party through a secret sharing algorithm.
  • the first data is used to determine the value of the first item.
  • the second data and the coefficient of the second term can be combined to determine the value of the second term.
  • the first data can be expressed as (1-y i ) ⁇ x i
  • the second data can be expressed as r i h ⁇ (x i )
  • the coefficient of the second term can be expressed as
  • the third data and the fourth data are respectively calculated by the first data party and the second data party through a secret sharing algorithm. According to the share of the third data, the share of the first data can be determined. According to the share of the fourth data, the share of the second data can be determined.
  • the third data can be expressed as (1-y i ) ⁇ x i > 0
  • the fourth data can be expressed as r i ⁇ h ⁇ (x i )> 1 .
  • the first data party may secretly share the first share according to the first share of the first parameter
  • the second data party may secretly share the first share according to the second share of the first parameter and the tag value.
  • the first data party and the second data party may respectively obtain a share of the first data.
  • the share obtained by the first data party may be used as the first share of the first data
  • the share obtained by the second data party may be used as the second share of the first data.
  • the sum of the first share of the first data and the second share of the first data is the first data.
  • the first data party may have first shares of multiple first parameters, and the second data party may have second shares of multiple first parameters.
  • the first data party can secretly share a share based on the first share of each first parameter
  • the second data party can secretly share a share based on the second share of the first parameter and the tag value corresponding to the first parameter.
  • First data The first data party can obtain a first share of the first data
  • the second data party can obtain a second share of the first data. It is worth noting that the correspondence between the first parameter and the label value can be understood as: the sample data corresponding to the label value and the sample data used to calculate the first parameter are the same sample data.
  • the first data party may secretly share a piece of third data according to the first share of each first parameter
  • the second data party may secretly share a piece of third data according to the tag value corresponding to the first parameter.
  • the first data party and the second data party may each obtain a share of the third data.
  • the share obtained by the first data party may be taken as the first share of the third data
  • the share obtained by the second data party may be taken as the second share of the third data.
  • the sum of the first share of the third data and the second share of the third data is the third data.
  • the first data party may directly use the first share of the third data as the first share of the first data.
  • the second data party may calculate the second share of the first parameter and the label value corresponding to the first parameter according to a preset calculation rule; may add the calculation result to the second share of the third data , The addition result can be used as the second share of the first data.
  • the first data party may secretly share the third data (1-y i ) ⁇ x i > 0 according to ⁇ x i > 0 and the second data party may secretly share the third data (1-y i ) ⁇ x i > 0 according to 1-y i .
  • the first data party can obtain the first share of the third data ⁇ [(1-y i ) ⁇ x i > 0 ]> 0 .
  • the second data party can obtain the second share of the third data ⁇ [(1-y i ) ⁇ x i > 0 ]> 1 .
  • the first data party may directly use the first share of the third data ⁇ [(1-y i ) ⁇ x i > 0 ]> 0 as the first share of the first data (1-y i ) ⁇ x i ⁇ ( 1-y i ) ⁇ x i > 0 .
  • the second data can be calculated according to ⁇ x i > 1 and y i to obtain (1-y i ) ⁇ x i > 1 ; the calculation result (1-y i ) ⁇ x i > 1 can be compared with the third data
  • the second share ⁇ [(1-y i ) ⁇ x i > 0 ]> 1 is added; the addition result can be used as the second share of the first data (1-y i ) ⁇ x i ⁇ (1-y i ) ⁇ x i > 1 . among them,
  • the first data party can obtain a first share of a plurality of first data.
  • the first data party may accumulate the first shares of the multiple first data; the first share of the value of the first item may be determined according to the accumulation result. Specifically, the first data party may divide the accumulation result by the quantity of the first data (that is, the quantity of the first parameter) to obtain the first share of the value of the first item.
  • the first data party can calculate As the first share of the value of the first item.
  • the second data party can obtain a second share of a plurality of first data.
  • the second data party may accumulate the second shares of the multiple first data; the second share of the value of the first item may be determined according to the accumulation result.
  • the first data party may divide the accumulation result by the quantity of the first data (that is, the quantity of the first parameter) to obtain the second share of the value of the first item.
  • the sum of the first share of the value of the first item and the second share of the value of the first item is the value of the first item.
  • the second data party can calculate As the second share of the value of the first item. among them,
  • the first data party may secretly share the second share according to the first share of the second parameter and the random number
  • the second data party may secretly share the second share according to the second share of the second parameter. data.
  • the first data party and the second data party may respectively obtain a share of the second data.
  • the share obtained by the first data party may be used as the first share of the second data
  • the share obtained by the second data party may be used as the second share of the second data.
  • the sum of the first share of the second data and the second share of the second data is the second data.
  • the number of second parameters can be multiple.
  • the first data party may generate multiple random numbers, and each random number may correspond to a second parameter (or a first share of the second parameter).
  • the random number can be used to mask the second parameter for privacy protection, thereby preventing the second data party from obtaining the specific second parameter.
  • the subsequent step S115 even if the second data party obtains the first share of the second data from the first data party, the first share of the second data is compared with the second share of the second data owned by itself. Plus, what is obtained is the product of the second parameter and the random number, and the specific second parameter cannot be obtained. For examples of related scenarios, refer to the subsequent step S115.
  • the first data party can secretly share one according to the first share of each second parameter and the random number corresponding to the second parameter.
  • the second data The first data party can obtain a first share of the second data, and the second data party can obtain a second share of the second data.
  • the first data party may secretly share a fourth piece of data according to each random number
  • the second data party may secretly share a fourth piece of data according to the second share of the second parameter corresponding to the random number.
  • the first data party and the second data party may respectively obtain a share of the fourth data.
  • the share obtained by the first data party may be taken as the first share of the fourth data
  • the share obtained by the second data party may be taken as the second share of the fourth data.
  • the sum of the first share of the fourth data and the second share of the fourth data is the fourth data.
  • the first data party may multiply the first share of the second parameter by the random number; may add the product result to the first share of the fourth data; may use the addition result as the second data
  • the second data party may directly use the second share of the fourth data as the second share of the second data.
  • the first party data can r i
  • the second party may be a data
  • the first data party can obtain the first share of the fourth data ⁇ [r i ⁇ h ⁇ (x i )> 1 ]> 0
  • the second data party can obtain the second share of the fourth data ⁇ [r i ⁇ h ⁇ (x i )> 1 ]> 1 .
  • the first data party may multiply ⁇ h ⁇ (x i )> 0 by r i ; the product result r i ⁇ h ⁇ (x i )> 0 and the first share of the fourth data ⁇ [r i ⁇ h ⁇ (x i )> 1 ]> 0 to add; the addition result can be used as the first share of the first data r i h ⁇ (x i ) ⁇ r i h ⁇ (x i )> 0 .
  • the second data party may directly use the second share of the fourth data ⁇ [r i ⁇ h ⁇ (x i )> 1 ]> 1 as the second share of the first data r i h ⁇ (x i ) ⁇ r i h ⁇ (x i )> 1 . among them,
  • the first data party may accumulate multiple random numbers to obtain the coefficient of the second term.
  • the first data party can calculate As the second term coefficient.
  • the first data party may send to the second data party the first share of the value of the first item, the first share of the multiple second data, and the share of the second item. coefficient.
  • the second data party may receive the first share of the value of the first item, the first share of multiple second data, and the coefficient of the second item.
  • the second data party may add the first share of the value of the first item and the second share of the value of the first item to obtain the value of the first item;
  • the first share of the second data and the second share of the second data are added to obtain the second data;
  • the value of the second item can be determined according to the second data and the coefficient of the second item.
  • the second data party may add the value of the first term and the value of the second term to obtain the value of the loss function.
  • the second data party may add the first share of each second data and the second share of the second data to obtain second data; multiple second data may be accumulated and multiplied; Multiply the result and the coefficient of the second term to determine the value of the second term.
  • the second data party can take the first share of the value of the first item And the second share of the value of the first item Add to get the value of the first term
  • the second party may be second data of the first share of the second share data ⁇ r i h ⁇ (x i )> 0 and the second data are added, to obtain a second data r i h ⁇ (x i) ; Can be calculated Get the value of the second term. It is worth noting that due to the use of random numbers, even if the second data party adds the first share of the second data and the second share of the second data, the second parameter h ⁇ (x i ) and The product r i h ⁇ (x i ) of the random number r i cannot obtain the specific second parameter h ⁇ (x i ), so that the second parameter h ⁇ (x i ) is concealed.
  • the second data party can take the value of the first item And the value of the second term Add to get the log loss function The value of.
  • the first data party and the second data party can use the secret sharing algorithm to jointly calculate the value of the loss function without leaking the data they own; thus, it is convenient to calculate the value of the loss function according to the value of the loss function. Measure the training effect of the data processing model, and then decide whether to terminate the training.
  • the second data party may send the value of the loss function to a trusted third party (TTP, Trusted Third Party), and the trusted third party decides whether to terminate the training.
  • TTP Trusted Third Party
  • the loss function may include a first term and a second term.
  • the first term and the second term are respectively different function terms in the loss function.
  • the first data party is the execution subject.
  • the first data party may be a data party that does not possess a tag value.
  • the first data party may have complete sample data; or, may have a part of data items of the sample data. Please refer to FIG. 2, this embodiment may include the following steps.
  • Step S21 secretly share the first data with the partner according to the share of the first parameter to obtain the first data share.
  • the cooperating party may be understood as a data party that performs cooperative security modeling with the first data party, and may specifically be the previous second data party.
  • the first data party may secretly share the first data with the partner according to the share of the first parameter to obtain the corresponding share.
  • Step S23 Determine the value of the first item according to the share of the first data.
  • the first data party may accumulate the share of the first data; the share of the first item value may be determined according to the accumulation result.
  • the share of the first item value may be determined according to the accumulation result.
  • Step S25 secretly share the second data with the partner according to the share of the second parameter and the random number to obtain the share of the second data.
  • the second data and the coefficient of the subsequent second term can be combined to determine the value of the second term.
  • the first data party can generate a random number corresponding to the share of the second parameter; can multiply the share of the second parameter by the random number; can secretly share the third data with the partner according to the random number to obtain the corresponding share ; You can add the product result to the share of the third data to get the share of the second data.
  • Step S27 Determine the coefficient of the second term according to the random number.
  • the first data party can accumulate the random number to obtain the coefficient of the second term.
  • the specific process please refer to the related description in the previous step S109, which will not be repeated here.
  • Step S29 Send the share of the first item, the share of the second data, and the coefficient of the second item to the partner, so that the partner can determine the value of the loss function.
  • the first data party can use the secret sharing algorithm to work with the partner to calculate the share of the first item, the share of the second data, and the second data without leaking the data it owns.
  • the coefficient of the item; the share of the value of the first item, the share of the second data, and the coefficient of the second item can be sent to the partner, so that the partner can determine the value of the loss function.
  • this specification also provides an embodiment of another method for determining the value of the loss function.
  • the loss function may include a first term and a second term.
  • the first term and the second term are respectively different function terms in the loss function.
  • This embodiment uses the second data party as the execution subject.
  • the second data party may be a data party possessing a tag value. Specifically, for example, the second data party may only possess the tag value; or, it may also possess part of the data items of the sample data. Referring to FIG. 3, this embodiment may include the following steps.
  • Step S31 Secretly share the first data with the partner according to the share and tag value of the first parameter to obtain the share of the first data.
  • the cooperating party may be understood as a data party that performs cooperative security modeling with the second data party, and may specifically be the previous first data party.
  • the second data party can calculate the share of the first parameter and the tag value according to a preset operation rule; can secretly share the third data with the partner according to the tag value to obtain the corresponding share; can compare the calculation result with the third data Add the share of the first data to get the share of the first data.
  • the specific process please refer to the related description in the previous step S101, which will not be repeated here.
  • Step S33 Determine the first share of the value of the first item according to the share of the first data.
  • the second data party may accumulate the share of the first data; the first share of the value of the first item may be determined according to the accumulation result.
  • the relevant description in the previous step S105 please refer to the relevant description in the previous step S105, which will not be repeated here.
  • Step S35 secretly share the second data with the partner according to the share of the second parameter to obtain the first share of the second data.
  • the combination of the first data and the coefficient of the subsequent second term can determine the value of the second term.
  • the relevant description in the previous step S107 please refer to the relevant description in the previous step S107, which will not be repeated here.
  • Step S37 Receive the second share of the first item value, the second share of the second data, and the coefficient of the second item sent by the partner.
  • Step S39 Determine the loss according to the first share of the first item value, the second share of the first item value, the first share of the second data, the second share of the second data, and the coefficient of the second item The value of the function.
  • the second data party may add the first share of the value of the first item and the second share of the value of the first item to obtain the value of the first item; may combine the first share of the second data with the second share of the first item. Add the second share of the data to get the second data; the value of the second item can be determined according to the coefficients of the second data and the second item; the value of the first item can be added to the value of the second item , Get the value of the loss function.
  • the specific process please refer to the related description in the previous step S115, which will not be repeated here.
  • the second data party can use the secret sharing algorithm to work with the partner to calculate the first share of the value of the first item and the first share of the second data without leaking the data it owns. Share. In this way, the second data party can determine the value of the loss function by combining the second share of the first item value from the partner, the second share of the second data, and the coefficient of the second item.
  • this specification also provides an embodiment of a device for determining the value of a loss function.
  • the loss function may include a first term and a second term.
  • the first term and the second term are respectively different function terms in the loss function.
  • This embodiment can be applied to the first data party.
  • the first data party may be a data party that does not possess a tag value. Specifically, for example, the first data party may have complete sample data; or, may have a part of data items of the sample data. Refer to FIG. 4, this embodiment may include the following units.
  • the first secret sharing unit 41 is configured to secretly share the first data with the partner according to the share of the first parameter to obtain the share of the first data;
  • the first determining unit 43 is configured to determine the share of the first item value according to the share of the first data
  • the second secret sharing unit 45 is configured to secretly share the second data with the partner according to the share of the second parameter and the random number to obtain the share of the second data;
  • the second determining unit 47 is configured to determine the coefficient of the second term according to the random number, and the coefficient of the second term and the first data are used to jointly determine the value of the second term;
  • the sending unit 49 is configured to send the share of the first item, the share of the second data, and the coefficient of the second item to the partner, so that the partner can determine the value of the loss function.
  • the loss function may include a first term and a second term.
  • the first term and the second term are respectively different function terms in the loss function.
  • This embodiment can be applied to a second data party.
  • the second data party may be a data party possessing a tag value. Specifically, for example, the second data party may only possess the tag value; or, it may also possess part of the data items of the sample data. Referring to FIG. 5, this embodiment may include the following units.
  • the first secret sharing unit 51 is configured to secretly share the first data with the partner according to the share and tag value of the first parameter to obtain the first share of the first data;
  • the first determining unit 53 is configured to determine the first share of the value of the first item according to the share of the first data
  • the second secret sharing unit 55 is configured to secretly share the second data with the partner according to the share of the second parameter to obtain the first share of the second data;
  • the receiving unit 57 is configured to receive the second share of the first item value, the second share of the second data, and the coefficient of the second item sent by the partner.
  • the coefficient of the second item and the second data are used to jointly determine the The value of the binomial;
  • the second determining unit 59 is used to determine the first share of the first item, the second share of the first item, the first share of the second data, the second share of the second data, and the coefficient of the second item , Determine the value of the loss function.
  • Fig. 6 is a schematic diagram of the hardware structure of an electronic device in this embodiment.
  • the electronic device may include one or more (only one is shown in the figure) processor, memory, and transmission module.
  • the hardware structure shown in FIG. 6 is only for illustration, and it does not limit the hardware structure of the above electronic device.
  • the electronic device may also include more or less component units than shown in FIG. 6; or, have a configuration different from that shown in FIG. 6.
  • the memory may include a high-speed random access memory; or, it may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory.
  • the storage may also include a remotely set network storage.
  • the remotely set network storage can be connected to the electronic device through a network such as the Internet, an intranet, a local area network, a mobile communication network, and the like.
  • the memory may be used to store program instructions or modules of application software, such as the program instructions or modules of the embodiment corresponding to FIG. 2 of this specification; and/or, the program instructions or modules of the embodiment corresponding to FIG. 3 of this specification.
  • the processor can be implemented in any suitable way.
  • the processor may take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program codes (for example, software or firmware) executable by the (micro)processor, logic gates, switches, special-purpose integrated Circuit (Application Specific Integrated Circuit, ASIC), programmable logic controller and embedded microcontroller form, etc.
  • the processor can read and execute program instructions or modules in the memory.
  • the transmission module can be used for data transmission via a network, for example, data transmission via a network such as the Internet, an intranet, a local area network, a mobile communication network, and the like.
  • a network such as the Internet, an intranet, a local area network, a mobile communication network, and the like.
  • a programmable logic device Programmable Logic Device, PLD
  • FPGA Field Programmable Gate Array
  • HDL Hardware Description Language
  • ABEL Advanced Boolean Expression Language
  • AHDL Altera Hardware Description Language
  • HDCal JHDL
  • Lava Lava
  • Lola MyHDL
  • PALASM RHDL
  • Verilog2 Verilog2
  • a typical implementation device is a computer.
  • the computer may be, for example, a personal computer, a laptop computer, a cell phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or Any combination of these devices.
  • This manual can be used in many general or special computer system environments or configurations.
  • program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types.
  • This specification can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network.
  • program modules can be located in local and remote computer storage media including storage devices.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Computational Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Algebra (AREA)
  • Databases & Information Systems (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Storage Device Security (AREA)

Abstract

一种损失函数取值的确定方法、装置和电子设备。所述方法用于确定损失函数的取值。所述损失函数包括第一项和第二项。所述方法包括:根据第一参数的份额与合作方秘密分享第一数据,得到第一数据的份额(S21);根据第一数据的份额,确定第一项取值的份额(S23);根据第二参数的份额和随机数与合作方秘密分享第二数据,得到第二数据的份额(S25);根据随机数确定第二项的系数(S27),第二项的系数和第二数据用于共同确定第二项的取值;向合作方发送第一项取值的份额、第二数据的份额和第二项的系数,以便合作方确定所述损失函数的取值(S29)。

Description

损失函数取值的确定方法、装置和电子设备 技术领域
本说明书实施例涉及计算机技术领域,特别涉及一种损失函数取值的确定方法、装置和电子设备。
背景技术
大数据时代,存在非常多的数据孤岛。数据通常分散存于不同的企业中,企业与企业之间由于竞争关系和隐私保护的考虑,并不是完全的互相信任。在一些情况下,企业与企业之间需要进行合作安全建模,以便在充分保护企业数据隐私的前提下,利用各方的数据对数据处理模型进行协作训练。
在合作安全建模的场景中,需要计算数据处理模型的损失函数的值;通过损失函数的值可以衡量数据处理模型的训练效果(例如过拟合、欠拟合等),进而决定是否终止训练。由于用于对数据处理模型进行训练的数据是分散在合作建模的各方的,在相关技术中,通常是将合作建模各方的数据汇总在独立的第三方,由该独立的第三方来计算损失函数的值。由于将合作建模各方的数据进行了汇总,这样容易造成企业数据的泄漏。
发明内容
本说明书实施例的目的是提供一种损失函数取值的确定方法、装置和电子设备,以在保护数据隐私的前提下,由建模的各数据方协作计算出损失函数的值。
为实现上述目的,本说明书中一个或多个实施例提供的技术方案如下。
根据本说明书一个或多个实施例的第一方面,提供了一种损失函数取值的确定方法,所述损失函数包括第一项和第二项;所述方法包括:根据第一参数的份额与合作方秘密分享第一数据,得到第一数据的份额;根据第一数据的份额,确定第一项取值的份额;根据第二参数的份额和随机数与合作方秘密分享第二数据,得到第二数据的份额;根据随机数确定第二项的系数,第二项的系数和第二数据用于共同确定第二项的取值;向合作方发送第一项取值的份额、第二数据的份额和第二项的系数,以便合作方确定所述损失函数的取值。
根据本说明书一个或多个实施例的第二方面,提供了一种损失函数取值的确定装置, 所述损失函数包括第一项和第二项;所述装置包括:第一秘密分享单元,用于根据第一参数的份额与合作方秘密分享第一数据,得到第一数据的份额;第一确定单元,用于根据第一数据的份额,确定第一项取值的份额;第二秘密分享单元,用于根据第二参数的份额和随机数与合作方秘密分享第二数据,得到第二数据的份额;确定单元,用于根据随机数确定第二项的系数,第二项的系数和第二数据用于共同确定第二项的取值;发送单元,用于向合作方发送第一项取值的份额、第二数据的份额和第二项的系数,以便合作方确定所述损失函数的取值。
根据本说明书一个或多个实施例的第三方面,提供了一种电子设备,包括:存储器,用于存储计算机指令;处理器,用于执行所述计算机指令以实现如第一方面所述的方法步骤。
根据本说明书一个或多个实施例的第四方面,提供了一种损失函数取值的确定方法,所述损失函数包括第一项和第二项;所述方法包括:根据第一参数的份额和标签值与合作方秘密分享第一数据,得到第一数据的份额;根据第一数据的份额,确定第一项取值的第一份额;根据第二参数的份额与合作方秘密分享第二数据,得到第二数据的第一份额;接收合作方发来的第一项取值的第二份额、第二数据的第二份额和第二项的系数,第二项的系数和第二数据用于共同确定第二项的取值;根据第一项取值的第一份额、第一项取值的第二份额、第二数据的第一份额、第二数据的第二份额和第二项的系数,确定所述损失函数的值。
根据本说明书一个或多个实施例的第五方面,提供了一种损失函数取值的确定装置,所述损失函数包括第一项和第二项;所述装置包括:第一秘密分享单元,用于根据第一参数的份额和标签值与合作方秘密分享第一数据,得到第一数据的份额;第一确定单元,用于根据第一数据的份额,确定第一项取值的第一份额;第二秘密分享单元,用于根据第二参数的份额与合作方秘密分享第二数据,得到第二数据的第一份额;接收单元,用于接收合作方发来的第一项取值的第二份额、第二数据的第二份额和第二项的系数,第二项的系数和第二数据用于共同确定第二项的取值;确定单元,用于根据第一项取值的第一份额、第一项取值的第二份额、第二数据的第一份额、第二数据的第二份额和第二项的系数,确定所述损失函数的值。
根据本说明书一个或多个实施例的第六方面,提供了一种电子设备,包括:存储器,用于存储计算机指令;处理器,用于执行所述计算机指令以实现如第四方面所述的方法步骤。
由以上本说明书实施例提供的技术方案可见,本说明书实施例中,第一数据方和第二数据方可以利用秘密分享算法,在不泄漏自身所拥有的数据的前提下,协作计算出损失函数的值;从而便于根据损失函数的值,来衡量数据处理模型的训练效果,进而决定是否终止训练。
附图说明
为了更清楚地说明本说明书实施例或现有技术中的技术方案,下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍,显而易见地,下面描述中的附图仅仅是本说明书中记载的一些实施例,对于本领域普通技术人员来讲,在不付出创造性劳动性的前提下,还可以根据这些附图获得其他的附图。
图1为本说明书实施例一种损失函数取值的确定方法的流程图;
图2为本说明书实施例一种损失函数取值的确定方法的流程图;
图3为本说明书实施例一种损失函数取值的确定方法的流程图;
图4为本说明书实施例一种损失函数取值的确定装置的功能结构图;
图5为本说明书实施例一种损失函数取值的确定装置的功能结构图;
图6为本说明书实施例一种电子设备的功能结构图。
具体实施方式
下面将结合本说明书实施例中的附图,对本说明书实施例中的技术方案进行清楚、完整地描述,显然,所描述的实施例仅仅是本说明书一部分实施例,而不是全部的实施例。基于本说明书中的实施例,本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例,都应当属于本说明书保护的范围。应当理解,尽管在本说明书可能采用术语第一、第二、第三等来描述各种信息,但这些信息不应限于这些术语。这些术语仅用来将同一类型的信息彼此区分开。例如,在不脱离本说明书范围的情况下,第一信息也可以被称为第二信息,类似地,第二信息也可以被称为第一信息。
秘密分享(SS,Secret Sharing)是一种保护数据隐私安全的算法。多个数据方可以在不泄漏自身数据的前提下,使用秘密分享算法进行协作计算,得到秘密信息。每个数据方可以获得该秘密信息的一份份额。单个数据方无法恢复该秘密信息。只有多个数据方一同协作才能恢复该秘密信息。例如数据方P 1拥有数据x 1,数据方P 2拥有数据x 2。 采用秘密分享算法,数据方P 1和数据方P 2可以进行协作计算,得到秘密信息y=y 1+y 2=x 1x 2。数据方P 1在计算后可以获得秘密信息y的份额y 1,数据方P 2在计算后可以获得秘密信息y的份额y 2
损失函数(Loss Function)可以用于衡量数据处理模型的预测值与真实值之间不一致的程度。损失函数的值越小,表示数据处理模型的鲁棒性越好。在对数据处理模型进行训练的过程中,可以通过计算损失函数的值,来衡量数据处理模型的训练效果(例如过拟合、欠拟合等),进而决定是否终止训练。所述数据处理模型包括但不限于逻辑回归模型、线性回归模型、神经网络模型等。不同的数据处理模型可以采用不同的损失函数来衡量。例如,逻辑回归模型可以采用对数损失函数(Logarithmic Loss Function)来衡量,线性回归模型可以采用平方损失函数(Square Loss)来衡量。
在合作安全建模的场景中,出于保护数据隐私的考虑,多个数据方可以在不泄漏自身所拥有的数据的前提下,基于自身拥有的数据,对数据处理模型进行协作训练。在一些场景示例中,进行合作安全建模的数据方的数量为两个。其中一个数据方可以拥有完整的样本数据,另一个数据方可以拥有样本数据的标签值;或者,其中一个数据方可以拥有样本数据中的一部分数据项,另一个数据方可以拥有样本数据的另一部分数据项和样本数据的标签值。具体地,例如,样本数据包括用户的储蓄金额和借贷金额。其中一个数据方可以拥有用户的储蓄金额,另一个数据方可以拥有用户的借贷金额和样本数据的标签值。
所述多个数据方需要协作计算损失函数的值,以决定是否终止训练。考虑到用于对数据处理模型进行训练的数据分散在合作建模的各数据方,若采用秘密分享算法,则合作建模的各数据方可以在不泄漏自身所拥有的数据的前提下,基于自身拥有的数据,协作计算出损失函数的值。
本说明书提供一种损失函数取值的确定方法的实施例。
请参阅图1,该实施例可以包括以下步骤。
步骤S101:第一数据方根据第一参数的第一份额,第二数据方根据第一参数的第二份额和标签值,秘密分享第一数据。第一数据方获得第一数据的第一份额,第二数据方获得第一数据的第二份额。
步骤S103:第一数据方根据第一数据的第一份额,确定第一项取值的第一份额。
步骤S105:第二数据方根据第一数据的第二份额,确定第一项取值的第二份额。
步骤S107:第一数据方根据第二参数的第一份额和随机数,第二数据方根据第二参数的第二份额,秘密分享第二数据。第一数据方获得第二数据的第一份额,第二数据方获得第二数据的第二份额。
步骤S109:第一数据方根据随机数确定第二项的系数。
步骤S111:第一数据方向第二数据方发送第一项取值的第一份额、第二数据的第一份额和第二项的系数。
步骤S113:第二数据方接收第一项取值的第一份额、第二数据的第一份额和第二项的系数。
步骤S115:第二数据方根据第一项取值的第一份额、第一项取值的第二份额、第二数据的第一份额、第二数据的第二份额和第二项的系数,确定所述损失函数的值。
下面介绍在实施例中涉及的一些术语。
(一)、第一项和第二项。所述第一项和所述第二项分别为所述损失函数中的函数项。
在一些场景示例中,所述损失函数可以为对数损失函数
Figure PCTCN2020070939-appb-000001
m表示样本数据的数量;x i表示第i个样本数据;y i表示样本数据x i的标签值;θ表示数据处理模型的模型参数;h θ(x i)表示数据处理模型的激励函数的取值,
Figure PCTCN2020070939-appb-000002
Figure PCTCN2020070939-appb-000003
那么,所述第一项可以为
Figure PCTCN2020070939-appb-000004
所述第二项可以为
Figure PCTCN2020070939-appb-000005
(二)、第一数据方和第二数据方。所述第一数据方和所述第二数据方分别为合作安全建模的双方。所述第一数据方可以为不拥有标签值的数据方,所述第二数据方可以为拥有标签值的数据方。例如,所述第一数据方可以拥有完整的样本数据,所述第二数据方可以拥有样本数据的标签值。或者,所述第一数据方可以拥有样本数据的一部分数据项,所述第二数据方可以拥有样本数据的另一部分数据项和标签值。所述标签值可以用于区分不同类型的样本数据,具体数值例如可以取自0和1。其中,所述数据方可以为电子设备。所述电子设备可以包括个人计算机、服务器、手持设备、便携式设备、平板型设备、多处理器装置;或者,还可以包括由以上任何多个装置或设备所构成的集群等。
(三)、第一参数和第二参数。所述第一参数和所述第二参数分别为所述第一数据方和所述第二数据方在合作安全建模过程中获得的中间结果。所述第一参数和所述第二参数不同。例如,所述第一参数可以为样本数据与数据处理模型的模型参数之间的乘积,所述第二参数可以为数据处理模型的激励函数的值。
在合作安全建模的过程中,所述第一数据方和所述第二数据方分别获得所述第一参数的一份份额。为了便于描述,可以将所述第一数据方获得的份额作为第一参数的第一份额,可以将所述第二数据方获得的份额作为第一参数的第二份额。第一参数的第一份额和第一参数的第二份额的和即为第一参数。此外,所述第一参数的数量可以为多个。如此所述第一数据方可以拥有多个第一参数的第一份额,所述第二数据方可以拥有多个第一参数的第二份额。
延续前面的场景示例,第一参数可以表示为θx i,第一参数的第一份额可以表示为<θx i0,第一参数的第二份额可以表示为<θx i1。其中<θx i0+<θx i1=θx i
在合作安全建模的过程中,所述第一数据方和所述第二数据方分别获得所述第二参数的一份份额。为了便于描述,可以将所述第一数据方获得的份额作为第二参数的第一份额,可以将所述第二数据方获得的份额作为第二参数的第二份额。第二参数的第一份额和第二参数的第二份额的和即为第二参数。此外,所述第二参数的数量可以为多个。如此所述第一数据方可以拥有多个第二参数的第一份额,所述第二数据方可以拥有多个第二参数的第二份额。
延续前面的场景示例,第二参数可以表示为
Figure PCTCN2020070939-appb-000006
第二参数的第一份额可以表示为<h θ(x i)> 0,第二参数的第二份额可以表示为<h θ(x i)> 1。其中<h θ(x i)> 0+<h θ(x i)> 1=h θ(x i)。
(四)、第一数据和第二数据。所述第一数据和所述第二数据分别由所述第一数据方和所述第二数据方通过秘密分享算法计算得到。所述第一数据用于确定第一项的取值。所述第二数据与所述第二项的系数相结合能够确定出第二项的取值。
延续前面的场景示例,第一数据可以表示为(1-y i)θx i,第二数据可以表示为r ih θ(x i),第二项的系数可以表示为
Figure PCTCN2020070939-appb-000007
(五)、第三数据和第四数据。所述第三数据和所述第四数据分别由所述第一数据方和所述第二数据方通过秘密分享算法计算得到。根据第三数据的份额能够确定出第一数据的份额。根据第四数据的份额能够确定出第二数据的份额。
延续前面的场景示例,第三数据可以表示为(1-y i)<θx i0,第四数据可以表示为r i<h θ(x i)> 1
在一些实施例中,在步骤S101中,所述第一数据方可以根据第一参数的第一份额,所述第二数据方可以根据第一参数的第二份额和标签值,秘密分享第一数据。所述第一数据方和所述第二数据方可以分别获得第一数据的一份份额。为了便于描述,可以将所述第一数据方获得的份额作为第一数据的第一份额,可以将所述第二数据方获得的份额作为第一数据的第二份额。第一数据的第一份额和第一数据的第二份额的和即为第一数据。
如前面所述,所述第一数据方可以拥有多个第一参数的第一份额,所述第二数据方可以拥有多个第一参数的第二份额。如此所述第一数据方可以根据每个第一参数的第一份额,所述第二数据方可以根据该第一参数的第二份额和与该第一参数相对应的标签值,秘密分享一个第一数据。所述第一数据方可以获得该第一数据的第一份额,所述第二数据方可以获得该第一数据的第二份额。值得说明的是,第一参数与标签值相对应可以理解为:标签值所对应的样本数据与用于计算第一参数的样本数据为同一样本数据。
进一步地,所述第一数据方可以根据每个第一参数的第一份额,所述第二数据方可以根据与该第一参数相对应的标签值,秘密分享一个第三数据。所述第一数据方和所述 第二数据方可以分别获得该第三数据的一份份额。为了便于描述,可以将所述第一数据方获得的份额作为该第三数据的第一份额,可以将所述第二数据方获得的份额作为该第三数据的第二份额。该第三数据的第一份额和该第三数据的第二份额的和即为该第三数据。如此所述第一数据方可以直接将该第三数据的第一份额作为第一数据的第一份额。所述第二数据方可以将该第一参数的第二份额与与该第一参数相对应的标签值按照预设运算规则进行运算;可以将运算结果与该第三数据的第二份额相加,可以将相加结果作为第一数据的第二份额。
延续前面的场景示例,所述第一数据方可以根据<θx i0,所述第二数据方可以根据1-y i,秘密分享第三数据(1-y i)<θx i0。所述第一数据方可以获得第三数据的第一份额<[(1-y i)<θx i0]> 0。所述第二数据方可以获得第三数据的第二份额<[(1-y i)<θx i0]> 1。其中,<[(1-y i)<θx i0]> 0+<[(1-y i)<θx i0]> 1=(1-y i)<θx i0
所述第一数据方可以直接将第三数据的第一份额<[(1-y i)<θx i0]> 0作为第一数据(1-y i)θx i的第一份额<(1-y i)θx i0。所述第二数据方可以根据<θx i1和y i计算得到(1-y i)<θx i1;可以将计算结果(1-y i)<θx i1与第三数据的第二份额<[(1-y i)<θx i0]> 1相加;可以将相加结果作为第一数据(1-y i)θx i的第二份额<(1-y i)θx i1。其中,
Figure PCTCN2020070939-appb-000008
在一些实施例中,经过步骤S101,所述第一数据方可以获得多个第一数据的第一份额。如此在步骤S103中,所述第一数据方可以对多个第一数据的第一份额进行累加;可以根据累加结果确定第一项取值的第一份额。具体地,所述第一数据方可以将累加结果与第一数据的数量(也即第一参数的数量)相除,得到第一项取值的第一份额。
延续前面的场景示例,所述第一数据方可以计算
Figure PCTCN2020070939-appb-000009
作为第一项取值的第一份额。
在一些实施例中,经过步骤S101,所述第二数据方可以获得多个第一数据的第二份额。如此在步骤S105中,所述第二数据方可以对多个第一数据的第二份额进行累加;可以根据累加结果确定第一项取值的第二份额。具体地,所述第一数据方可以将累加结果与第一数据的数量(也即第一参数的数量)相除,得到第一项取值的第二份额。第一 项取值的第一份额和第一项取值的第二份额的和即为第一项的取值。
延续前面的场景示例,所述第二数据方可以计算
Figure PCTCN2020070939-appb-000010
作为第一项取值的第二份额。其中,
Figure PCTCN2020070939-appb-000011
在一些实施例中,在步骤S107中,所述第一数据方可以根据第二参数的第一份额和随机数,所述第二数据方可以根据第二参数的第二份额,秘密分享第二数据。所述第一数据方和所述第二数据方可以分别获得第二数据的一份份额。为了便于描述,可以将所述第一数据方获得的份额作为第二数据的第一份额,可以将所述第二数据方获得的份额作为第二数据的第二份额。第二数据的第一份额和第二数据的第二份额的和即为第二数据。
如前面所述,第二参数的数量可以为多个。所述第一数据方可以生成多个随机数,每个随机数可以对应一个第二参数(或第二参数的第一份额)。随机数可以用于掩盖第二参数,以进行隐私保护,从而防止第二数据方获得具体的第二参数。这样在后续的步骤S115中,即使第二数据方获得了来自第一数据方的第二数据的第一份额,进而将第二数据的第一份额与自身拥有的第二数据的第二份额相加,获得的也是第二参数与随机数的乘积,而无法获得具体的第二参数。相关场景示例可以参见后续的步骤S115。如此所述第一数据方可以根据每个第二参数的第一份额和与该第二参数相对应的随机数,所述第二数据方可以根据该第二参数的第二份额,秘密分享一个第二数据。所述第一数据方可以获得该第二数据的第一份额,所述第二数据方可以获得该第二数据的第二份额。
进一步地,所述第一数据方可以根据每个随机数,所述第二数据方可以根据与该随机数相对应的第二参数的第二份额,秘密分享一个第四数据。所述第一数据方和所述第二数据方可以分别获得该第四数据的一份份额。为了便于描述,可以将所述第一数据方获得的份额作为该第四数据的第一份额,可以将所述第二数据方获得的份额作为该第四数据的第二份额。该第四数据的第一份额和该第四数据的第二份额的和即为该第四数据。如此所述第一数据方可以将该第二参数的第一份额与该随机数相乘;可以将乘积结果与该第四数据的第一份额相加;可以将相加结果作为第二数据的第一份额。所述第二数据方可以直接将该第四数据的第二份额作为第二数据的第二份额。
延续前面的场景示例,所述第一数据方可以根据r i,所述第二数据方可以根据<h θ(x i)> 1,秘密分享第四数据r i<h θ(x i)> 1。所述第一数据方可以获得第四数据的第一 份额<[r i<h θ(x i)> 1]> 0,所述第二数据方可以获得第四数据的第二份额<[r i<h θ(x i)> 1]> 1。其中,<[r i<h θ(x i)> 1]> 0+<[r i<h θ(x i)> 1]> 1=r i<h θ(x i)> 1
所述第一数据方可以将<h θ(x i)> 0与r i相乘;可以将乘积结果r i<h θ(x i)> 0与第四数据的第一份额<[r i<h θ(x i)> 1]> 0相加;可以将相加结果作为第一数据r ih θ(x i)的第一份额<r ih θ(x i)> 0。所述第二数据方可以直接将第四数据的第二份额<[r i<h θ(x i)> 1]> 1作为第一数据r ih θ(x i)的第二份额<r ih θ(x i)> 1。其中,
<r ih θ(x i)> 0+<r ih θ(x i)> 1
=r i<h θ(x i)> 0+<[r i<h θ(x i)> 1]> 0+<[r i<h θ(x i)> 1]> 1
=r ih θ(x i)
在一些实施例中,在步骤S109中,所述第一数据方可以将多个随机数进行累乘,得到第二项的系数。延续前面的场景示例,所述第一数据方可以计算
Figure PCTCN2020070939-appb-000012
作为第二项的系数。
在一些实施例中,在步骤S111中,所述第一数据方可以向所述第二数据方发送第一项取值的第一份额、多个第二数据的第一份额和第二项的系数。在步骤S113中,所述第二数据方可以接收第一项取值的第一份额、多个第二数据的第一份额和第二项的系数。
在一些实施例中,在步骤S115中,所述第二数据方可以将第一项取值的第一份额和第一项取值的第二份额相加,得到第一项的取值;可以将第二数据的第一份额和第二数据的第二份额相加,得到第二数据;可以根据第二数据和第二项的系数,确定第二项的取值。所述第二数据方可以将第一项的取值和第二项的取值相加,得到所述损失函数的值。其中,所述第二数据方可以将每个第二数据的第一份额和该第二数据的第二份额相加,得到第二数据;可以将多个第二数据进行累乘;可以根据累乘结果与第二项的系数,确定第二项的取值。
延续前面的场景示例,所述第二数据方可以将第一项取值的第一份额
Figure PCTCN2020070939-appb-000013
和第一项取值的第二份额
Figure PCTCN2020070939-appb-000014
相加,得到第一项的取值
Figure PCTCN2020070939-appb-000015
所述第二数据方可以将第二数据的第一份额<r ih θ(x i)> 0和第二数据的第二份额相加, 得到第二数据r ih θ(x i);可以计算
Figure PCTCN2020070939-appb-000016
得到第二项的取值。值得说明的是,由于使用了随机数,所述第二数据方即使将第二数据的第一份额和第二数据的第二份额相加,获得的也是第二参数h θ(x i)与随机数r i的乘积r ih θ(x i),而无法获得具体的第二参数h θ(x i),从而实现了对第二参数h θ(x i)的掩盖。
所述第二数据方可以将第一项的取值
Figure PCTCN2020070939-appb-000017
和第二项的取值
Figure PCTCN2020070939-appb-000018
相加,得到对数损失函数
Figure PCTCN2020070939-appb-000019
的取值。
在本实施例中,第一数据方和第二数据方可以利用秘密分享算法,在不泄漏自身所拥有的数据的前提下,协作计算出损失函数的值;从而便于根据损失函数的值,来衡量数据处理模型的训练效果,进而决定是否终止训练。例如,第二数据方可以将损失函数的值发送给可信任的第三方(TTP,Trusted Third Party),由可信任的第三方来决定是否终止训练。
基于同样的发明构思,本说明书还提供另一种损失函数取值的确定方法的实施例。所述损失函数可以包括第一项和第二项。所述第一项和所述第二项分别为所述损失函数中不同的函数项。该实施例以第一数据方为执行主体。所述第一数据方可以为不拥有标签值的数据方。具体地,例如,所述第一数据方可以拥有完整的样本数据;或者,可以拥有样本数据的一部分数据项。请参阅图2,该实施例可以包括以下步骤。
步骤S21:根据第一参数的份额与合作方秘密分享第一数据,得到第一数据份额。
所述合作方可以理解为与所述第一数据方进行合作安全建模的数据方,具体可以为前面的第二数据方。所述第一数据方可以根据第一参数的份额与合作方秘密分享第一数据,得到相应的份额。具体过程可以参见前面步骤S101中的相关描述,在此不再赘述。
步骤S23:根据第一数据的份额,确定第一项取值的份额。
所述第一数据方可以对第一数据的份额进行累加;可以根据累加结果确定第一项取值的份额。具体过程可以参见前面步骤S103中的相关描述,在此不再赘述。
步骤S25:根据第二参数的份额和随机数与合作方秘密分享第二数据,得到第二数据的份额。
第二数据与后续第二项的系数相结合能够确定出第二项的取值。所述第一数据方可以生成与第二参数的份额相对应的随机数;可以将第二参数的份额与随机数相乘;可以根据随机数与合作方秘密分享第三数据,得到相应的份额;可以将乘积结果与第三数据的份额相加,得到第二数据的份额。具体过程可以参见前面步骤S107中的相关描述,在此不再赘述。
步骤S27:根据随机数确定第二项的系数。
所述第一数据方可以对随机数进行累乘,得到第二项的系数。具体过程可以参见前面步骤S109中的相关描述,在此不再赘述。
步骤S29:向合作方发送第一项取值的份额、第二数据的份额和第二项的系数,以便合作方确定所述损失函数的取值。
在本实施例中,第一数据方可以利用秘密分享算法,在不泄漏自身所拥有的数据的前提下,与合作方协作计算出第一项取值的份额、第二数据的份额和第二项的系数;可以向合作方发送第一项取值的份额、第二数据的份额和第二项的系数,以便由合作方确定出所述损失函数的取值。
基于同样的发明构思,本说明书还提供另一种损失函数取值的确定方法的实施例。所述损失函数可以包括第一项和第二项。所述第一项和所述第二项分别为所述损失函数中不同的函数项。该实施例以第二数据方为执行主体。所述第二数据方可以为拥有标签值的数据方。具体地,例如,所述第二数据方可以仅拥有标签值;或者,还可以拥有样本数据的一部分数据项。请参阅图3,该实施例可以包括以下步骤。
步骤S31:根据第一参数的份额和标签值与合作方秘密分享第一数据,得到第一数据的份额。
所述合作方可以理解为与所述第二数据方进行合作安全建模的数据方,具体可以为前面的第一数据方。所述第二数据方可以将第一参数的份额与标签值按照预设运算规则进行运算;可以根据标签值与合作方秘密分享第三数据,得到相应的份额;可以将运算结果与第三数据的份额相加,得到第一数据的份额。具体过程可以参见前面步骤S101中的相关描述,在此不再赘述。
步骤S33:根据第一数据的份额,确定第一项取值的第一份额。
所述第二数据方可以对第一数据的份额进行累加;可以根据累加结果确定第一项取值的第一份额。具体过程可以参见前面步骤S105中的相关描述,在此不再赘述。
步骤S35:根据第二参数的份额与合作方秘密分享第二数据,得到第二数据的第一份额。
第一数据与后续第二项的系数相结合能够确定出第二项的取值。具体过程可以参见前面步骤S107中的相关描述,在此不再赘述。
步骤S37:接收合作方发来的第一项取值的第二份额、第二数据的第二份额和第二项的系数。
步骤S39:根据第一项取值的第一份额、第一项取值的第二份额、第二数据的第一份额、第二数据的第二份额和第二项的系数,确定所述损失函数的值。
所述第二数据方可以将第一项取值的第一份额和第一项取值的第二份额相加,得到第一项的取值;可以将第二数据的第一份额和第二数据的第二份额相加,得到第二数据;可以根据第二数据和第二项的系数,确定第二项的取值;可以将第一项的取值和第二项的取值相加,得到所述损失函数的值。具体过程可以参见前面步骤S115中的相关描述,在此不再赘述。
在本实施例中,第二数据方可以利用秘密分享算法,在不泄漏自身所拥有的数据的前提下,与合作方协作计算出第一项取值的第一份额、第二数据的第一份额。这样所述第二数据方结合来自合作方的第一项取值的第二份额、第二数据的第二份额和第二项的系数,便可以确定出损失函数的取值。
基于同样的发明构思,本说明书还提供一种损失函数取值的确定装置的实施例。所述损失函数可以包括第一项和第二项。所述第一项和所述第二项分别为所述损失函数中不同的函数项。该实施例可以应用于第一数据方。所述第一数据方可以为不拥有标签值的数据方。具体地,例如,所述第一数据方可以拥有完整的样本数据;或者,可以拥有样本数据的一部分数据项。请参阅图4,该实施例可以包括以下单元。
第一秘密分享单元41,用于根据第一参数的份额与合作方秘密分享第一数据,得到第一数据的份额;
第一确定单元43,用于根据第一数据的份额,确定第一项取值的份额;
第二秘密分享单元45,用于根据第二参数的份额和随机数与合作方秘密分享第二数据,得到第二数据的份额;
第二确定单元47,用于根据随机数确定第二项的系数,第二项的系数和第一数据用 于共同确定第二项的取值;
发送单元49,用于向合作方发送第一项取值的份额、第二数据的份额和第二项的系数,以便合作方确定所述损失函数的取值。
基于同样的发明构思,本说明书还提供另一种损失函数取值的确定装置的实施例。所述损失函数可以包括第一项和第二项。所述第一项和所述第二项分别为所述损失函数中不同的函数项。该实施例可以应用于第二数据方。所述第二数据方可以为拥有标签值的数据方。具体地,例如,所述第二数据方可以仅拥有标签值;或者,还可以拥有样本数据的一部分数据项。请参阅图5,该实施例可以包括以下单元。
第一秘密分享单元51,用于根据第一参数的份额和标签值与合作方秘密分享第一数据,得到第一数据的第一份额;
第一确定单元53,用于根据第一数据的份额,确定第一项取值的第一份额;
第二秘密分享单元55,用于根据第二参数的份额与合作方秘密分享第二数据,得到第二数据的第一份额;
接收单元57,用于接收合作方发来的第一项取值的第二份额、第二数据的第二份额和第二项的系数,第二项的系数和第二数据用于共同确定第二项的取值;
第二确定单元59,用于根据第一项取值的第一份额、第一项取值的第二份额、第二数据的第一份额、第二数据的第二份额和第二项的系数,确定所述损失函数的值。
下面介绍本说明书电子设备的一个实施例。图6是该实施例中一种电子设备的硬件结构示意图。如图6所示,所述电子设备可以包括一个或多个(图中仅示出一个)处理器、存储器和传输模块。当然,本领域普通技术人员可以理解,图6所示的硬件结构仅为示意,其并不对上述电子设备的硬件结构造成限定。在实际中所述电子设备还可以包括比图6所示更多或者更少的组件单元;或者,具有与图6所示不同的配置。
所述存储器可以包括高速随机存储器;或者,还可以包括非易失性存储器,例如一个或者多个磁性存储装置、闪存、或者其他非易失性固态存储器。当然,所述存储器还可以包括远程设置的网络存储器。所述远程设置的网络存储器可以通过诸如互联网、企业内部网、局域网、移动通信网等网络连接至所述电子设备。所述存储器可以用于存储应用软件的程序指令或模块,例如本说明书图2所对应实施例的程序指令或模块;和/或,本说明书图3所对应实施例的程序指令或模块。
所述处理器可以按任何适当的方式实现。例如,所述处理器可以采取例如微处理器或处理器以及存储可由该(微)处理器执行的计算机可读程序代码(例如软件或固件)的计算机可读介质、逻辑门、开关、专用集成电路(Application Specific Integrated Circuit,ASIC)、可编程逻辑控制器和嵌入微控制器的形式等等。所述处理器可以读取并执行所述存储器中的程序指令或模块。
所述传输模块可以用于经由网络进行数据传输,例如经由诸如互联网、企业内部网、局域网、移动通信网等网络进行数据传输。
需要说明的是,本说明书中的各个实施例均采用递进的方式描述,各个实施例之间相同或相似的部分互相参见即可,每个实施例重点说明的都是与其它实施例的不同之处。尤其,对于装置实施例和电子设备实施例而言,由于其基本相似于方法实施例,所以描述的比较简单,相关之处参见方法实施例的部分说明即可。另外,可以理解的是,本领域技术人员在阅读本说明书文件之后,可以无需创造性劳动想到将本说明书列举的部分或全部实施例进行任意组合,这些组合也在本说明书公开和保护的范围内。
在20世纪90年代,对于一个技术的改进可以很明显地区分是硬件上的改进(例如,对二极管、晶体管、开关等电路结构的改进)还是软件上的改进(对于方法流程的改进)。然而,随着技术的发展,当今的很多方法流程的改进已经可以视为硬件电路结构的直接改进。设计人员几乎都通过将改进的方法流程编程到硬件电路中来得到相应的硬件电路结构。因此,不能说一个方法流程的改进就不能用硬件实体模块来实现。例如,可编程逻辑器件(Programmable Logic Device,PLD)(例如现场可编程门阵列(Field Programmable Gate Array,FPGA))就是这样一种集成电路,其逻辑功能由用户对器件编程来确定。由设计人员自行编程来把一个数字系统“集成”在一片PLD上,而不需要请芯片制造厂商来设计和制作专用的集成电路芯片2。而且,如今,取代手工地制作集成电路芯片,这种编程也多半改用“逻辑编译器(logic compiler)”软件来实现,它与程序开发撰写时所用的软件编译器相类似,而要编译之前的原始代码也得用特定的编程语言来撰写,此称之为硬件描述语言(Hardware Description Language,HDL),而HDL也并非仅有一种,而是有许多种,如ABEL(Advanced Boolean Expression Language)、AHDL(Altera Hardware Description Language)、Confluence、CUPL(Cornell University Programming Language)、HDCal、JHDL(Java Hardware Description Language)、Lava、Lola、MyHDL、PALASM、RHDL(Ruby Hardware Description Language)等,目前最普遍使用的是VHDL(Very-High-Speed Integrated Circuit Hardware Description  Language)与Verilog2。本领域技术人员也应该清楚,只需要将方法流程用上述几种硬件描述语言稍作逻辑编程并编程到集成电路中,就可以很容易得到实现该逻辑方法流程的硬件电路。
上述实施例阐明的系统、装置、模块或单元,具体可以由计算机芯片或实体实现,或者由具有某种功能的产品来实现。一种典型的实现设备为计算机。具体的,计算机例如可以为个人计算机、膝上型计算机、蜂窝电话、相机电话、智能电话、个人数字助理、媒体播放器、导航设备、电子邮件设备、游戏控制台、平板计算机、可穿戴设备或者这些设备中的任何设备的组合。
通过以上的实施方式的描述可知,本领域的技术人员可以清楚地了解到本说明书可借助软件加必需的通用硬件平台的方式来实现。基于这样的理解,本说明书的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来,该计算机软件产品可以存储在存储介质中,如ROM/RAM、磁碟、光盘等,包括若干指令用以使得一台计算机设备(可以是个人计算机,服务器,或者网络设备等)执行本说明书各个实施例或者实施例的某些部分所述的方法。
本说明书可用于众多通用或专用的计算机系统环境或配置中。例如:个人计算机、服务器计算机、手持设备或便携式设备、平板型设备、多处理器系统、基于微处理器的系统、置顶盒、可编程的消费电子设备、网络PC、小型计算机、大型计算机、包括以上任何系统或设备的分布式计算环境等等。
本说明书可以在由计算机执行的计算机可执行指令的一般上下文中描述,例如程序模块。一般地,程序模块包括执行特定任务或实现特定抽象数据类型的例程、程序、对象、组件、数据结构等等。也可以在分布式计算环境中实践本说明书,在这些分布式计算环境中,由通过通信网络而被连接的远程处理设备来执行任务。在分布式计算环境中,程序模块可以位于包括存储设备在内的本地和远程计算机存储介质中。
虽然通过实施例描绘了本说明书,本领域普通技术人员知道,本说明书有许多变形和变化而不脱离本说明书的精神,希望所附的权利要求包括这些变形和变化而不脱离本说明书的精神。

Claims (12)

  1. 一种损失函数取值的确定方法,所述损失函数包括第一项和第二项;所述方法包括:
    根据第一参数的份额与合作方秘密分享第一数据,得到第一数据的份额;
    根据第一数据的份额,确定第一项取值的份额;
    根据第二参数的份额和随机数与合作方秘密分享第二数据,得到第二数据的份额;
    根据随机数确定第二项的系数,第二项的系数和第二数据用于共同确定第二项的取值;
    向合作方发送第一项取值的份额、第二数据的份额和第二项的系数,以便合作方确定所述损失函数的取值。
  2. 如权利要求1所述的方法,所述确定第一项取值的份额,包括:
    对第一数据的份额进行累加;
    根据累加结果确定第一项取值的份额。
  3. 如权利要求1所述的方法,所述方法包括:
    生成与第二参数的份额相对应的随机数。
  4. 如权利要求1或3所述的方法,所述根据随机数确定第二项的系数,包括:
    对随机数进行累乘,得到第二项的系数。
  5. 如权利要求1或3所述的方法,所述根据第二参数的份额和随机数与合作方秘密分享第二数据,得到第二数据的份额,包括:
    将第二参数的份额与随机数相乘;
    根据随机数与合作方秘密分享第三数据,得到相应的份额;
    将乘积结果与第三数据的份额相加,得到第二数据的份额。
  6. 一种损失函数取值的确定装置,所述损失函数包括第一项和第二项;所述装置包括:
    第一秘密分享单元,用于根据第一参数的份额与合作方秘密分享第一数据,得到第一数据的份额;
    第一确定单元,用于根据第一数据的份额,确定第一项取值的份额;
    第二秘密分享单元,用于根据第二参数的份额和随机数与合作方秘密分享第二数据,得到第二数据的份额;
    第二确定单元,用于根据随机数确定第二项的系数,第二项的系数和第二数据用于共同确定第二项的取值;
    发送单元,用于向合作方发送第一项取值的份额、第二数据的份额和第二项的系数,以便合作方确定所述损失函数的取值。
  7. 一种电子设备,包括:
    存储器,用于存储计算机指令;
    处理器,用于执行所述计算机指令以实现如权利要求1-5中任一项所述的方法步骤。
  8. 一种损失函数取值的确定方法,所述损失函数包括第一项和第二项;所述方法包括:
    根据第一参数的份额和标签值与合作方秘密分享第一数据,得到第一数据的份额;
    根据第一数据的份额,确定第一项取值的第一份额;
    根据第二参数的份额与合作方秘密分享第二数据,得到第二数据的第一份额;
    接收合作方发来的第一项取值的第二份额、第二数据的第二份额和第二项的系数,第二项的系数和第二数据用于共同确定第二项的取值;
    根据第一项取值的第一份额、第一项取值的第二份额、第二数据的第一份额、第二数据的第二份额和第二项的系数,确定所述损失函数的值。
  9. 如权利要求8所述的方法,所述根据第一参数的份额和标签值与合作方秘密分享第一数据,得到第一数据的份额,包括:
    将第一参数的份额与标签值按照预设运算规则进行运算;
    根据标签值与合作方秘密分享第三数据,得到相应的份额;
    将运算结果与第三数据的份额相加,得到第一数据的份额。
  10. 如权利要求8所述的方法,所述根据第一项取值的第一份额、第一项取值的第二份额、第二数据的第一份额、第二数据的第二份额和第二项的系数,确定所述损失函数的值,包括:
    将第一项取值的第一份额和第一项取值的第二份额相加,得到第一项的取值;
    将第二数据的第一份额和第二数据的第二份额相加,得到第二数据;
    根据第二数据和第二项的系数,确定第二项的取值;
    将第一项的取值和第二项的取值相加,得到所述损失函数的值。
  11. 一种损失函数取值的确定装置,所述损失函数包括第一项和第二项;所述装置包括:
    第一秘密分享单元,用于根据第一参数的份额和标签值与合作方秘密分享第一数据,得到第一数据的份额;
    第一确定单元,用于根据第一数据的份额,确定第一项取值的第一份额;
    第二秘密分享单元,用于根据第二参数的份额与合作方秘密分享第二数据,得到第二数据的第一份额;
    接收单元,用于接收合作方发来的第一项取值的第二份额、第二数据的第二份额和第二项的系数,第二项的系数和第二数据用于共同确定第二项的取值;
    第二确定单元,用于根据第一项取值的第一份额、第一项取值的第二份额、第二数据的第一份额、第二数据的第二份额和第二项的系数,确定所述损失函数的值。
  12. 一种电子设备,包括:
    存储器,用于存储计算机指令;
    处理器,用于执行所述计算机指令以实现如权利要求8-10中任一项所述的方法步骤。
PCT/CN2020/070939 2019-05-23 2020-01-08 损失函数取值的确定方法、装置和电子设备 Ceased WO2020233137A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/786,337 US10956597B2 (en) 2019-05-23 2020-02-10 Loss function value determination method and device and electronic equipment

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201910434666.0A CN110263294B (zh) 2019-05-23 2019-05-23 损失函数取值的确定方法、装置和电子设备
CN201910434666.0 2019-05-23

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/786,337 Continuation US10956597B2 (en) 2019-05-23 2020-02-10 Loss function value determination method and device and electronic equipment

Publications (1)

Publication Number Publication Date
WO2020233137A1 true WO2020233137A1 (zh) 2020-11-26

Family

ID=67915183

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/070939 Ceased WO2020233137A1 (zh) 2019-05-23 2020-01-08 损失函数取值的确定方法、装置和电子设备

Country Status (3)

Country Link
CN (1) CN110263294B (zh)
TW (1) TWI710981B (zh)
WO (1) WO2020233137A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114648130A (zh) * 2022-02-07 2022-06-21 北京航空航天大学 纵向联邦学习方法、装置、电子设备及存储介质

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110263294B (zh) * 2019-05-23 2020-08-04 阿里巴巴集团控股有限公司 损失函数取值的确定方法、装置和电子设备
US10956597B2 (en) 2019-05-23 2021-03-23 Advanced New Technologies Co., Ltd. Loss function value determination method and device and electronic equipment
CN111737757B (zh) * 2020-07-31 2020-11-17 支付宝(杭州)信息技术有限公司 针对隐私数据进行安全运算的方法和装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120330867A1 (en) * 2011-06-27 2012-12-27 International Business Machines Corporation Systems and methods for large-scale randomized optimization for problems with decomposable loss functions
US20170300811A1 (en) * 2016-04-14 2017-10-19 Linkedin Corporation Dynamic loss function based on statistics in loss layer of deep convolutional neural network
CN108418810A (zh) * 2018-02-08 2018-08-17 中国人民解放军国防科技大学 一种基于Hadamard矩阵的秘密分享方法
CN108764666A (zh) * 2018-05-15 2018-11-06 国网上海市电力公司 基于多质量损失函数综合的用户暂降经济损失评估方法
CN110263294A (zh) * 2019-05-23 2019-09-20 阿里巴巴集团控股有限公司 损失函数取值的确定方法、装置和电子设备

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8468244B2 (en) * 2007-01-05 2013-06-18 Digital Doors, Inc. Digital information infrastructure and method for security designated data and with granular data stores
US9052709B2 (en) * 2010-07-30 2015-06-09 Kla-Tencor Corporation Method and system for providing process tool correctables
CN107153630B (zh) * 2016-03-04 2020-11-06 阿里巴巴集团控股有限公司 一种机器学习系统的训练方法和训练系统
US11321609B2 (en) * 2016-10-19 2022-05-03 Samsung Electronics Co., Ltd Method and apparatus for neural network quantization
CN106548210B (zh) * 2016-10-31 2021-02-05 腾讯科技(深圳)有限公司 基于机器学习模型训练的信贷用户分类方法及装置
US10536437B2 (en) * 2017-01-31 2020-01-14 Hewlett Packard Enterprise Development Lp Performing privacy-preserving multi-party analytics on vertically partitioned local data
CN109214404A (zh) * 2017-07-07 2019-01-15 阿里巴巴集团控股有限公司 基于隐私保护的训练样本生成方法和装置
CN107679859B (zh) * 2017-07-18 2020-08-25 中国银联股份有限公司 一种基于迁移深度学习的风险识别方法以及系统
CN109426861A (zh) * 2017-08-16 2019-03-05 阿里巴巴集团控股有限公司 数据加密、机器学习模型训练方法、装置及电子设备
CN110832596B (zh) * 2017-10-16 2021-03-26 因美纳有限公司 基于深度学习的深度卷积神经网络训练方法
CN109492420B (zh) * 2018-12-28 2021-07-20 深圳前海微众银行股份有限公司 基于联邦学习的模型参数训练方法、终端、系统及介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20120330867A1 (en) * 2011-06-27 2012-12-27 International Business Machines Corporation Systems and methods for large-scale randomized optimization for problems with decomposable loss functions
US20170300811A1 (en) * 2016-04-14 2017-10-19 Linkedin Corporation Dynamic loss function based on statistics in loss layer of deep convolutional neural network
CN108418810A (zh) * 2018-02-08 2018-08-17 中国人民解放军国防科技大学 一种基于Hadamard矩阵的秘密分享方法
CN108764666A (zh) * 2018-05-15 2018-11-06 国网上海市电力公司 基于多质量损失函数综合的用户暂降经济损失评估方法
CN110263294A (zh) * 2019-05-23 2019-09-20 阿里巴巴集团控股有限公司 损失函数取值的确定方法、装置和电子设备

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114648130A (zh) * 2022-02-07 2022-06-21 北京航空航天大学 纵向联邦学习方法、装置、电子设备及存储介质
CN114648130B (zh) * 2022-02-07 2024-04-16 北京航空航天大学 纵向联邦学习方法、装置、电子设备及存储介质

Also Published As

Publication number Publication date
CN110263294A (zh) 2019-09-20
TW202044241A (zh) 2020-12-01
TWI710981B (zh) 2020-11-21
CN110263294B (zh) 2020-08-04

Similar Documents

Publication Publication Date Title
TWI745861B (zh) 資料處理方法、裝置和電子設備
TWI761778B (zh) 模型參數確定方法、裝置和電子設備
TWI734041B (zh) 一種資料審計的方法及裝置
TWI740375B (zh) 資料交互方法、裝置和電子設備
TWI724809B (zh) 模型參數確定方法、裝置和電子設備
CN110162551B (zh) 数据处理方法、装置和电子设备
TWI710981B (zh) 損失函數取值的確定方法、裝置和電子設備
TWI717221B (zh) 模型參數確定方法、裝置和電子設備
TWI788662B (zh) 安全認證方法、訓練安全認證模型的方法、安全認證裝置、安全認證模型的訓練裝置、電子設備及電腦可讀儲存媒體
CN104769605A (zh) 用于图形处理单元的安全环境
CN110580409A (zh) 模型参数确定方法、装置和电子设备
US20210042640A1 (en) Determining model parameters using secret sharing
CN110391895A (zh) 数据预处理方法、密文数据获取方法、装置和电子设备
CN111967035A (zh) 模型训练方法、装置和电子设备
TW202103149A (zh) 資料互動方法、裝置、伺服器和電子設備
US20200293908A1 (en) Performing data processing based on decision tree
TW202103151A (zh) 資料處理方法、裝置和電子設備
WO2020088008A1 (zh) 基于区块链的生物特征信息处理方法和装置、终端设备
CN112507323A (zh) 基于单向网络的模型训练方法、装置和计算设备
CN116629381A (zh) 一种联邦迁移学习方法、装置、存储介质及电子设备
CN111324778A (zh) 数据、业务处理方法、装置及电子设备
CN111460515B (zh) 数据匹配方法、装置和电子设备
US10956597B2 (en) Loss function value determination method and device and electronic equipment
CN112671689B (zh) 数据上链方法、装置、电子设备及计算机存储介质
CN114091417A (zh) 表格拆分方法、装置、设备及介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20810150

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20810150

Country of ref document: EP

Kind code of ref document: A1