CN112668020A - Feature interleaving method, device, readable storage medium and computer program product - Google Patents
Feature interleaving method, device, readable storage medium and computer program product Download PDFInfo
- Publication number
- CN112668020A CN112668020A CN202011554365.0A CN202011554365A CN112668020A CN 112668020 A CN112668020 A CN 112668020A CN 202011554365 A CN202011554365 A CN 202011554365A CN 112668020 A CN112668020 A CN 112668020A
- Authority
- CN
- China
- Prior art keywords
- feature
- encrypted
- cross
- sample
- characteristic
- 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.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 89
- 238000004590 computer program Methods 0.000 title claims abstract description 14
- 238000011156 evaluation Methods 0.000 claims abstract description 122
- 230000000694 effects Effects 0.000 claims abstract description 79
- 230000003993 interaction Effects 0.000 claims abstract description 18
- 238000004364 calculation method Methods 0.000 claims abstract description 17
- 238000012549 training Methods 0.000 claims description 11
- 238000000926 separation method Methods 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 5
- 238000010801 machine learning Methods 0.000 description 16
- 238000004891 communication Methods 0.000 description 6
- 230000007547 defect Effects 0.000 description 5
- 238000010586 diagram Methods 0.000 description 5
- 238000005516 engineering process Methods 0.000 description 5
- 238000013473 artificial intelligence Methods 0.000 description 3
- 230000002708 enhancing effect Effects 0.000 description 3
- 238000011161 development Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The application discloses a feature crossing method, a device, a readable storage medium and a computer program product, wherein the feature crossing method is applied to a third-party device, and comprises the following steps: generating an encryption public key and an encryption private key, and sending the encryption public key to each participant device, so that a first device in each participant device performs federated cross feature calculation with each second device in each participant device based on the acquired first local feature and the encryption public key to generate an encryption cross feature, and further receives the encryption cross feature sent by the first device, and decrypts the encryption cross feature based on the encryption private key to obtain a target cross feature, and further performs feature cross effect evaluation on the target cross feature by performing feature evaluation interaction with a label device with sample label information to obtain a feature cross effect evaluation result. The application solves the technical problem of poor characteristic cross interpretability.
Description
Technical Field
The present application relates to the field of artificial intelligence in financial technology (Fintech), and in particular to a feature intersection method, apparatus, readable storage medium and computer program product.
Background
With the continuous development of financial technologies, especially internet technology and finance, more and more technologies (such as distributed, Blockchain, artificial intelligence and the like) are applied to the financial field, but the financial industry also puts higher requirements on the technologies, such as higher requirements on the distribution of backlog of the financial industry.
With the continuous development of computer software and artificial intelligence, the application field of machine learning is more and more extensive, at present, when longitudinal federal learning is performed, feature crossing is generally required to search features with good training effect for federal learning, at present, feature crossing is generally performed through a machine learning model, in the method, the feature crossing process is completed by an internal model network (such as a hidden layer) of the machine learning model, the model outputs a model prediction result obtained based on the cross features, and then technical personnel cannot observe the effect of a plurality of cross features corresponding to the model prediction result, so that the interpretability of the feature crossing is poor.
Disclosure of Invention
The present application mainly aims to provide a feature crossing method, a device, a readable storage medium and a computer program product, which aim to solve the technical problem of poor interpretability of feature crossing in the prior art.
In order to achieve the above object, the present application provides a feature crossing method, which is applied to a third-party device, and the feature crossing method includes:
generating an encryption public key and an encryption private key, and sending the encryption public key to each participant device, so that a first device in each participant device performs federated cross feature calculation with each second device in each participant device based on the acquired first local feature and the encryption public key, and generates encryption cross features;
receiving the encrypted cross feature sent by the first equipment, and decrypting the encrypted cross feature based on the encrypted private key to obtain a target cross feature;
and performing characteristic evaluation interaction with label side equipment with sample label information, and performing characteristic cross effect evaluation on the target cross characteristics to obtain a characteristic cross effect evaluation result.
In order to achieve the above object, the present application provides a feature crossing method, which is applied to a first device, the feature crossing method including:
receiving an encrypted public key sent by third-party equipment, and encrypting the acquired first local feature into a first encrypted local feature based on the encrypted public key;
receiving a second local encryption feature sent by a second device, wherein the second device is used for encrypting the obtained second local feature into a second encrypted local feature based on the encryption public key;
performing homomorphic encryption multiplication on the first encryption local feature and the second encryption local feature to obtain an encryption cross feature;
and sending the encrypted cross feature to the third-party equipment so that the third-party equipment can decrypt the encrypted cross feature into a target cross feature based on an encrypted private key corresponding to the encrypted public key and evaluate the feature cross effect of the target cross feature.
In order to achieve the above object, the present application provides a feature crossing method, which is applied to a second device, the feature crossing method including:
receiving an encrypted public key sent by third-party equipment, and encrypting the acquired second local feature into a second encrypted local feature based on the encrypted public key;
and sending the second encrypted local feature to the first device, so that the first device can calculate the encrypted cross feature based on the first encrypted local feature and the second encrypted local feature generated by local encryption.
In order to achieve the above object, the present application provides a feature crossing method, which is applied to a label side device, and the feature crossing method includes:
acquiring encrypted sample label information, and sending the encrypted sample label information to third-party equipment so that the third-party equipment can generate encrypted box sample label information corresponding to each cross characteristic sample box of the acquired target cross characteristic based on the encrypted sample label information;
receiving the label information of each encrypted sample in a sub-box, and decrypting the label information of each encrypted sample in a sub-box to obtain the label information of each sample in the sub-box;
calculating a characteristic evaluation value of each cross characteristic sample bin based on the label information of each bin sample;
feeding back each of the feature evaluation values to the third-party device.
The present application further provides a feature crossing device, the feature crossing device is a virtual device, and the feature crossing device is applied to a third-party device, the feature crossing device includes:
the key management module is used for generating an encrypted public key and an encrypted private key and sending the encrypted public key to each participant device so that the first device in each participant device performs federated cross feature calculation with each second device in each participant device based on the acquired first local feature and the encrypted public key to generate an encrypted cross feature;
the decryption module is used for receiving the encrypted cross feature sent by the first equipment and decrypting the encrypted cross feature based on the encrypted private key to obtain a target cross feature;
and the evaluation module is used for evaluating the characteristic cross effect of the target cross characteristic by performing characteristic evaluation interaction with the label side equipment with the sample label information to obtain a characteristic cross effect evaluation result.
The present application further provides a feature crossing device, the feature crossing device is a virtual device, and the feature crossing device is applied to a first device, the feature crossing device includes:
the encryption module is used for receiving an encrypted public key sent by third-party equipment and encrypting the acquired first local feature into a first encrypted local feature based on the encrypted public key;
a receiving module, configured to receive a second local encryption feature sent by a second device, where the second device is configured to encrypt the obtained second local feature into a second encrypted local feature based on the encryption public key;
the multiplication module is used for carrying out homomorphic encryption multiplication on the first encryption local characteristic and the second encryption local characteristic to obtain an encryption cross characteristic;
and the sending module is used for sending the encrypted cross feature to the third-party equipment so that the third-party equipment can decrypt the encrypted cross feature into a target cross feature based on an encrypted private key corresponding to the encrypted public key and evaluate the feature cross effect of the target cross feature.
The present application further provides a feature crossing device, the feature crossing device is a virtual device, and the feature crossing device is applied to a second device, the feature crossing device includes:
the encryption module is used for receiving an encrypted public key sent by third-party equipment and encrypting the acquired second local feature into a second encrypted local feature based on the encrypted public key;
and the sending module is used for sending the second encrypted local feature to the first equipment so that the first equipment can calculate the encrypted cross feature based on the first encrypted local feature and the second encrypted local feature generated by local encryption.
The present application further provides a feature crossing device, the feature crossing device is a virtual device, and the feature crossing device is applied to a label side device, the feature crossing device includes:
the sending module is used for obtaining encrypted sample label information and sending the encrypted sample label information to third-party equipment so that the third-party equipment can generate encrypted sub-box sample label information corresponding to each cross characteristic sample sub-box of the obtained target cross characteristic based on the encrypted sample label information;
the decryption module is used for receiving the encrypted sub-box sample label information and decrypting the encrypted sub-box sample label information to obtain the sub-box sample label information;
the calculation module is used for calculating the characteristic evaluation value of each cross characteristic sample bin based on the label information of each bin sample;
and the feedback module is used for feeding back each characteristic evaluation value to the third-party equipment.
The present application further provides a feature crossing device, where the feature crossing device is an entity device, the feature crossing device includes: a memory, a processor and a program of the feature interleaving method stored on the memory and executable on the processor, the program of the feature interleaving method being executable by the processor to implement the steps of the feature interleaving method as described above.
The present application also provides a readable storage medium having stored thereon a program for implementing a feature intersection method, the program implementing the steps of the feature intersection method as described above when executed by a processor.
The present application also provides a computer program product comprising a computer program which, when executed by a processor, performs the steps of the feature interleaving method as described above.
Compared with the technical means of performing feature crossing through a machine learning model adopted in the prior art, the feature crossing method, the device, the readable storage medium and the computer program product firstly generate an encryption public key and an encryption private key, further send the encryption public key to each participant device, so that a first device in each participant device performs federated crossing feature calculation with each second device in each participant device based on the acquired first local feature and the encryption public key to generate an encryption crossing feature, further receive the encryption crossing feature sent by the first device, decrypt the encryption crossing feature based on the encryption private key to acquire a target crossing feature, and therefore a third party device can acquire the target crossing feature acquired by performing feature crossing on two or more known features from each participant device, further realizing the purpose of understanding the source of the target cross feature, improving the interpretability of the feature cross, further realizing the evaluation of the quality of the feature cross effect by carrying out the feature evaluation interaction with the label side equipment with the sample label information, carrying out the feature cross effect evaluation on the target cross feature, obtaining the evaluation result of the feature cross effect, further realizing the evaluation of the quality of the feature cross effect by carrying out the joint evaluation with the label side equipment with the sample label, further directly observing the quality of the feature cross effect by technical personnel, further enhancing the interpretability of the feature cross, overcoming the defects that when the feature cross is usually carried out by a machine learning model in the prior art, the feature cross process is finished by an internal model network (such as a hidden layer) of the machine learning model, the model output is the model prediction result obtained based on the cross feature, and further the technical personnel can not observe the quality of a plurality of cross features corresponding to the model prediction result, furthermore, the interpretability of the feature intersection is improved.
Drawings
The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and together with the description, serve to explain the principles of the application.
In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings needed to be used in the description of the embodiments or the prior art will be briefly described below, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a schematic flow chart of a first embodiment of a feature intersection method of the present application;
fig. 2 is a schematic diagram illustrating interaction among a first device, a second device, a third-party device, and a tag-side device in the feature interleaving method of the present application;
FIG. 3 is a schematic flow chart of a second embodiment of the feature crossing method of the present application;
FIG. 4 is a schematic flow chart of a third embodiment of the feature crossing method of the present application;
FIG. 5 is a schematic flow chart of a fourth embodiment of the feature crossing method of the present application;
FIG. 6 is a schematic diagram of an apparatus configuration of a hardware operating environment according to an embodiment of the present application;
fig. 7 is a schematic diagram of a hardware architecture according to an embodiment of the present application.
The objectives, features, and advantages of the present application will be further described with reference to the accompanying drawings.
Detailed Description
It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.
In a first embodiment of the feature crossing method of the present application, referring to fig. 1, the feature crossing method is applied to a third-party device, and the feature crossing method includes:
step S10, generating an encrypted public key and an encrypted private key, and sending the encrypted public key to each participant device, so that the first device in each participant device performs federated cross feature calculation with each second device in each participant device based on the acquired first local feature and the encrypted public key, and generates an encrypted cross feature;
in this embodiment, it should be noted that the third-party device belongs to a trusted third party, where the participant device includes a first device and second devices, where after receiving the encryption public key, each participant encrypts its local feature into an encryption feature, and collects the encryption features at the first device to obtain an encryption cross feature, and sends the encryption cross feature to the third-party device through the first device, where the local feature may be represented by a vector, for example, it is assumed that the local feature is a vector (a, B, C), where a is a feature value corresponding to a sample a, B is a feature value corresponding to a sample B, and C is a feature value corresponding to a sample C.
Additionally, it should be noted that the encrypted cross feature is obtained by the homomorphic cryptographic multiplication of a first encrypted local feature generated based on the first local feature and the encrypted public key and a second encrypted local cross feature sent by the second device by the first device.
Generating an encryption public key and an encryption private key, and sending the encryption public key to each participant device, so that a first device in each participant device performs federated cross feature calculation with each second device in each participant device based on the acquired first local feature and the encryption public key to generate an encryption cross feature, and specifically, generates an encryption public key and an encryption private key in a preset encryption mode, wherein the preset encryption mode includes a multiplicative homomorphic encryption mode and a fully homomorphic encryption mode, wherein, because the characteristic cross is a multiplication between the features, when the preset encryption mode is the multiplicative homomorphic encryption mode, the operation complexity at the time of the characteristic cross is low, the calculation efficiency at the time of the characteristic cross can be improved, and the encryption public key is sent to each participant device, and further, the first device in each participant device encrypts the acquired first local feature based on the encryption public key, the method comprises the steps that first encrypted local features are obtained, each second device encrypts the obtained second local features based on the encryption public key to obtain each second encrypted local feature, each second device sends the corresponding second encrypted local feature to the first device, the first device performs homomorphic encryption multiplication on the first encrypted local features and the corresponding second encrypted local features after receiving the corresponding second encrypted local features to obtain encrypted cross features, and the first device sends the encrypted cross features to third-party devices.
Step S20, receiving the encrypted cross feature sent by the first device, and decrypting the encrypted cross feature based on the encrypted private key to obtain a target cross feature;
in this embodiment, it should be noted that the encrypted public key and the encrypted private key are a pair of public and private keys, the target cross feature is a decrypted encrypted cross feature, and since the target cross feature is held by a third-party device, each participating-party device cannot reversely deduce local features of other parameter-party devices through the target cross feature, so as to implement privacy protection in the feature cross process, where a specific process of generating the encrypted cross feature may refer to steps a10 to a30 in the second embodiment, and is not described herein again.
And step S30, performing characteristic evaluation interaction with the label side equipment with the sample label information, and performing characteristic cross effect evaluation on the target cross characteristics to obtain a characteristic cross effect evaluation result.
In this embodiment, it should be noted that, the labeler device is a device that holds sample label information of the target cross feature, the feature evaluation interaction process is a process in which the third-party device performs, by interacting with the labeler device, feature cross effect evaluation on the target cross feature in combination with the sample label information of the labeler device on the premise that private data is not disclosed, where the sample label information includes a first sample label value and a second sample label value of the target cross feature in the labeler device, where the first sample label value is a positive sample label that identifies the target cross feature, the second sample label value is a negative sample label that identifies the target cross feature, a sum of the first sample label value and the second sample label value is 1, for example, assuming that the first sample label value is y, the second sample tag value is 1-y.
Performing characteristic evaluation interaction with a label side device with sample label information to perform characteristic cross effect evaluation on the target cross characteristics to obtain a characteristic cross effect evaluation result, specifically, performing sample ID alignment on the third side device and the label side device to enable the label side device to obtain target sample IDs corresponding to target cross characteristic samples corresponding to the target cross characteristics, further generating an addition homomorphic encryption second encryption key by the label side device, encrypting a first sample label value corresponding to a target sample corresponding to each target sample ID and a corresponding second sample label value based on the second encryption key to obtain each first encryption sample label value and each second encryption sample label value, and further sending each first encryption sample label value and each second encryption sample label value to the third side device, the third-party device performs binning on the target cross feature samples corresponding to the target cross features to obtain cross feature sample binning, and counts the sum of first encrypted sample label values and the sum of second encrypted sample label values corresponding to the target cross feature samples in the cross feature sample binning, and then the third-party device sends the sum of the first encrypted sample label values and the sum of the second encrypted sample label values to the tagger device to enable the tagger device to calculate a feature evaluation value based on the sum of the first encrypted sample label values and the sum of the second encrypted sample label values, wherein the feature evaluation value comprises an IV value, a WOE value and the like, and sends the feature evaluation value to the third-party device, and then the third-party device performs feature cross effect evaluation on the target cross features based on the feature evaluation value, and obtaining a characteristic cross effect evaluation result.
The method comprises the following steps of performing characteristic evaluation interaction with label side equipment with sample label information, performing characteristic cross effect evaluation on the target cross characteristics, and obtaining a characteristic cross effect evaluation result, wherein the step of performing the characteristic evaluation interaction with the label side equipment with the sample label information comprises the following steps:
step S31, receiving encrypted sample label information sent by the labeler device, where the encrypted sample label information is obtained by encrypting the sample label information by the labeler device;
in this embodiment, it should be noted that the sample tag information includes a first sample tag value and a second sample tag value of a target sample corresponding to each target sample ID in the labeler device, and the encrypted sample information includes a first encrypted sample tag value corresponding to each first sample tag value and a second encrypted sample tag value corresponding to each second sample tag value.
Step S32, performing box separation on each target cross feature sample corresponding to the target cross feature to obtain each cross feature sample box separation;
in this embodiment, each target cross feature sample corresponding to the target cross feature is subjected to binning to obtain each cross feature sample bin, and specifically, each target cross feature sample is subjected to binning according to a feature value of each target cross feature sample in the target cross feature to obtain each cross feature sample bin.
Step S33, based on the encrypted sample label information, counting the encrypted sub-box sample label information corresponding to each cross characteristic sample sub-box;
in this embodiment, based on the encrypted sample label information, statistics is performed on encrypted binning sample label information corresponding to each cross feature sample bin, specifically, the following steps are performed for each cross feature sample bin:
calculating the sum of first encrypted sample label values corresponding to each target cross characteristic sample in the cross characteristic sample sub-box to obtain first encrypted sub-box sample label information, calculating the sum of second encrypted sample label values corresponding to each target cross characteristic sample in the cross characteristic sample sub-box to obtain second encrypted sub-box sample label information, and taking the first encrypted sub-box sample label information corresponding to each cross characteristic sample sub-box and the corresponding second encrypted sub-box sample label information as encrypted sub-box sample label information.
Wherein the encrypted sample label information at least comprises a sample identifier, a first encrypted sample label value corresponding to the sample identifier and a second encrypted sample label value corresponding to the sample identifier, the encrypted binning sample label information comprises first encrypted binning sample label information and second encrypted binning sample label information,
the step of counting the encrypted sub-box sample label information corresponding to each cross feature sample sub-box based on the encrypted sample label information includes:
step S331, based on each sample identifier, determining each first encrypted sample tag value corresponding to each cross feature sample sub-box and each second encrypted sample tag value corresponding to each cross feature sample sub-box;
in this embodiment, it should be noted that the sample identifier is a target sample ID for performing sample ID alignment determination on the third-party device and the tagger device.
Based on each sample identifier, determining each first encrypted sample tag value and each second encrypted sample tag value respectively corresponding to each cross feature sample bin, and specifically, based on each target sample ID, determining a first encrypted sample tag value and a second encrypted sample tag value corresponding to each target cross feature sample in each cross feature sample bin, where one target sample ID corresponds to one target cross feature sample and corresponds to one first encrypted sample tag value and one second encrypted sample tag value.
Step S332, calculating the sum of the label values of the first encrypted samples corresponding to the cross characteristic sample sub-boxes respectively to obtain label information of the first encrypted sub-boxes corresponding to the cross characteristic sample sub-boxes;
in this embodiment, the sum of the first encrypted sample tag values corresponding to each cross feature sample bin is calculated to obtain first encrypted bin sample tag information corresponding to each cross feature sample bin, specifically, homomorphic encrypted addition is performed on each first encrypted sample tag value corresponding to each cross feature sample bin to obtain the sum of the first encrypted sample tag values corresponding to each cross feature sample bin, and the sum of each first encrypted sample tag value is used as the first encrypted bin sample tag information corresponding to each cross feature sample bin.
Step S333, calculating a sum of each second encrypted sample label value corresponding to each cross feature sample bin, and obtaining second encrypted bin sample label information corresponding to each cross feature sample bin.
In this embodiment, the sum of the second encrypted sample tag values corresponding to each cross feature sample bin is calculated to obtain second encrypted bin sample tag information corresponding to each cross feature sample bin, specifically, homomorphic encrypted addition is performed on each second encrypted sample tag value corresponding to each cross feature sample bin to obtain the sum of the second encrypted sample tag values corresponding to each cross feature sample bin, and the sum of each second encrypted sample tag value is used as the second encrypted bin sample tag information corresponding to each cross feature sample bin.
Step S34, sending each encrypted sub-box sample label information to the label side equipment, so that the label side equipment can calculate a characteristic evaluation value corresponding to each cross characteristic sample sub-box based on the encrypted sub-box sample label information;
in this embodiment, each encrypted sample label information is sent to the label side device, so that the label side device calculates a feature evaluation value corresponding to each cross feature sample label based on the encrypted sample label information, specifically, a sum of first encrypted sample label values corresponding to each cross feature sample label and a sum of second encrypted sample label values are sent to the label side device, so that the label side device decrypts the sum of the first encrypted sample label values and the sum of the second encrypted sample label values to obtain the sum of first sample label values corresponding to each cross feature sample label and the sum of corresponding second sample label values, and further, the label side device obtains the sum of first sample label values and the sum of second sample label values corresponding to each cross feature sample label based on a preset IV value calculation formula, the sum of the first sample label values, and the sum of the second sample label values, calculating an IV value corresponding to each cross feature sample bin, and taking each IV value as each feature evaluation value, wherein the preset feature evaluation value calculation formula is as follows:
wherein IViFor the IV value corresponding to the ith cross-feature sample bin, p is the sum of all the first sample tag values, yiIs the sum of the first sample label values corresponding to the ith cross feature sample sub-box, niThe sum of label values of second samples corresponding to the ith cross feature sample sub-box, wherein the label side equipmentThe specific process of calculating the feature evaluation value may refer to steps C10 to C40 in the fourth embodiment, and will not be described herein again.
Step S35, receiving each feature evaluation value sent by the tagger device, and evaluating a feature crossing effect of the target crossing feature based on each feature evaluation value, to obtain a result of evaluating the feature crossing effect.
In this embodiment, each feature evaluation value sent by the tag side device is received, and based on each feature evaluation value, a feature crossing effect of the target crossing feature is evaluated to obtain the feature crossing effect evaluation result, specifically, each feature evaluation value sent by the tag side device is received, each feature evaluation value is compared with a preset feature evaluation value threshold, if the feature evaluation value is greater than the preset feature evaluation value threshold, the feature crossing effect of the target crossing feature at a cross feature sample bin corresponding to the feature evaluation value is proved to be good, and the feature evaluation value is taken as a target feature evaluation value, if the feature evaluation value is less than or equal to the preset feature evaluation value threshold, the feature crossing effect of the target crossing feature at a cross feature sample bin corresponding to the feature evaluation value is proved to be poor, and further determining an evaluation value ratio of a target feature evaluation value in each feature evaluation value, and further evaluating a feature crossing effect of the target crossing feature based on the evaluation value ratio to obtain a feature crossing effect evaluation result, wherein the higher the evaluation value ratio is, the better the feature crossing effect is.
After the step of performing feature evaluation interaction with the label side device having the sample label information, performing feature cross effect evaluation on the target cross feature, and obtaining a feature cross effect evaluation result, the feature cross method further includes:
step S40, based on the characteristic cross effect evaluation result, performing characteristic screening on each generated target cross characteristic to obtain federal training characteristic data;
in this embodiment, it should be noted that at least one object intersection feature exists in the third-party device.
And screening the characteristics of each generated target cross characteristic based on the characteristic cross effect evaluation result to obtain federal training characteristic data, specifically, screening each target characteristic of which the corresponding evaluation value ratio is greater than a preset evaluation value ratio threshold value from each target cross characteristic generated by third-party equipment based on the characteristic cross effect evaluation result as the federal training characteristic data, wherein the federal training characteristic data at least comprises one target characteristic.
And step S50, carrying out federal learning modeling with each participant device based on the federal training characteristic data, and generating a target federal model.
In this embodiment, it should be noted that, when the federal learning modeling is performed, the third-party device is a party involved in the federal learning modeling, and is configured to provide the federal training feature data, and each party device is configured to provide respective local features, so as to expand the feature dimension used for the federal learning modeling, and further improve the effect of the federal learning modeling.
Fig. 2 is a schematic diagram of interaction among a first device, a second device, a third-party device and a tag-side device, where a is the first device, B is the second device, C is the third-party device, D is the tag-side device, label data is sample tag data of the tag-side device, IV value is a feature evaluation value,
in order to be a first local feature,
in order to be a second local feature,
for the purpose of the first encrypted local feature,
for the purpose of the second encrypted local feature,
in order to encrypt the cross-over feature,
is an object intersection feature.
Compared with the technical means of performing feature crossing through a machine learning model adopted in the prior art, the embodiment of the application firstly generates an encryption public key and an encryption private key, and then sends the encryption public key to each participant device, so that a first device in each participant device performs federated crossing feature calculation with each second device in each participant device based on the acquired first local feature and the encryption public key to generate an encryption crossing feature, further receives the encryption crossing feature sent by the first device, decrypts the encryption crossing feature based on the encryption private key to acquire a target crossing feature, and therefore, a third party device can acquire the target crossing feature acquired by performing feature crossing on two or more known features from each participant device, further realizing the purpose of understanding the source of the target cross feature, improving the interpretability of the feature cross, further realizing the evaluation of the quality of the feature cross effect by carrying out the feature evaluation interaction with the label side equipment with the sample label information, carrying out the feature cross effect evaluation on the target cross feature, obtaining the evaluation result of the feature cross effect, further realizing the evaluation of the quality of the feature cross effect by carrying out the joint evaluation with the label side equipment with the sample label, further directly observing the quality of the feature cross effect by technical personnel, further enhancing the interpretability of the feature cross, overcoming the defects that when the feature cross is usually carried out by a machine learning model in the prior art, the feature cross process is finished by an internal model network (such as a hidden layer) of the machine learning model, the model output is the model prediction result obtained based on the cross feature, and further the technical personnel can not observe the quality of a plurality of cross features corresponding to the model prediction result, furthermore, the interpretability of the feature intersection is improved.
Further, referring to fig. 3, based on the first embodiment in the present application, in another embodiment of the present application, the feature crossing method is applied to the first device, and the feature crossing method includes:
step A10, receiving an encrypted public key sent by a third-party device, and encrypting the acquired first local feature into a first encrypted local feature based on the encrypted public key;
in this embodiment, it should be noted that the encryption public key is a public key of multiplicative homomorphic encryption.
Step A20, receiving a second local encryption feature sent by a second device, where the second device is configured to encrypt the obtained second local feature into a second encrypted local feature based on the encryption public key;
in this embodiment, it should be noted that the third party device encrypts the public key to the first device and each second device in a manner.
Step A30, performing homomorphic encryption multiplication on the first encrypted local feature and the second encrypted local feature to obtain an encrypted cross feature;
in this embodiment, homomorphic encryption multiplication is performed on the first encrypted local feature and the second encrypted local feature to obtain an encrypted cross feature, specifically, a product of the first encrypted local feature and the second encrypted local feature is calculated to obtain the encrypted cross feature, where a calculation process of the encrypted cross feature is as follows:
[[X1]]*[[X2]]＝[[X1X2]]
wherein [ [ X ]1]]For the first encrypted local feature, [ [ X ]2]]For the second encrypted local feature, [ [ X ]1X2]]The encryption cross feature.
Step A40, sending the encrypted cross feature to the third-party equipment, so that the third-party equipment decrypts the encrypted cross feature into a target cross feature based on an encrypted private key corresponding to the encrypted public key, and evaluates a feature cross effect of the target cross feature.
In this embodiment, it should be noted that, the target cross feature is held by the third-party device, and since either one of the first device and the second device can obtain the target cross feature and can reversely derive the local feature of the other device, that is, the private data, the first device and the second device cannot hold the target cross feature.
Sending the encrypted cross feature to the third-party equipment so that the third-party equipment decrypts the encrypted cross feature into a target cross feature based on an encrypted private key corresponding to the encrypted public key and evaluates a feature cross effect of the target cross feature, specifically, sending the encrypted cross feature to the third-party equipment, further decrypting the encrypted cross feature into the target cross feature by the third-party equipment based on the encrypted private key corresponding to the encrypted public key, further binning each target cross feature sample corresponding to the target cross feature by the third-party equipment to obtain each cross feature sample bin, calculating an IV value corresponding to each cross feature sample bin by sample tag information of joint label side equipment, and further evaluating the feature cross effect of the target cross feature based on each IV value, the specific process of the third-party device evaluating the feature crossing effect of the target crossing feature may refer to steps S31 to S34 in the first embodiment, which are not described herein again.
Compared with the technical means of performing feature crossing through a machine learning model adopted in the prior art, the feature crossing method comprises the steps of firstly encrypting the acquired first local feature into the first encrypted local feature based on the encrypted public key after receiving the encrypted public key sent by third-party equipment, and then receiving the second local encrypted feature sent by second equipment, wherein the second equipment is used for encrypting the acquired second local feature into the second encrypted local feature based on the encrypted public key, and further performing homomorphic encryption multiplication on the first encrypted local feature and the second encrypted local feature to obtain the encrypted cross feature, and further sending the encrypted cross feature to the third-party equipment so that the third-party equipment can decrypt the encrypted cross feature into the target cross feature based on the encrypted private key corresponding to the encrypted public key, and evaluating the characteristic cross effect of the target cross characteristic, thereby realizing the evaluation of the characteristic cross effect, further enabling a technician to directly observe the quality of the characteristic cross effect, further enhancing the interpretability of the characteristic cross, and overcoming the technical defect that the interpretability of the characteristic cross is poor because the characteristic cross process is finished by an internal model network (such as a hidden layer) of a machine learning model when the characteristic cross is generally carried out by the machine learning model in the prior art, the model outputs a model prediction result obtained based on the cross characteristic, and the technician cannot observe the quality of a plurality of cross characteristics corresponding to the model prediction result, thereby improving the interpretability of the characteristic cross.
Further, referring to fig. 4, based on the first embodiment and the second embodiment in the present application, in another embodiment of the present application, the feature crossing method is applied to the second device, and the feature crossing method includes:
step B10, receiving an encrypted public key sent by a third-party device, and encrypting the obtained second local feature into a second encrypted local feature based on the encrypted public key;
in this embodiment, it should be noted that the encryption public key is a public key of multiplicative homomorphic encryption.
Step B20, sending the second encrypted local feature to the first device, so that the first device calculates an encrypted cross-feature based on the first encrypted local feature and the second encrypted local feature generated by local encryption.
In this embodiment, the second encrypted local feature is sent to the first device, so that the first device calculates an encrypted cross feature based on the first encrypted local feature and the second encrypted local feature generated by local encryption, and specifically, the second encrypted local feature is sent to the first device, so that the first device calculates a product of the first encrypted local feature and the second encrypted local feature generated by local encryption, and obtains an encrypted cross feature, and then the first device sends the encrypted cross feature to the third party device, and then the third party device decrypts the encrypted cross feature based on the encrypted private key of the encrypted public key, obtains a target cross feature, and evaluates a feature cross effect of the target cross feature, where a specific process of generating the encrypted cross feature may refer to steps a10 to a30 in the second embodiment, here, the specific process of the third-party device evaluating the feature crossing effect of the target crossing feature may refer to steps S31 to S34 in the first embodiment, which are not described herein again.
The embodiment of the application provides a method for calculating encrypted cross features, that is, an encrypted public key sent by a third-party device is received, an acquired second local feature is encrypted into a second encrypted local feature based on the encrypted public key, and the second encrypted local feature is sent to a first device, so that the first device calculates encrypted cross features based on a first encrypted local feature generated by local encryption and the second encrypted local feature, and after the first device sends the encrypted cross features to the third-party device, the third-party device can obtain target cross features obtained by feature cross of two or more known features from each participant device, thereby achieving the purpose of explaining the source of the target cross features, improving the interpretability of feature cross, and further performing feature evaluation interaction with a label-party device having sample label information, evaluating the characteristic cross effect of the target cross characteristics to obtain an evaluation result of the characteristic cross effect, further, the quality of the characteristic cross effect can be evaluated by joint evaluation with the label side equipment with the sample label, so that a technician can directly observe the quality of the characteristic cross effect, the interpretability of feature intersection is further enhanced, so when the feature intersection is performed by a machine learning model in the prior art, since the feature crossing process is completed by an internal model network (such as a hidden layer) of the machine learning model, the model outputs a model prediction result obtained based on the crossing features, and then technical personnel can not observe the effect of a plurality of cross features corresponding to the model prediction result, and further lay the foundation for the technical defect of poor interpretability of feature cross.
Further, referring to fig. 5, based on the first embodiment, the second embodiment and the third embodiment in the present application, in another embodiment of the present application, the feature crossing method is applied to a labeler side device, and the feature crossing method includes:
step C10, acquiring encrypted sample label information, and sending the encrypted sample label information to third-party equipment, so that the third-party equipment can generate encrypted sub-box sample label information corresponding to each cross feature sample sub-box of the acquired target cross feature based on the encrypted sample label information;
in this embodiment, it should be noted that, before obtaining the encrypted sample tag information, the third-party device and the tagger device need to perform sample ID alignment, so that the tagger device obtains the target sample IDs corresponding to the target cross feature samples corresponding to the target cross feature.
Obtaining encrypted sample label information, sending the encrypted sample label information to third-party equipment, so that the third-party equipment generates encrypted box sample label information corresponding to each cross characteristic sample box of the obtained target cross characteristic based on the encrypted sample label information, specifically, performing sample ID alignment with the third-party equipment to obtain a target sample ID of each target cross characteristic sample of the target cross characteristic, further obtaining a first sample label value corresponding to each target sample ID and a corresponding second sample label value, further encrypting each first sample label value and each second sample label value to obtain each first encrypted sample label value and each second encrypted sample label value, and sending each first encrypted sample label value and each second encrypted sample label value to the third-party equipment as encrypted sample label information, and then the third-party equipment counts the sum of the first encrypted sample label values corresponding to the target cross characteristic samples in each cross characteristic sample bin of the obtained target cross characteristic and the sum of the corresponding second encrypted sample label values to obtain the sum of the first encrypted sample label values corresponding to each cross characteristic sample bin and the sum of the corresponding second encrypted sample label values, and takes the sum of the first encrypted sample label values corresponding to each cross characteristic sample bin and the sum of the corresponding second encrypted sample label values as the encrypted sub-bin sample label information corresponding to each cross characteristic sample bin.
Step C20, receiving the label information of each encrypted sub-box sample, and decrypting the label information of each encrypted sub-box sample to obtain the label information of each sub-box sample;
in this embodiment, each encrypted sample label information is received and decrypted to obtain each sample label information, specifically, a sum of first encrypted sample label values corresponding to each cross feature sample sub-box and a sum of corresponding second encrypted sample label values are received and decrypted to obtain a sum of first sample label values corresponding to each cross feature sample sub-box and a sum of corresponding second sample label values, and a sum of first sample label values corresponding to each cross feature sample sub-box and a sum of corresponding second sample label values are obtained and used as the sample label information corresponding to each cross feature sample sub-box.
Step C30, calculating the characteristic evaluation value of each cross characteristic sample bin based on the label information of each bin sample;
in this embodiment, a feature evaluation value of each cross feature sample bin is calculated based on each of the bin sample label information, specifically, an IV value corresponding to each of the cross feature sample bins is calculated based on a preset IV value calculation formula, a sum of each first sample label value, and a sum of each second sample label value, and each IV value is taken as each of the feature evaluation values, where the preset feature evaluation value calculation formula is as follows:
wherein IViFor the IV value corresponding to the ith cross-feature sample bin, p is the sum of all the first sample tag values, yiIs the sum of the first sample label values corresponding to the ith cross feature sample sub-box, niAnd the sum of the label values of the second samples corresponding to the ith cross feature sample bin.
Step C40, feeding back each of the feature evaluation values to the third-party device.
In this embodiment, each feature evaluation value is fed back to the third party device, so that the third party device evaluates the feature crossing effect of the target crossing feature based on each feature evaluation value, where the specific process of the third party device evaluating the feature crossing effect of the target crossing feature based on each feature evaluation value may refer to step S35 in the first embodiment, and is not described herein again.
The embodiment of the community provides a method for calculating a feature evaluation value, that is, encrypted sample label information is obtained and sent to a third-party device, so that the third-party device generates encrypted sub-box sample label information corresponding to each cross feature sample sub-box of the obtained target cross feature based on the encrypted sample label information, further receives each encrypted sub-box sample label information, decrypts each encrypted sub-box sample label information to obtain each sub-box sample label information, further calculates the feature evaluation value of each cross feature sample sub-box based on each sub-box sample label information, and further after each feature evaluation value is sent to the third-party device, the third-party device can evaluate the quality of the feature cross effect of the target cross feature based on each feature evaluation value, furthermore, technicians can directly observe the quality of the characteristic cross effect, so that the interpretability of the characteristic cross is further enhanced, and a foundation is laid for overcoming the technical defects that when the characteristic cross is generally carried out through a machine learning model in the prior art, because the characteristic cross process is completed by an internal model network (such as a hidden layer) of the machine learning model, the model outputs a model prediction result obtained based on the cross characteristic, the technicians cannot observe the quality of the effect of a plurality of cross characteristics corresponding to the model prediction result, and the interpretability of the characteristic cross is poor.
Referring to fig. 6, fig. 6 is a schematic device structure diagram of a hardware operating environment according to an embodiment of the present application.
As shown in fig. 6, the feature crossing apparatus may include: a processor 1001, such as a CPU, a memory 1005, and a communication bus 1002. The communication bus 1002 is used for realizing connection communication between the processor 1001 and the memory 1005. The memory 1005 may be a high-speed RAM memory or a non-volatile memory (e.g., a magnetic disk memory). The memory 1005 may alternatively be a memory device separate from the processor 1001 described above.
Optionally, the feature crossing device may further include a rectangular user interface, a network interface, a camera, RF (Radio Frequency) circuitry, a sensor, audio circuitry, a WiFi module, and so forth. The rectangular user interface may comprise a Display screen (Display), an input sub-module such as a Keyboard (Keyboard), and the optional rectangular user interface may also comprise a standard wired interface, a wireless interface. The network interface may optionally include a standard wired interface, a wireless interface (e.g., WI-FI interface).
Those skilled in the art will appreciate that the feature crossing device configuration shown in fig. 6 does not constitute a limitation of the feature crossing device and may include more or fewer components than shown, or some components in combination, or a different arrangement of components.
As shown in fig. 6, a memory 1005, which is a kind of computer storage medium, may include therein an operating system, a network communication module, and a feature intersection program. An operating system is a program that manages and controls the hardware and software resources of a feature crossbar device, supporting the operation of the feature crossbar program, as well as other software and/or programs. The network communication module is used to enable communication between the various components within the memory 1005, as well as with other hardware and software in the feature crossing system.
In the feature interleaving apparatus shown in fig. 6, the processor 1001 is configured to execute a feature interleaving program stored in the memory 1005 to implement the steps of the feature interleaving method described in any one of the above.
The specific implementation of the feature interleaving apparatus of the present application is substantially the same as the embodiments of the feature interleaving method, and is not described herein again.
The embodiment of the present application further provides a feature crossing apparatus, and the feature crossing apparatus is applied to a third-party device, the feature crossing apparatus includes:
the key management module is used for generating an encrypted public key and an encrypted private key and sending the encrypted public key to each participant device so that the first device in each participant device performs federated cross feature calculation with each second device in each participant device based on the acquired first local feature and the encrypted public key to generate an encrypted cross feature;
the decryption module is used for receiving the encrypted cross feature sent by the first equipment and decrypting the encrypted cross feature based on the encrypted private key to obtain a target cross feature;
and the evaluation module is used for evaluating the characteristic cross effect of the target cross characteristic by performing characteristic evaluation interaction with the label side equipment with the sample label information to obtain a characteristic cross effect evaluation result.
Optionally, the evaluation module is further configured to:
receiving encrypted sample label information sent by the label side equipment, wherein the encrypted sample label information is obtained by encrypting the sample label information by the label side equipment;
performing box separation on each target cross feature sample corresponding to the target cross feature to obtain each cross feature sample box separation;
counting encrypted sub-box sample label information corresponding to each cross characteristic sample sub-box based on the encrypted sample label information;
sending the label information of each encrypted sub-box sample to the label side equipment so that the label side equipment can calculate a characteristic evaluation value corresponding to each cross characteristic sample sub-box based on the label information of the encrypted sub-box sample;
and receiving each feature evaluation value sent by the tag side equipment, evaluating a feature cross effect of the target cross feature based on each feature evaluation value, and obtaining a feature cross effect evaluation result.
Optionally, the evaluation module is further configured to:
determining each first encrypted sample label value and each second encrypted sample label value respectively corresponding to each cross characteristic sample sub-box based on each sample identifier;
calculating the sum of the label values of the first encrypted samples corresponding to the cross characteristic sample sub-boxes respectively to obtain the label information of the first encrypted sub-boxes corresponding to the cross characteristic sample sub-boxes;
and calculating the sum of the label values of the second encrypted samples corresponding to the cross characteristic sample sub-boxes respectively to obtain the label information of the second encrypted sub-boxes corresponding to the cross characteristic sample sub-boxes.
Optionally, the feature crossing device is further configured to:
based on the characteristic cross effect evaluation result, performing characteristic screening on each generated target cross characteristic to obtain federal training characteristic data;
and carrying out federal learning modeling with each participant device based on the federal training characteristic data to generate a target federal model.
The specific implementation of the feature interleaving apparatus of the present application is substantially the same as the embodiments of the feature interleaving method described above, and is not described herein again.
The embodiment of the present application further provides a feature crossing apparatus, and the feature crossing apparatus is applied to a first device, the feature crossing apparatus includes:
the encryption module is used for receiving an encrypted public key sent by third-party equipment and encrypting the acquired first local feature into a first encrypted local feature based on the encrypted public key;
a receiving module, configured to receive a second local encryption feature sent by a second device, where the second device is configured to encrypt the obtained second local feature into a second encrypted local feature based on the encryption public key;
the multiplication module is used for carrying out homomorphic encryption multiplication on the first encryption local characteristic and the second encryption local characteristic to obtain an encryption cross characteristic;
and the sending module is used for sending the encrypted cross feature to the third-party equipment so that the third-party equipment can decrypt the encrypted cross feature into a target cross feature based on an encrypted private key corresponding to the encrypted public key and evaluate the feature cross effect of the target cross feature.
The specific implementation of the feature interleaving apparatus of the present application is substantially the same as the embodiments of the feature interleaving method described above, and is not described herein again.
The embodiment of the present application further provides a feature crossing apparatus, and the feature crossing apparatus is applied to the second device, the feature crossing apparatus includes:
the encryption module is used for receiving an encrypted public key sent by third-party equipment and encrypting the acquired second local feature into a second encrypted local feature based on the encrypted public key;
and the sending module is used for sending the second encrypted local feature to the first equipment so that the first equipment can calculate the encrypted cross feature based on the first encrypted local feature and the second encrypted local feature generated by local encryption.
The specific implementation of the feature interleaving apparatus of the present application is substantially the same as the embodiments of the feature interleaving method described above, and is not described herein again.
The embodiment of the present application further provides a feature crossing apparatus, and the feature crossing apparatus is applied to a label side device, the feature crossing apparatus includes:
the sending module is used for obtaining encrypted sample label information and sending the encrypted sample label information to third-party equipment so that the third-party equipment can generate encrypted sub-box sample label information corresponding to each cross characteristic sample sub-box of the obtained target cross characteristic based on the encrypted sample label information;
the decryption module is used for receiving the encrypted sub-box sample label information and decrypting the encrypted sub-box sample label information to obtain the sub-box sample label information;
the calculation module is used for calculating the characteristic evaluation value of each cross characteristic sample bin based on the label information of each bin sample;
and the feedback module is used for feeding back each characteristic evaluation value to the third-party equipment.
The specific implementation of the feature interleaving apparatus of the present application is substantially the same as the embodiments of the feature interleaving method described above, and is not described herein again.
The present application provides a readable storage medium, and the readable storage medium stores one or more programs, which can be further executed by one or more processors for implementing the steps of the feature interleaving method described in any one of the above.
The specific implementation of the readable storage medium of the present application is substantially the same as the embodiments of the feature interleaving method, and is not described herein again.
Embodiments of the present application provide a computer program product comprising one or more computer programs, which are also executable by one or more processors for implementing the steps of the feature interleaving method of any one of the above.
The specific implementation of the computer program product of the present application is substantially the same as the embodiments of the feature interleaving method, and is not described herein again.
The above description is only a preferred embodiment of the present application, and not intended to limit the scope of the present application, and all modifications of equivalent structures and equivalent processes, which are made by the contents of the specification and the drawings, or which are directly or indirectly applied to other related technical fields, are included in the scope of the present application.
Claims (11)
1. A feature crossing method applied to a third-party device, the feature crossing method comprising:
generating an encryption public key and an encryption private key, and sending the encryption public key to each participant device, so that a first device in each participant device performs federated cross feature calculation with each second device in each participant device based on the acquired first local feature and the encryption public key, and generates encryption cross features;
receiving the encrypted cross feature sent by the first equipment, and decrypting the encrypted cross feature based on the encrypted private key to obtain a target cross feature;
and performing characteristic evaluation interaction with label side equipment with sample label information, and performing characteristic cross effect evaluation on the target cross characteristics to obtain a characteristic cross effect evaluation result.
2. The feature intersection method of claim 1, wherein the step of performing feature intersection effect evaluation on the target intersection feature through feature evaluation interaction with a tagger device with sample tag information to obtain a feature intersection effect evaluation result comprises:
receiving encrypted sample label information sent by the label side equipment, wherein the encrypted sample label information is obtained by encrypting the sample label information by the label side equipment;
performing box separation on each target cross feature sample corresponding to the target cross feature to obtain each cross feature sample box separation;
counting encrypted sub-box sample label information corresponding to each cross characteristic sample sub-box based on the encrypted sample label information;
sending the label information of each encrypted sub-box sample to the label side equipment so that the label side equipment can calculate a characteristic evaluation value corresponding to each cross characteristic sample sub-box based on the label information of the encrypted sub-box sample;
and receiving each feature evaluation value sent by the tag side equipment, evaluating a feature cross effect of the target cross feature based on each feature evaluation value, and obtaining a feature cross effect evaluation result.
3. The method of claim 2, wherein the encrypted sample tag information includes at least a sample identification and a corresponding first encrypted sample tag value and a corresponding second encrypted sample tag value for the sample identification, the encrypted binned sample tag information including first encrypted binned sample tag information and second encrypted binned sample tag information,
the step of counting the encrypted sub-box sample label information corresponding to each cross feature sample sub-box based on the encrypted sample label information includes:
determining each first encrypted sample label value and each second encrypted sample label value respectively corresponding to each cross characteristic sample sub-box based on each sample identifier;
calculating the sum of the label values of the first encrypted samples corresponding to the cross characteristic sample sub-boxes respectively to obtain the label information of the first encrypted sub-boxes corresponding to the cross characteristic sample sub-boxes;
and calculating the sum of the label values of the second encrypted samples corresponding to the cross characteristic sample sub-boxes respectively to obtain the label information of the second encrypted sub-boxes corresponding to the cross characteristic sample sub-boxes.
4. The method of claim 1, wherein after the step of obtaining the cross-feature effect evaluation result by evaluating the cross-feature effect of the target cross-feature through a cross-feature evaluation interaction with a tagger device having sample tag information, the method further comprises:
based on the characteristic cross effect evaluation result, performing characteristic screening on each generated target cross characteristic to obtain federal training characteristic data;
and carrying out federal learning modeling with each participant device based on the federal training characteristic data to generate a target federal model.
5. The signature interleaving method according to claim 1, wherein the encrypted cross signature is obtained by the first device performing homomorphic cryptographic multiplication on a first encrypted local signature generated based on the first local signature and the encrypted public key, and a second encrypted local cross signature transmitted by the second device.
6. A feature crossing method applied to a first device, the feature crossing method comprising:
receiving an encrypted public key sent by third-party equipment, and encrypting the acquired first local feature into a first encrypted local feature based on the encrypted public key;
receiving a second local encryption feature sent by a second device, wherein the second device is used for encrypting the obtained second local feature into a second encrypted local feature based on the encryption public key;
performing homomorphic encryption multiplication on the first encryption local feature and the second encryption local feature to obtain an encryption cross feature;
and sending the encrypted cross feature to the third-party equipment so that the third-party equipment can decrypt the encrypted cross feature into a target cross feature based on an encrypted private key corresponding to the encrypted public key and evaluate the feature cross effect of the target cross feature.
7. A feature crossing method applied to a second device, the feature crossing method comprising:
receiving an encrypted public key sent by third-party equipment, and encrypting the acquired second local feature into a second encrypted local feature based on the encrypted public key;
and sending the second encrypted local feature to the first device, so that the first device can calculate the encrypted cross feature based on the first encrypted local feature and the second encrypted local feature generated by local encryption.
8. A feature crossing method applied to a labeler device, the feature crossing method comprising:
acquiring encrypted sample label information, and sending the encrypted sample label information to third-party equipment so that the third-party equipment can generate encrypted box sample label information corresponding to each cross characteristic sample box of the acquired target cross characteristic based on the encrypted sample label information;
receiving the label information of each encrypted sample in a sub-box, and decrypting the label information of each encrypted sample in a sub-box to obtain the label information of each sample in the sub-box;
calculating a characteristic evaluation value of each cross characteristic sample bin based on the label information of each bin sample;
feeding back each of the feature evaluation values to the third-party device.
9. A feature intersection apparatus, characterized in that the feature intersection apparatus comprises: a memory, a processor, and a program stored on the memory for implementing the feature interleaving method,
the memory is used for storing a program for realizing the feature crossing method;
the processor is configured to execute a program implementing the feature intersection method to implement the steps of the feature intersection method according to any one of claims 1 to 5 or 6 or 7 or 8.
10. A readable storage medium, characterized in that the readable storage medium has stored thereon a program for implementing a feature intersection method, the program being executed by a processor to implement the steps of the feature intersection method according to any one of claims 1 to 5 or 6 or 7 or 8.
11. A computer program product comprising a computer program, characterized in that the computer program realizes the steps of the feature intersection method of any one of claims 1 to 5 or 6 or 7 or 8 when executed by a processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011554365.0A CN112668020A (en) | 2020-12-24 | 2020-12-24 | Feature interleaving method, device, readable storage medium and computer program product |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011554365.0A CN112668020A (en) | 2020-12-24 | 2020-12-24 | Feature interleaving method, device, readable storage medium and computer program product |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112668020A true CN112668020A (en) | 2021-04-16 |
Family
ID=75410066
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011554365.0A Pending CN112668020A (en) | 2020-12-24 | 2020-12-24 | Feature interleaving method, device, readable storage medium and computer program product |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112668020A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113836559A (en) * | 2021-09-28 | 2021-12-24 | 中国银联股份有限公司 | Sample alignment method, device, equipment and storage medium in federated learning |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030158960A1 (en) * | 2000-05-22 | 2003-08-21 | Engberg Stephan J. | System and method for establishing a privacy communication path |
| CN107682380A (en) * | 2017-11-23 | 2018-02-09 | 上海众人网络安全技术有限公司 | A kind of method and device of cross-certification |
| CN111147248A (en) * | 2019-11-27 | 2020-05-12 | 北京旷视科技有限公司 | Encrypted transmission method, device and system of face feature library and storage medium |
| CN111523679A (en) * | 2020-04-26 | 2020-08-11 | 深圳前海微众银行股份有限公司 | Feature binning method, device and readable storage medium |
| CN111967615A (en) * | 2020-09-25 | 2020-11-20 | 北京百度网讯科技有限公司 | Multi-model training method and system based on feature extraction, electronic device and medium |
-
2020
- 2020-12-24 CN CN202011554365.0A patent/CN112668020A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20030158960A1 (en) * | 2000-05-22 | 2003-08-21 | Engberg Stephan J. | System and method for establishing a privacy communication path |
| CN107682380A (en) * | 2017-11-23 | 2018-02-09 | 上海众人网络安全技术有限公司 | A kind of method and device of cross-certification |
| CN111147248A (en) * | 2019-11-27 | 2020-05-12 | 北京旷视科技有限公司 | Encrypted transmission method, device and system of face feature library and storage medium |
| CN111523679A (en) * | 2020-04-26 | 2020-08-11 | 深圳前海微众银行股份有限公司 | Feature binning method, device and readable storage medium |
| CN111967615A (en) * | 2020-09-25 | 2020-11-20 | 北京百度网讯科技有限公司 | Multi-model training method and system based on feature extraction, electronic device and medium |
Non-Patent Citations (2)
| Title |
|---|
| FUGKEAW, S 等: "CLOUD-CAT: A Collaborative Access Control Tool for Data Outsourced in Cloud Computing", 《 2015 TENTH INTERNATIONAL CONFERENCE ON DIGITAL INFORMATION MANAGEMENT (ICDIM)》, 13 September 2016 (2016-09-13), pages 194 - 199 * |
| 康立;唐小虎;范佳;: "基于认证的高效公钥加密算法", 电子学报, no. 10, 15 October 2008 (2008-10-15), pages 201 - 205 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113836559A (en) * | 2021-09-28 | 2021-12-24 | 中国银联股份有限公司 | Sample alignment method, device, equipment and storage medium in federated learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111598254B (en) | Federal learning modeling method, device and readable storage medium | |
| CN109886417B (en) | Model parameter training method, device, equipment and medium based on federal learning | |
| CN110851869B (en) | Sensitive information processing method, device and readable storage medium | |
| CN113505894B (en) | Longitudinal federal learning linear regression and logistic regression model training method and device | |
| US9158925B2 (en) | Server-aided private set intersection (PSI) with data transfer | |
| CN109635462A (en) | Model parameter training method, device, equipment and medium based on federation's study | |
| Curty et al. | Comment on “Arbitrated quantum-signature scheme” | |
| Song et al. | Protection of image ROI using chaos-based encryption and DCNN-based object detection | |
| CN112926073A (en) | Federal learning modeling optimization method, apparatus, medium, and computer program product | |
| CN110519268B (en) | Voting method, device, equipment, system and storage medium based on block chain | |
| Xu et al. | Proof-carrying cloud computation: The case of convex optimization | |
| CN107204997A (en) | The method and apparatus for managing cloud storage data | |
| US20230254113A1 (en) | Correlation coefficient acquisition method, electronic device and non-transitory computer readable storage medium | |
| Wang et al. | An efficient image encryption algorithm based on a novel chaotic map | |
| CN112231309A (en) | Method, device, terminal equipment and medium for removing duplicate of longitudinal federal data statistics | |
| Nardo et al. | A reliable chaos-based cryptography using Galois field | |
| CN111523679A (en) | Feature binning method, device and readable storage medium | |
| US11101981B2 (en) | Generating a pseudorandom number based on a portion of shares used in a cryptographic operation | |
| CN112668020A (en) | Feature interleaving method, device, readable storage medium and computer program product | |
| CN110851472A (en) | Sample matching method, device and readable storage medium | |
| CN112001452B (en) | Feature selection method, device, equipment and readable storage medium | |
| CN113806760B (en) | Method, device, electronic equipment and storage medium for acquiring correlation coefficient between features | |
| CN114462626A (en) | Federal model training method and device, terminal equipment and storage medium | |
| CN113132109B (en) | Electronic deposit certificate management method and device based on block chain and electronic equipment | |
| CN112149140B (en) | Prediction method, prediction device, prediction equipment and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |