CN109299728A - Federal learning method, system and readable storage medium storing program for executing - Google Patents
Federal learning method, system and readable storage medium storing program for executing Download PDFInfo
- Publication number
- CN109299728A CN109299728A CN201810918868.8A CN201810918868A CN109299728A CN 109299728 A CN109299728 A CN 109299728A CN 201810918868 A CN201810918868 A CN 201810918868A CN 109299728 A CN109299728 A CN 109299728A
- Authority
- CN
- China
- Prior art keywords
- data terminal
- sum
- tree
- derivative
- dervative
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F18/2148—Generating training patterns; Bootstrap methods, e.g. bagging or boosting characterised by the process organisation or structure, e.g. boosting cascade
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a kind of federal learning method, system and readable storage medium storing program for executing, federal learning method carries out federal training to multi-party training sample the following steps are included: data terminal is based on gradient decline tree GBDT algorithm, to construct gradient tree-model, wherein, the data terminal be it is multiple, the gradient tree-model includes more regression trees, and the regression tree includes multiple cut-points, the training sample includes multiple features, and the feature and the cut-point correspond；The data terminal is based on the gradient tree-model, treats forecast sample and carries out associated prediction, with the predicted value of determination sample to be predicted.The present invention carries out federal training to multi-party training sample by GBDT algorithm, realizes gradient tree model foundation, by gradient tree-model, is suitable for the sweeping scene of data volume, can meet real production environment needs well；It treats forecast sample and carries out associated prediction, realize the prediction for treating forecast sample.
Description
Technical field
The present invention relates to big data processing technology fields, more particularly to federal learning method, system and readable storage medium storing program for executing.
Background technique
Currently, it is predominantly stayed in theoretical research and academic paper about the federal machine learning scheme of secret protection, root
It is found according to investigation, is limited to Form of Technique and practical application, industry is without relevant technical application at present.
Currently existing secret protection federation Learning Scheme often appears in academic paper, and more in paper is for letter
The simple structure method of single algorithm model such as logistic regression or single decision tree decision tree,
Such as ID3, C4.5.Deficiency is understood to realistic problem, more rests on theory stage, lacks the thinking to real production environment,
It is difficult to be applied directly in industry practical application scene.
Summary of the invention
The main purpose of the present invention is to provide a kind of federal learning method, system and readable storage medium storing program for executing, it is intended to solve
The technical issues of solving folk prescription or the corresponding sample training inefficiency of both sides in the prior art.
To achieve the above object, the present invention provides a kind of federal learning method, and federation's learning method includes following step
It is rapid:
Data terminal is based on gradient decline tree GBDT algorithm and carries out federal training to multi-party training sample, to construct gradient tree
Model, wherein the data terminal be it is multiple, the gradient tree-model includes more regression trees, and the regression tree includes multiple
Cut-point, the training sample include multiple features, and the feature and the cut-point correspond；
The data terminal is based on the gradient tree-model, treats forecast sample and carries out associated prediction, to be predicted with determination
The predicted value of sample.
Preferably, the multi-party training sample includes that each data terminal is stored with training sample, Ge Gesuo respectively
State training sample sample characteristics having the same.
Preferably, each data terminal is based on gradient decline tree GBDT algorithm and carries out federal instruction to multi-party training sample
Practice, includes: the step of gradient tree-model to construct
When constructing epicycle regression tree, for the node to be processed of epicycle regression tree, each data terminal is obtained by last round of
To first gradient tree-model predicted to obtain the first derivative and second dervative of the local loss function to training sample；
Each data terminal determines the corresponding segmentation point set of all partitioning schemes of the sample characteristics of itself；
Based on each cut-point in the segmentation point set, each data terminal carries out Secure and the first meter is calculated
Calculate result；
Each data terminal is led based on the single order that the cut-point of itself and first calculated result obtain being divided into left branch
The sum of the sum of first derivative of the sum of the sum of number and second dervative, right branch and second dervative；
Each data terminal to the sum of the sum of first derivative of the left branch and second dervative, the first derivative of right branch it
Summation is carried out with the data terminal for executing with the sum of second dervative where being sent to the cut-point after cryptographic operation to summarize, and is obtained
Summarized results；
The summarized results is sent to coordination terminal by the data terminal where the cut-point, for the coordination terminal
The sum of the sum of the sum of first derivative of left branch and second dervative, the first derivative of right branch are obtained after being decrypted and second order is led
The sum of number, the sum of first derivative based on the left branch and the sum of the sum of second dervative, the first derivative of right branch and second order
The sum of derivative calculates the corresponding yield value of the cut-point, calculates the best cutting point based on the yield value, and will be described optimal
Cut-point is back to corresponding first data terminal of the best cutting point；
When receiving the best cutting point, the best cutting point is sent to the second number by first data terminal
It is saved according to terminal, and the node to be processed is divided to obtain two new nodes to be processed, wherein second data
Terminal is in each data terminal for saving the data terminal of gradient tree-model.
Preferably, each data terminal obtains being divided into left point based on the cut-point of itself and first calculated result
The step of the sum of the sum of the sum of first derivative and second dervative of branch, the first derivative of right branch and the sum of second dervative includes:
Each data terminal is based on removing the segmentation in the cut-point of itself and first calculated result and each data terminal
The corresponding local of the 4th data terminal outside the corresponding third data terminal of point is compared to training sample, is obtained first and is compared
As a result；
Based on first comparison result, each data terminal obtains being divided into the sum of first derivative of left branch to be led with second order
The sum of the sum of first derivative of the sum of number, right branch and second dervative.
Preferably, each data terminal is to the sum of the sum of first derivative of the left branch and second dervative, right branch
The data terminal that executes where being sent to the cut-point after cryptographic operation of the sum of the sum of first derivative and second dervative carry out
The step of summation summarizes, and obtains summarized results include:
Each data terminal executes cryptographic operation to the sum of the sum of first derivative of the left branch, the first derivative of right branch
Obtain the first encrypted result；
Each data terminal executes cryptographic operation to the sum of the sum of second dervative of the left branch, the second dervative of right branch
Obtain the second encrypted result；
First encrypted result and the second encrypted result are carried out summation and summarized by each data terminal, obtain summarized results,
So that the summarized results is sent to coordination terminal by each data terminal, wherein first encrypted result and the second encrypted result
It is decrypted by coordinating the private key that terminal is retained.
Preferably, each data terminal by last round of obtained first gradient tree-model predicted to obtain it is local to
Before the step of first derivative and second dervative of the loss function of training sample, federation's learning method further include:
Each data terminal receives the public key that the coordination terminal is sent, so that each data terminal is to the single order of the left branch
The sum of the sum of the sum of first derivative of the sum of derivative, right branch, second dervative of the left branch, the second dervative of right branch point
It Zhi Hang not cryptographic operation.
Preferably, described when receiving the best cutting point, first data terminal is by the best cutting point
It is sent to the preservation of the second data terminal, and the step of obtaining two new nodes to be processed is divided to the node to be processed
Later, the federal learning method further include:
When generating regression tree of the new node to be processed to construct gradient tree-model, each data terminal judges that epicycle returns
Whether tree reaches leaf condition；
If so, the new node Stop node to be processed division, obtains a regression tree of gradient tree-model, it is no
Then, each data terminal is updated described in local sample data entrance to be trained using the new corresponding sample data of node to be processed
Each data terminal is predicted the loss function obtained Ben Di to training sample by last round of obtained first gradient tree-model
First derivative and second dervative the step of.
Preferably, described to be based on the gradient tree-model, it treats forecast sample and carries out associated prediction, with determination sample to be predicted
The step of this predicted value includes:
5th data terminal traverses the corresponding regression tree of the gradient tree-model, wherein the 5th data terminal is each
Possess the data terminal of gradient tree-model in data terminal；
5th data terminal by comparing the local sample to be predicted of the 5th data terminal the first data point
The second comparison result is obtained with the attribute value of current first traverse node, judges to work as described in entrance based on second comparison result
The left subtree or right subtree of preceding first traverse node are based on institute until entering the leaf node of current first traverse node
It states leaf node and obtains the first prediction result；
Or；
6th data terminal traverses the corresponding regression tree of the gradient tree-model, wherein the 6th data terminal is each
Data terminal in data terminal in addition to the 5th data terminal；
6th data terminal by the attribute value of current second traverse node of the 6th data terminal with it is described
The attribute value of current first traverse node of 5th data terminal carries out Secure calculating, obtains the second calculated result, for
Attribute value of 5th terminal based on current first traverse node of second comparison of computational results and current second time described
The attribute value of joint-running point obtains third comparison result, and the 5th data terminal is based on third comparison result judgement and enters institute
The left subtree or right subtree of current first traverse node are stated, until enter the leaf node of current first traverse node,
5th data terminal is based on being sent to the 6th data terminal after the leaf node obtains the second prediction result.
In addition, to achieve the above object, the present invention also provides a kind of system, the system comprises: memory, processor and
It is stored in the federal learning program that can be run on the memory and on the processor, federation's learning program is described
The step of described in any item federal learning methods among the above are realized when processor executes.
In addition, to achieve the above object, the present invention also provides a kind of readable storage medium storing program for executing, being deposited on the readable storage medium storing program for executing
Federal learning program is contained, federation's learning program realizes described in any item federal study among the above when being executed by processor
The step of method.
In the present invention, federal training is carried out to multi-party training sample based on gradient decline tree GBDT algorithm, to construct gradient
Tree-model realizes the sample training of the corresponding data terminal of multi-party training sample, is suitable for the sweeping scene of data volume, can be with
Meet real production environment needs well；Solve the problems, such as folk prescription or the corresponding sample training inefficiency of both sides；Pass through ladder
Tree-model is spent, forecast sample is treated and carries out associated prediction, obtains the predicted value of sample to be predicted, the pre- of forecast sample is treated in realization
It surveys.The present invention carries out federal training to multi-party training sample by GBDT algorithm, realizes gradient tree model foundation, passes through gradient tree
Model treats forecast sample and carries out associated prediction, realizes the prediction for treating forecast sample.
Detailed description of the invention
Fig. 1 is the system hardware structure schematic diagram that the embodiment of the present invention is related to；
Fig. 2 is the flow diagram of the federal learning method first embodiment of the present invention；
Fig. 3 is the flow diagram of the federal learning method second embodiment of the present invention；
Fig. 4 is the flow diagram in the federal learning method second embodiment of the present invention；
Fig. 5 is the flow diagram of the federal learning method 3rd embodiment of the present invention；
Fig. 6 is the flow diagram of the federal learning method fourth embodiment of the present invention；
Fig. 7 is the flow diagram of the present invention the 5th embodiment of federal learning method；
Fig. 8 is the flow diagram of the present invention the 5th embodiment of federal learning method.
The object of the invention is realized, the embodiments will be further described with reference to the accompanying drawings for functional characteristics and advantage.
Specific embodiment
It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not intended to limit the present invention.
As shown in Figure 1, Fig. 1 is the system structure diagram for the hardware running environment that the embodiment of the present invention is related to.
As shown in Figure 1, the system may include: processor 1001, such as CPU, network interface 1004, user interface
1003, memory 1005, communication bus 1002.Wherein, communication bus 1002 is for realizing the connection communication between these components.
User interface 1003 may include display screen (Display), input unit such as keyboard (Keyboard), optional user interface
1003 can also include standard wireline interface and wireless interface.Network interface 1004 optionally may include that the wired of standard connects
Mouth, wireless interface (such as WI-FI interface).Memory 1005 can be high speed RAM memory, be also possible to stable memory
(non-volatile memory), such as magnetic disk storage.Memory 1005 optionally can also be independently of aforementioned processor
1001 storage device.
Optionally, system can also include camera, RF (Radio Frequency, radio frequency) circuit, sensor, audio
Circuit, WiFi module etc..Certainly, system can also configure gyroscope, barometer, hygrometer, thermometer, infrared sensor etc.
Other sensors, details are not described herein.
It will be understood by those skilled in the art that the restriction of the not structure paired systems of system structure shown in Fig. 1, can wrap
It includes than illustrating more or fewer components, perhaps combines certain components or different component layouts.
As shown in Figure 1, as may include that operating system, network are logical in a kind of memory 1005 of computer storage medium
Believe module, user interface section and federal learning program.
In the system shown in figure 1, network interface 1004 is mainly used for connecting background apparatus, is counted with background server
According to communication；User interface 1003 is mainly used for connecting client (user terminal), carries out data communication with client；And processor
1001 can be used for calling the federal learning program stored in memory 1005, and execute following operation:
Data terminal is based on gradient decline tree GBDT algorithm and carries out federal training to multi-party training sample, to construct gradient tree
Model, wherein the data terminal be it is multiple, the gradient tree-model includes more regression trees, and the regression tree includes multiple
Cut-point, the training sample include multiple features, and the feature and the cut-point correspond；
The data terminal is based on the gradient tree-model, treats forecast sample and carries out associated prediction, to be predicted with determination
The predicted value of sample.
Further, processor 1001 can call the federal learning program stored in memory 1005, also execute following
Operation:
When constructing epicycle regression tree, for the node to be processed of epicycle regression tree, each data terminal is obtained by last round of
To first gradient tree-model predicted to obtain the first derivative and second dervative of the local loss function to training sample；
Each data terminal determines the corresponding segmentation point set of all partitioning schemes of the sample characteristics of itself；
Based on each cut-point in the segmentation point set, each data terminal carries out Secure and the first meter is calculated
Calculate result；
Each data terminal is led based on the single order that the cut-point of itself and first calculated result obtain being divided into left branch
The sum of the sum of first derivative of the sum of the sum of number and second dervative, right branch and second dervative；
Each data terminal to the sum of the sum of first derivative of the left branch and second dervative, the first derivative of right branch it
Summation is carried out with the data terminal for executing with the sum of second dervative where being sent to the cut-point after cryptographic operation to summarize, and is obtained
Summarized results；
The summarized results is sent to coordination terminal by the data terminal where the cut-point, for the coordination terminal
The sum of the sum of the sum of first derivative of left branch and second dervative, the first derivative of right branch are obtained after being decrypted and second order is led
The sum of number, the sum of first derivative based on the left branch and the sum of the sum of second dervative, the first derivative of right branch and second order
The sum of derivative calculates the corresponding yield value of the cut-point, calculates the best cutting point based on the yield value, and will be described optimal
Cut-point is back to corresponding first data terminal of the best cutting point；
When receiving the best cutting point, the best cutting point is sent to the second number by first data terminal
It is saved according to terminal, and the node to be processed is divided to obtain two new nodes to be processed, wherein second data
Terminal is in each data terminal for saving the data terminal of gradient tree-model.
Further, processor 1001 can call the federal learning program stored in memory 1005, also execute following
Operation:
Each data terminal is based on removing the segmentation in the cut-point of itself and first calculated result and each data terminal
The corresponding local of the 4th data terminal outside the corresponding third data terminal of point is compared to training sample, is obtained first and is compared
As a result；
Based on first comparison result, each data terminal obtains being divided into the sum of first derivative of left branch to be led with second order
The sum of the sum of first derivative of the sum of number, right branch and second dervative.
Further, processor 1001 can call the federal learning program stored in memory 1005, also execute following
Operation:
Each data terminal executes cryptographic operation to the sum of the sum of first derivative of the left branch, the first derivative of right branch
Obtain the first encrypted result；
Each data terminal executes cryptographic operation to the sum of the sum of second dervative of the left branch, the second dervative of right branch
Obtain the second encrypted result；
First encrypted result and the second encrypted result are carried out summation and summarized by each data terminal, obtain summarized results,
So that the summarized results is sent to coordination terminal by each data terminal, wherein first encrypted result and the second encrypted result
It is decrypted by coordinating the private key that terminal is retained.
Further, processor 1001 can call the federal learning program stored in memory 1005, also execute following
Operation:
Each data terminal receives the public key that the coordination terminal is sent, so that each data terminal is to the single order of the left branch
The sum of the sum of the sum of first derivative of the sum of derivative, right branch, second dervative of the left branch, the second dervative of right branch point
It Zhi Hang not cryptographic operation.
Further, processor 1001 can call the federal learning program stored in 1005, also execute following operation:
When generating regression tree of the new node to be processed to construct gradient tree-model, each data terminal judges that epicycle returns
Whether tree reaches leaf condition；
If so, the new node Stop node to be processed division, obtains a regression tree of gradient tree-model, it is no
Then, each data terminal is updated described in local sample data entrance to be trained using the new corresponding sample data of node to be processed
Each data terminal is predicted the loss function obtained Ben Di to training sample by last round of obtained first gradient tree-model
First derivative and second dervative the step of.
Further, processor 1001 can call the federal learning program stored in memory 1005, also execute following
Operation:
5th data terminal traverses the corresponding regression tree of the gradient tree-model, wherein the 5th data terminal is each
Possess the data terminal of gradient tree-model in data terminal；
5th data terminal by comparing the local sample to be predicted of the 5th data terminal the first data point
The second comparison result is obtained with the attribute value of current first traverse node, judges to work as described in entrance based on second comparison result
The left subtree or right subtree of preceding first traverse node are based on institute until entering the leaf node of current first traverse node
It states leaf node and obtains the first prediction result；
Or；
6th data terminal traverses the corresponding regression tree of the gradient tree-model, wherein the 6th data terminal is each
Data terminal in data terminal in addition to the 5th data terminal；
6th data terminal by the attribute value of current second traverse node of the 6th data terminal with it is described
The attribute value of current first traverse node of 5th data terminal carries out Secure calculating, obtains the second calculated result, for
Attribute value of 5th terminal based on current first traverse node of second comparison of computational results and current second time described
The attribute value of joint-running point obtains third comparison result, and the 5th data terminal is based on third comparison result judgement and enters institute
The left subtree or right subtree of current first traverse node are stated, until enter the leaf node of current first traverse node,
5th data terminal is based on being sent to the 6th data terminal after the leaf node obtains the second prediction result.
It is the flow diagram of the federal learning method first embodiment of the present invention referring to Fig. 2, Fig. 2.
In the first embodiment, federal learning method includes:
Step S10, data terminal is based on gradient decline tree GBDT algorithm and carries out federal training to multi-party training sample, with structure
Build gradient tree-model, wherein the data terminal be it is multiple, the gradient tree-model includes more regression trees, the regression tree
Including multiple cut-points, the training sample includes multiple features, and the feature and the cut-point correspond.
GBDT full name gradient decline tree is to the best several of true fitting of distribution inside conventional machines learning algorithm
One of algorithm, before several years ago deep learning is propagated its belief on a large scale not yet, GBDT is to yield unusually brilliant results in various contests.Reason is general
Have it is several, first is that effect is really quite well.Second is that can be used to classify can be used for returning, third is that feature can be screened.
GBDT (Gradient Boosting Decision Tree, the decision tree of iteration), is a kind of decision Tree algorithms of iteration, should
Algorithm is made of more decision trees, and the conclusion of all trees, which adds up, does final result.It at the beginning of being suggested just and SVM together
It is considered as generalization ability (generalization) stronger algorithm.In recent years more because being used to search for the engineering of sequence
It practises model and causes everybody concern.
In the present embodiment, tree GBDT algorithm is declined using gradient, federal training is carried out to multi-party training sample, to construct ladder
Spend tree-model.GBDT is mainly made of three concepts: Regression Decision Tree, i.e. DT, Gradient
Boosting, i.e. GB, Shrinkage, the important evolution branch of one of algorithm, major part source code all presses version realization at present.
Decision tree (DT, Decision Tree), is commonly referred to as regression tree in GBDT algorithm, passes through the gradient of federal training building
Tree-model includes more regression trees, and a cut-point of regression tree corresponds to a feature of training sample.
By carrying out federal training to multi-party training sample simultaneously, training effectiveness is effectively improved, data volume scale is suitable for
Big scene can be very good to meet real production environment needs.
Step S20, the data terminal are based on the gradient tree-model, treat forecast sample and carry out associated prediction, with true
The predicted value of fixed sample to be predicted.
In the present embodiment, data terminal is based on gradient tree-model, treats forecast sample and carries out associated prediction, to determine to pre-
The predicted value of test sample sheet.Input multi-party sample characteristics XownerAnd sample class label Yowner, and XOwner_o={[xi,1, xI, 2...
xi,dim], i=1 ... N, N Xowner_oSample number }, wherein dim is sample characteristics dimension size, each side sample characteristics dimension dim
Equal, each characteristic dimension meaning is consistent, such as [loan value, loan duration, debt situation].It is special to multi-party sample based on GBDT algorithm
Levy XownerAnd sample class label YownerAfter carrying out federal training, gradient tree-model is obtained, by gradient tree-model, is treated pre-
Test sample this progress associated prediction, so that it is determined that the predicted value of sample to be predicted.
In the present invention, federal training is carried out to multi-party training sample based on gradient decline tree GBDT algorithm, to construct gradient
Tree-model realizes the sample training of the corresponding data terminal of multi-party training sample, is suitable for the sweeping scene of data volume, can be with
Meet real production environment needs well；Solve the problems, such as folk prescription or the corresponding sample training inefficiency of both sides；Pass through ladder
Tree-model is spent, forecast sample is treated and carries out associated prediction, obtains the predicted value of sample to be predicted, the pre- of forecast sample is treated in realization
It surveys.The present invention carries out federal training to multi-party training sample by GBDT algorithm, realizes gradient tree model foundation, passes through gradient tree
Model treats forecast sample and carries out associated prediction, realizes the prediction for treating forecast sample.
Further, the multi-party training sample includes that each data terminal is stored with training sample respectively, each
The training sample sample characteristics having the same.
In the present embodiment, multi-party training sample includes the corresponding training sample of multiple data terminals, each training sample
Sample characteristics having the same are simultaneously stored in each data terminal local.Since data terminal has multiple, the data of each data terminal
By transversally cutting, i.e. every sample characteristics are completely existed in a copy of it data, and there is only in a copy of it data,
Cutting several pieces save parallel convenient for subsequent training.
As shown in following table one, the data that data terminal X includes are as shown in Table 1:
| ID certificate number/telephone number | Age |
| X1 | 10 |
| X2 | 20 |
| X3 | 30 |
| X4 | 40 |
| X5 | 50 |
Table one: data terminal X
It is 3 parts that table one, which is carried out transversally cutting, obtains the training sample A as shown in following table two
| ID certificate number/telephone number | Age |
| X1 | 10 |
Table two: training sample A
| ID certificate number/telephone number | Age |
| X2 | 20 |
| X3 | 30 |
Table three: training sample B
| ID certificate number/telephone number | Age |
| X4 | 40 |
| X5 | 50 |
Table four: training sample C
As shown in table two, table three and table four, transversally cutting is carried out into three parts by the data possessed data terminal X, is obtained
To training sample A, training sample B and training sample C, three training sample sample characteristics having the same, and it is maintained in number
According in terminal X.
Based on first embodiment, the second embodiment of the federal learning method of the present invention, as shown in Figure 3-4, step S10 are proposed
Include:
Step S11, when constructing epicycle regression tree, for the node to be processed of epicycle regression tree, each data terminal passes through
Last round of obtained first gradient tree-model is predicted to obtain the first derivative and two of the local loss function to training sample
Order derivative；
Step S12, each data terminal determine the corresponding segmentation point set of all partitioning schemes of the sample characteristics of itself；
Step S13, based on each cut-point in the segmentation point set, each data terminal carries out Secure and calculates
To the first calculated result；
Step S14, each data terminal obtain being divided into left branch based on the cut-point of itself and first calculated result
The sum of the sum of first derivative and second dervative, the sum of the sum of the first derivative of right branch and second dervative；
Step S15, each data terminal to the sum of the sum of first derivative of the left branch and second dervative, right branch one
Data terminal where being sent to the cut-point after the sum of the sum of order derivative and second dervative execution cryptographic operation is summed
Summarize, obtains summarized results；
Step S16, the summarized results is sent to coordination terminal by the data terminal where the cut-point, for described
Coordinate to obtain the sum of the sum of the sum of first derivative of left branch and second dervative, the first derivative of right branch after terminal is decrypted
And the sum of second dervative, the sum of first derivative based on the left branch and the sum of second dervative, the first derivative of right branch it
With and the sum of second dervative calculate the corresponding yield value of the cut-point, the best cutting point is calculated based on the yield value, and will
The best cutting point is back to corresponding first data terminal of the best cutting point；
Step S17, when receiving the best cutting point, first data terminal sends the best cutting point
It is saved to the second data terminal, and the node to be processed is divided to obtain two new nodes to be processed, wherein is described
Second data terminal is in each data terminal for saving the data terminal of gradient tree-model.
In the present embodiment, when constructing epicycle regression tree, for the node to be processed of epicycle regression tree, each data terminal is logical
Cross last round of obtained first gradient tree-model predicted to obtain the local loss function to training sample first derivative and
Second dervative.It is the t regression tree for epicycle regression tree, last round of obtained first gradient tree-model is the t-1 tree.If
T=1 then is predicted to carry out by first gradient tree-model itself pre- by last round of obtained first gradient tree-model
It surveys.
Based on each cut-point in the segmentation point set, each data terminal carries out Secure and the first meter is calculated
Calculate result.Secure is calculated can be solved using Yao Shi circuit.Yao Shi millionaires' problem is the typical case that Secure calculates
Millionaires' problem is converted to set intersection problem by using 0 coding and 1 coding by problem, is proposed a kind of based on commutative
The millionaires' problem efficient solutions of encryption function, and Security Proof has been carried out, the program is without complicated module exponent
Operation, encryption and decryption operation are O (n), and theory of communication number is 4.Certainly, other modes can also be used in the Secure calculating of this case, and
It is not limited to the mode of Yao Shi circuit, as long as can be carried out Secure calculating.
It is specific as shown in figure 4, model owning side H is equivalent to the second data terminal in the present embodiment, data owning side X1 and
Data owning side X2 is equivalent to each data terminal, and coordination side C is equivalent to coordination terminal.For each tree node node to be processed,
If not up to stopping design conditions, i.e., under the conditions of leaf, the second data terminal such as model owning side H is for each column feature, choosing
Several cut-points out, while all data owning sides i.e. data owning side X1 and data owning side X2 being notified to prepare about itself
Each column Image Segmentation Methods Based on Features point Candidate Set of data.Other are notified for each (feature, dividing candidate point) P, cut-point P place side X
Data side carries out Secure calculating (assuming that there is other N number of data participants, N number of participant both participates in the calculating of P), by meter
After calculation, other data roots according to and P the first comparison result, obtain the first derivative for being divided into left and right branch and G, second order
Encrypted result [[G]], [[H]] are sent to model owning side H summation and summarized by derivative and H, all sides, obtain (P, Sum
([[G]], Sum ([[H]]).After result etc. all candidate point P calculates, each dividing candidate point place side will be all
(Sum [[G]], Sum [[H]]) is sent to coordination side C decryption, and carries out optimum segmentation value selection, and coordination side C selects optimal point
It is after cutting value S, the value is corresponding (Sum [[G]), Sum [[H]]] partition value S place side Y is returned to, Y is data owning side herein
The corresponding P of S is sent to model owning side H by X1, Y, while increasing by two new node Node1, Node2 to be processed.
The calculation formula of yield value is as follows:
Wherein, GainFor yield value, GLFor the first derivative of left branch, GRFor the first derivative of right branch, HLFor left branch
Second dervative, HRFor the second dervative of right branch, γ is the complexity cost that new node to be processed is added and introduces.
By taking table one to four as an example, when constructing t regression tree of gradient tree, wherein t=1,2,3 ... N, first each
Data terminal, calculates the first derivative and second dervative of the respectively local loss function to training sample, and the above table one to four is
Example, obtained gradient are A (gAhA)、B(gB1hB1)、A(gB2hB2)、C(gC1hC1)、C(gC2hC2), then, in each data terminal, really
The corresponding segmentation point set of all partitioning schemes of the sample characteristics of fixed each data terminal for the above table one to table four, obtains
Cut-point PA、PBAnd Pc, such as assume PA(age≤10), PB(age≤20), PC(age≤40)；In each data terminal, notice
Other data terminal combined calculations respectively divide the sum of the first derivative of the corresponding left branch of all cut-points in point set GLWith
The sum of second dervative HL, the sum of the sum of first derivative and the second dervative of right branch；It, will be described by public key in each data terminal
The sum of the first derivative of left branch GLWith the sum of second dervative HL, right branch the sum of first derivative GRWith the sum of second dervative HR
It is sent to model owning side H after encrypting respectively and sum and summarizes, obtains summarized results (P, Sum ([[G]]), Sum ([[H]]))
It is sent to coordination terminal；Coordinate terminal all Sum ([[G]]), Sum ([[H]]) are decrypted by private key, obtains left point
The sum of first derivative of branch and the sum of the sum of second dervative, the first derivative of right branch and the sum of second dervative, and according to above-mentioned
Formula calculates the corresponding yield value G of each cut-pointain.According to yield value Gain, coordinate terminal and calculate cut-point PA、PBAnd Pc
In the best cutting point, as the best cutting point be PB, then by the best cutting point PBIt is back to PBIn the training sample B at place, instruction
Data terminal where practicing sample B is data terminal X；After the data terminal receives the best cutting point, it is sent to
First data terminal saves, and is divided to obtain two new nodes to be processed to the node to be processed, wherein described the
One data terminal is in each data terminal for saving the data terminal of gradient tree-model.
Based on second embodiment, the 3rd embodiment of the federal learning method of the present invention is proposed, as shown in figure 5, step S14 packet
It includes:
Step S141, each data terminal in the cut-point of itself and first calculated result and each data terminal based on removing
The corresponding local of the 4th data terminal outside the corresponding third data terminal of the cut-point is compared to training sample, is obtained
First comparison result；
Step S142, is based on first comparison result, each data terminal obtain being divided into left branch first derivative it
With the sum of the sum of first derivative with the sum of second dervative, right branch and second dervative.
In the present embodiment, the data terminal where each cut-point notifies other data terminals to carry out Secure calculating,
Carried out after the first calculated result is calculated in Secure in each data terminal, each data terminal based on itself cut-point and
First calculated result and the 4th data terminal in each data terminal in addition to the corresponding third data terminal of the cut-point
Corresponding local is compared to training sample, obtains the first comparison result；Based on first comparison result, each data terminal
Obtain being divided into the sum of the sum of the sum of first derivative of left branch and second dervative, the first derivative of right branch and second dervative it
With.
Based on first calculated result, is recorded as such as this dimension of monthly pay, there are three data terminal X1, be
1000,2000,3000；There are three records for the same characteristic dimension of data terminal X2, are 2000,3000,4000, it is assumed that number
Cut-point according to terminal X1 is 1500 and 2500, and the cut-point of data terminal X2 is 2500 and 3500, then for data terminal
For X1, he wish to know in all monthly pay data (union of X1 and X2 and A)<1500 and>1500 and, X1's
2500, the 3500 of 2500, X2 are similarly.It is specifically relatively three records 1000,2000 of cut-point 1500 and data terminal X1,
The union that three of 3000 and data terminal X2 records 2000,3000,4000 are formed be (1000,2000,3000,2000,
3000,4000) be compared, union be (1000,2000,3000,2000,3000,4000) in find out < 1500 and
>1500 all records, it is evident that herein<1500 be (1000) to be (2000,3000,4000), also, data greater than 1500
Terminal X2 gives data terminal X1 calculating<the cost derivative of 1500 sums>1500, and<the cost derivative of 1500 sums>1500
Ciphertext is sent to data terminal X1 by public key.
Due to data terminal X2 to data terminal X1 send be cost derivative ciphertext, by public key encryption, therefore, number
The physical record in data terminal X2 will not be obtained according to terminal X1, therefore, realizes the secret protection of data terminal X2.
Based on 3rd embodiment, the fourth embodiment of the federal learning method of the present invention is proposed, as shown in fig. 6, step S15 packet
It includes:
Step S151, each data terminal hold the sum of the sum of first derivative of the left branch, the first derivative of right branch
Row cryptographic operation obtains the first encrypted result；
Step S152, each data terminal hold the sum of the sum of second dervative of the left branch, the second dervative of right branch
Row cryptographic operation obtains the second encrypted result；
First encrypted result and the second encrypted result are carried out summation and summarized, obtained by step S153, each data terminal
Summarized results, so that the summarized results is sent to coordination terminal by each data terminal, wherein first encrypted result and
Two encrypted results are decrypted by coordinating the private key that terminal is retained.
Each data terminal to the sum of the sum of first derivative of the left branch and second dervative, the first derivative of right branch it
Summation is carried out with the data terminal for executing with the sum of second dervative where being sent to the cut-point after cryptographic operation to summarize, and is obtained
Summarized results specifically: each data terminal executes the sum of the sum of first derivative of the left branch, the first derivative of right branch
Cryptographic operation obtains the first encrypted result [[G]]；Each data terminal to the sum of second dervative of the left branch, right branch two
The sum of order derivative executes cryptographic operation and obtains the second encrypted result [[H]]；In each data terminal, by first encrypted result
[[G]] and the second encrypted result [[H]] are sent to coordination terminal.Each data terminal is by the encrypted result of G, H [[G]], [[H]]
It is sent to other data terminals and sum and summarize, obtain (P, Sum ([[G]], Sum ([[H]]).By Sum ([[G]], Sum
([[H]] is sent to coordination terminal, coordinates terminal and passes through the private key retained to the first encrypted result [[G]] and the second encrypted result
[[H]] is decrypted.
The public key sent by coordinating terminal so that each data terminal encrypted to obtain the first encrypted result and second plus
It is close to be decrypted when as a result, decrypting by the private key of coordination side, guarantee that data will not be revealed between each data terminal.
Further, before step S10, federation's learning method further include: each data terminal receives the coordination
Terminal send public key, for each data terminal to the sum of the sum of first derivative of the left branch, the first derivative of right branch,
The sum of the sum of second dervative of the left branch, second dervative of right branch execute cryptographic operation respectively.
Public key is sent to each data terminal by coordinating terminal, so that each data terminal is to the first derivative of the left branch
The sum of, the sum of the sum of the first derivative of right branch, the second dervative of the left branch, the sum of the second dervative of right branch hold respectively
Row cryptographic operation obtains the first encrypted result and the second encrypted result, realizes and coordinates terminal in Coordination Treatment to each data end
The data at end maintain secrecy.
Based on second embodiment, propose the 5th embodiment of the federal learning method of the present invention, as shown in fig. 7, step S17 it
Afterwards, federal learning method further include:
Step S18, when generating regression tree of the new node to be processed to construct gradient tree-model, each data terminal judgement
Whether epicycle regression tree reaches leaf condition；
Step S191, if so, the new node Stop node to be processed division, one for obtaining gradient tree-model return
Gui Shu；
Step S192, if it is not, each data terminal is updated locally using the new corresponding sample data of node to be processed wait instruct
Experienced sample data enters step S11.
After the data terminal receives the best cutting point, it is sent to the preservation of the first data terminal, and to described
Node to be processed is divided to obtain two new nodes to be processed, is realized the processing of a node, is being handled a section
After point, when generating regression tree of the new node to construct gradient tree-model, judge whether epicycle regression tree reaches leaf condition,
If so, Stop node divides, a regression tree of gradient tree-model is obtained, otherwise, uses the corresponding sample data of new node
It updates local sample data to be trained and enters next round node split.In general, a regression tree has multiple leaf nodes,
With the growth of regression tree, when regression tree branch, handled sample size was constantly reduced, and regression tree is to data totality pearl generation
Table degree is constantly declining, and when carrying out branch to root node, it is then processing that processing, which is whole samples, then branch down,
The sample under grouping under different grouping.In order to avoid over-fitting, we stop division under conditions of n omicronn-leaf, otherwise, will be after
It is continuous to carry out next round node split.
Based on first embodiment, the sixth embodiment of the federal learning method of the present invention is proposed, step S20 includes:
5th data terminal traverses the corresponding regression tree of the gradient tree-model, wherein the 5th data terminal is each
Possess the data terminal of gradient tree-model in data terminal；
5th data terminal by comparing the local sample to be predicted of the 5th data terminal the first data point
The second comparison result is obtained with the attribute value of current first traverse node, judges to work as described in entrance based on second comparison result
The left subtree or right subtree of preceding first traverse node are based on institute until entering the leaf node of current first traverse node
It states leaf node and obtains the first prediction result；
Or；
6th data terminal traverses the corresponding regression tree of the gradient tree-model, wherein the 6th data terminal is each
Data terminal in data terminal in addition to the 5th data terminal；
6th data terminal by the attribute value of current second traverse node of the 6th data terminal with it is described
The attribute value of current first traverse node of 5th data terminal carries out Secure calculating, obtains the second calculated result, for
Attribute value of 5th terminal based on current first traverse node of second comparison of computational results and current second time described
The attribute value of joint-running point obtains third comparison result, and the 5th data terminal is based on third comparison result judgement and enters institute
The left subtree or right subtree of current first traverse node are stated, until enter the leaf node of current first traverse node,
5th data terminal is based on being sent to the 6th data terminal after the leaf node obtains the second prediction result.
For the 5th data terminal: it is local directly directly to be predicted according to common GBDT rule, that is, traverse the process of tree.Specifically
Are as follows: the 5th data terminal is by comparing the first data point of the local sample to be predicted of the 5th data terminal and current
The attribute value of first traverse node obtains the second comparison result, enters described current first based on second comparison result judgement
The left subtree or right subtree of traverse node are based on the leaf until entering the leaf node of current first traverse node
Node obtains the first prediction result.For example, the node N, N of current traversal tree are (monthly pay < 1500), then if monthly pay 2000,
The right subtree of node N is entered, otherwise enters left subtree, gets to leaf node, the first prediction result can be obtained.
For the 6th data terminal of the data terminal in addition to the 5th data terminal, the 6th data terminal traverses institute
The corresponding regression tree of gradient tree-model is stated, the 6th data terminal is saved by current second traversal of the 6th data terminal
The attribute value of current first traverse node of the attribute value and the 5th data terminal of point carries out Secure calculating, obtains the
Two calculated results, attribute value for the 5th terminal based on current first traverse node of second comparison of computational results and
The attribute value of current second traverse node obtains third comparison result, and the 5th data terminal is based on the third and compares
As a result judgement enters the left subtree or right subtree of current first traverse node, until entering current first traverse node
Leaf node, the 5th data terminal be based on the leaf node obtain the second prediction result after be sent to the 6th data
Terminal.
For the 6th data terminal in addition to the 5th data terminal, needs to be implemented n times and predict process, N GBDT as follows
The quantity of middle decision tree.For each decision tree, it is assumed that be currently located at point node (root node for being initially positioned in tree), B will be pre-
The node node of measured data each (feature name, characteristic value) and model owning side H carry out Secure calculating, in not leak data
While, A obtains (feature name, the binary values comparison result with node), selects next branch node node_ of node
Present node node is updated to node_child by child, A, is repeated this process until going to leaf node, is obtained predicted value.
Such as: present node is monthly pay < 1500, it is assumed that there is 2 attribute values, one is monthly pay, and one is the age,
It when one wheel compares, needs to compare monthly pay and age, in order to maintain secrecy, the 5th data terminal does not need to inform the 6th data terminal
Specific age data and monthly pay data, it is assumed that subtree is the age, can set a false monthly pay data, and there are one true
Real age data, then, the 5th data terminal is all compared against two data, obtains third comparison result, based on the
The judgement of three comparison results enters the left subtree or right subtree of the present node, until entering the leaf section of the present node
Point is based on being sent to the 6th data terminal after the leaf node obtains the second prediction result in the 5th data terminal, this
When, what the 6th data terminal obtained is only the second prediction result, it can not learn other any data of the 5th data terminal, from
And realize the secrecy of prediction process.
Referring to Fig. 8, Fig. 8 is the flow diagram of the fifth embodiment of the present invention, model owning side H is equivalent to the 5th data
Terminal locally predicts that just traverses the node of oneself if it is model owning side；If being non-model owning side prediction, that is just
After calculating with model owning side's Secure, the node of model owning side, also, the data confidentiality of model owning side are traversed, only
The second prediction result is only sent to other data sides.
In addition, the embodiment of the present invention also proposes a kind of readable storage medium storing program for executing, federation is stored on the readable storage medium storing program for executing
The step of learning program, federation's learning program realizes federal learning method as described above when being executed by processor.
The specific embodiment of readable storage medium storing program for executing of the present invention and each embodiment of above-mentioned federal learning method are essentially identical,
This will not be repeated here.
It should be noted that, in this document, the terms "include", "comprise" or its any other variant are intended to non-row
His property includes, so that the process, method, article or the device that include a series of elements not only include those elements, and
And further include other elements that are not explicitly listed, or further include for this process, method, article or device institute it is intrinsic
Element.In the absence of more restrictions, the element limited by sentence "including a ...", it is not excluded that including being somebody's turn to do
There is also other identical elements in the process, method of element, article or device.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
Through the above description of the embodiments, those skilled in the art can be understood that above-described embodiment side
Method can be realized by means of software and necessary general hardware platform, naturally it is also possible to by hardware, but in many cases
The former is more preferably embodiment.Based on this understanding, technical solution of the present invention substantially in other words does the prior art
The part contributed out can be embodied in the form of software products, which is stored in one as described above
In storage medium (such as ROM/RAM, magnetic disk, CD), including some instructions are used so that appliance arrangement (it can be mobile phone,
Computer, device, air conditioner or network equipment etc.) execute method described in each embodiment of the present invention.
The above is only a preferred embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other relevant skills
Art field, is included within the scope of the present invention.
Claims (10)
1. it is a kind of federation learning method, which is characterized in that it is described federation learning method the following steps are included:
Data terminal is based on gradient decline tree GBDT algorithm and carries out federal training to multi-party training sample, to construct gradient tree mould
Type, wherein the data terminal be it is multiple, the gradient tree-model includes more regression trees, and the regression tree includes multiple points
Cutpoint, the training sample include multiple features, and the feature and the cut-point correspond；
The data terminal is based on the gradient tree-model, treats forecast sample and carries out associated prediction, with determination sample to be predicted
Predicted value.
2. federation's learning method as described in claim 1, which is characterized in that the multi-party training sample includes each number
It is stored with training sample, each training sample sample characteristics having the same respectively according to terminal.
3. federation's learning method as claimed in claim 2, which is characterized in that each data terminal is based on gradient decline tree
GBDT algorithm carries out federal training to multi-party training sample, includes: the step of gradient tree-model to construct
When constructing epicycle regression tree, for the node to be processed of epicycle regression tree, each data terminal is obtained by last round of
First gradient tree-model is predicted to obtain the first derivative and second dervative of the local loss function to training sample；
Each data terminal determines the corresponding segmentation point set of all partitioning schemes of the sample characteristics of itself；
Based on each cut-point in the segmentation point set, each data terminal carries out Secure and the first calculating knot is calculated
Fruit；
Each data terminal based on the cut-point of itself and first calculated result obtain being divided into left branch first derivative it
With the sum of the sum of first derivative with the sum of second dervative, right branch and second dervative；
Each data terminal to the sum of the sum of the sum of first derivative of the left branch and second dervative, the first derivative of right branch with
Data terminal where being sent to the cut-point after the sum of second dervative execution cryptographic operation carries out summation and summarizes, and is summarized
As a result；
The summarized results is sent to coordination terminal by the data terminal where the cut-point, for coordination terminal progress
Obtained after decryption the sum of the sum of the sum of first derivative of left branch and second dervative, the first derivative of right branch and second dervative it
With, the sum of first derivative based on the left branch and the sum of the sum of second dervative, the first derivative of right branch and second dervative
The sum of calculate the corresponding yield value of the cut-point, the best cutting point is calculated based on the yield value, and by the optimum segmentation
Point is back to corresponding first data terminal of the best cutting point；
When receiving the best cutting point, the best cutting point is sent to the second data end by first data terminal
End saves, and is divided to obtain two new nodes to be processed to the node to be processed, wherein second data terminal
For the data terminal for being used to save gradient tree-model in each data terminal.
4. federation's learning method as claimed in claim 3, which is characterized in that cut-point of each data terminal based on itself
And first calculated result obtains being divided into the sum of the sum of first derivative of left branch and second dervative, the single order of right branch is led
The step of the sum of number and the sum of second dervative includes:
Each data terminal is based on removing the cut-point pair in the cut-point of itself and first calculated result and each data terminal
The corresponding local of the 4th data terminal outside the third data terminal answered is compared to training sample, is obtained first and is compared knot
Fruit；
Based on first comparison result, each data terminal obtain being divided into the sum of first derivative of left branch and second dervative it
With the sum of the sum of the first derivative of right branch and second dervative.
5. federation's learning method as claimed in claim 4, which is characterized in that each data terminal to the left branch one
The sum of the sum of first derivative of the sum of the sum of order derivative and second dervative, right branch and second dervative are sent after executing cryptographic operation
To the data terminal where the cut-point carry out summation summarize, obtain summarized results the step of include:
Each data terminal executes cryptographic operation to the sum of the sum of first derivative of the left branch, the first derivative of right branch and obtains
First encrypted result；
Each data terminal executes cryptographic operation to the sum of the sum of second dervative of the left branch, the second dervative of right branch and obtains
Second encrypted result；
First encrypted result and the second encrypted result are carried out summation and summarized by each data terminal, obtain summarized results, for
The summarized results is sent to coordination terminal by each data terminal, wherein first encrypted result and the second encrypted result pass through
Coordinate the private key that terminal is retained to be decrypted.
6. federation's learning method as claimed in claim 5, which is characterized in that each data terminal is obtained by last round of
First gradient tree-model is predicted the step of obtaining local first derivative and second dervative to the loss function of training sample
Before, the federal learning method further include:
Each data terminal receives the public key that the coordination terminal is sent, so that each data terminal is to the first derivative of the left branch
The sum of, the sum of the sum of the first derivative of right branch, the second dervative of the left branch, the sum of the second dervative of right branch hold respectively
Row cryptographic operation.
7. federation's learning method as claimed in claim 3, which is characterized in that it is described when receiving the best cutting point,
The best cutting point is sent to the second data terminal and saved by first data terminal, and is carried out to the node to be processed
After division obtains the step of two new nodes to be processed, federation's learning method further include:
When generating regression tree of the new node to be processed to construct gradient tree-model, each data terminal judges that epicycle regression tree is
It is no to reach leaf condition；
If so, the new node Stop node to be processed division, obtains a regression tree of gradient tree-model, otherwise, respectively
Data terminal updates local sample data to be trained using the new corresponding sample data of node to be processed and enters each number
It is predicted to obtain the one of the local loss function to training sample by last round of obtained first gradient tree-model according to terminal
The step of order derivative and second dervative.
8. federation's learning method as described in claim 1, which is characterized in that it is described to be based on the gradient tree-model, it treats pre-
Test sample this progress associated prediction includes: with the step of predicted value of determination sample to be predicted
5th data terminal traverses the corresponding regression tree of the gradient tree-model, wherein the 5th data terminal is each data
Possess the data terminal of gradient tree-model in terminal；
5th data terminal by comparing the local sample to be predicted of the 5th data terminal the first data point with work as
The attribute value of preceding first traverse node obtains the second comparison result, enters described current the based on second comparison result judgement
The left subtree or right subtree of one traverse node are based on the leaf until entering the leaf node of current first traverse node
Child node obtains the first prediction result；
Or；
6th data terminal traverses the corresponding regression tree of the gradient tree-model, wherein the 6th data terminal is each data
Data terminal in terminal in addition to the 5th data terminal；
The attribute value and the described 5th that 6th data terminal passes through current second traverse node of the 6th data terminal
The attribute value of current first traverse node of data terminal carries out Secure calculating, the second calculated result is obtained, for described
Attribute value and the current second traversal section of 5th terminal based on current first traverse node of second comparison of computational results
The attribute value of point obtains third comparison result, and the 5th data terminal is based on working as described in third comparison result judgement entrance
The left subtree or right subtree of preceding first traverse node, until entering the leaf node of current first traverse node, described
5th data terminal is based on being sent to the 6th data terminal after the leaf node obtains the second prediction result.
9. a kind of system, which is characterized in that the system comprises: it memory, processor and is stored on the memory and can
The federal learning program run on the processor, federation's learning program realize such as right when being executed by the processor
It is required that the step of federal learning method described in any one of 1 to 8.
10. a kind of readable storage medium storing program for executing, which is characterized in that federal learning program is stored on the readable storage medium storing program for executing, it is described
It is realized when federal learning program is executed by processor such as the step of federal learning method described in any item of the claim 1 to 8.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810918868.8A CN109299728B (en) | 2018-08-10 | 2018-08-10 | Sample joint prediction method, system and medium based on construction of gradient tree model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810918868.8A CN109299728B (en) | 2018-08-10 | 2018-08-10 | Sample joint prediction method, system and medium based on construction of gradient tree model |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109299728A true CN109299728A (en) | 2019-02-01 |
| CN109299728B CN109299728B (en) | 2023-06-27 |
Family
ID=65170261
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810918868.8A Active CN109299728B (en) | 2018-08-10 | 2018-08-10 | Sample joint prediction method, system and medium based on construction of gradient tree model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109299728B (en) |
Cited By (55)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210233A (en) * | 2019-04-19 | 2019-09-06 | 平安科技（深圳）有限公司 | Joint mapping method, apparatus, storage medium and the computer equipment of prediction model |
| CN110443378A (en) * | 2019-08-02 | 2019-11-12 | 深圳前海微众银行股份有限公司 | Feature correlation analysis method, device and readable storage medium storing program for executing in federation's study |
| CN110443408A (en) * | 2019-07-04 | 2019-11-12 | 特斯联（北京）科技有限公司 | Travel forecasting approaches and device |
| CN110569659A (en) * | 2019-07-01 | 2019-12-13 | 阿里巴巴集团控股有限公司 | data processing method and device and electronic equipment |
| CN110598870A (en) * | 2019-09-02 | 2019-12-20 | 深圳前海微众银行股份有限公司 | Method and device for federated learning |
| CN110728317A (en) * | 2019-09-30 | 2020-01-24 | 腾讯科技（深圳）有限公司 | Training method and system of decision tree model, storage medium and prediction method |
| CN110782340A (en) * | 2019-10-25 | 2020-02-11 | 深圳前海微众银行股份有限公司 | Interactive modeling method, device and equipment of decision tree model and storage medium |
| CN110851785A (en) * | 2019-11-14 | 2020-02-28 | 深圳前海微众银行股份有限公司 | Longitudinal federated learning optimization method, device, equipment and storage medium |
| CN110851786A (en) * | 2019-11-14 | 2020-02-28 | 深圳前海微众银行股份有限公司 | Longitudinal federated learning optimization method, device, equipment and storage medium |
| CN110968886A (en) * | 2019-12-20 | 2020-04-07 | 支付宝(杭州)信息技术有限公司 | Method and system for screening training samples of machine learning model |
| CN110991552A (en) * | 2019-12-12 | 2020-04-10 | 支付宝(杭州)信息技术有限公司 | Isolated forest model construction and prediction method and device based on federal learning |
| CN110990829A (en) * | 2019-11-21 | 2020-04-10 | 支付宝(杭州)信息技术有限公司 | Method, device and equipment for training GBDT model in trusted execution environment |
| CN111046425A (en) * | 2019-12-12 | 2020-04-21 | 支付宝(杭州)信息技术有限公司 | Method and device for risk identification by combining multiple parties |
| CN111079939A (en) * | 2019-11-28 | 2020-04-28 | 支付宝(杭州)信息技术有限公司 | Machine learning model feature screening method and device based on data privacy protection |
| CN111091197A (en) * | 2019-11-21 | 2020-05-01 | 支付宝(杭州)信息技术有限公司 | Method, device and equipment for training GBDT model in trusted execution environment |
| CN111144576A (en) * | 2019-12-13 | 2020-05-12 | 支付宝(杭州)信息技术有限公司 | Model training method and device and electronic equipment |
| CN111178408A (en) * | 2019-12-19 | 2020-05-19 | 中国科学院计算技术研究所 | Health monitoring model construction method and system based on federal random forest learning |
| CN111242385A (en) * | 2020-01-19 | 2020-06-05 | 苏宁云计算有限公司 | Prediction method, device and system of gradient lifting tree model |
| CN111310933A (en) * | 2020-02-11 | 2020-06-19 | 深圳前海微众银行股份有限公司 | Feature dependence graph calculation optimization method, device and equipment and readable storage medium |
| CN111340614A (en) * | 2020-02-28 | 2020-06-26 | 深圳前海微众银行股份有限公司 | Sample sampling method and device based on federal learning and readable storage medium |
| CN111340150A (en) * | 2020-05-22 | 2020-06-26 | 支付宝(杭州)信息技术有限公司 | Method and device for training first classification model |
| CN111339275A (en) * | 2020-02-27 | 2020-06-26 | 深圳大学 | Method and device for matching answer information, server and storage medium |
| CN111353554A (en) * | 2020-05-09 | 2020-06-30 | 支付宝(杭州)信息技术有限公司 | Method and device for predicting missing user service attributes |
| CN111598186A (en) * | 2020-06-05 | 2020-08-28 | 腾讯科技（深圳）有限公司 | Decision model training method, prediction method and device based on longitudinal federal learning |
| CN111695697A (en) * | 2020-06-12 | 2020-09-22 | 深圳前海微众银行股份有限公司 | Multi-party combined decision tree construction method and device and readable storage medium |
| CN111783139A (en) * | 2020-06-29 | 2020-10-16 | 京东数字科技控股有限公司 | Federal learning classification tree construction method, model construction method and terminal equipment |
| 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 |
| CN112001500A (en) * | 2020-08-13 | 2020-11-27 | 星环信息科技（上海）有限公司 | Model training method, device and storage medium based on longitudinal federated learning system |
| CN112085758A (en) * | 2020-09-04 | 2020-12-15 | 西北工业大学 | Edge-end fused terminal context adaptive model segmentation method |
| CN112183759A (en) * | 2019-07-04 | 2021-01-05 | 创新先进技术有限公司 | Model training method, device and system |
| CN112199706A (en) * | 2020-10-26 | 2021-01-08 | 支付宝(杭州)信息技术有限公司 | Tree model training method and business prediction method based on multi-party safety calculation |
| WO2021008017A1 (en) * | 2019-07-17 | 2021-01-21 | 深圳前海微众银行股份有限公司 | Federation learning method, system, terminal device and storage medium |
| CN112308157A (en) * | 2020-11-05 | 2021-02-02 | 浙江大学 | Decision tree-oriented transverse federated learning method |
| CN112464174A (en) * | 2020-10-27 | 2021-03-09 | 华控清交信息科技(北京)有限公司 | Method and device for verifying multi-party secure computing software and device for verifying |
| CN112712182A (en) * | 2021-03-29 | 2021-04-27 | 腾讯科技（深圳）有限公司 | Model training method and device based on federal learning and storage medium |
| CN112836830A (en) * | 2021-02-01 | 2021-05-25 | 广西师范大学 | Method for voting and training in parallel by using federated gradient boosting decision tree |
| CN112884164A (en) * | 2021-03-18 | 2021-06-01 | 中国地质大学（北京） | Federal machine learning migration method and system for intelligent mobile terminal |
| CN112989399A (en) * | 2021-05-18 | 2021-06-18 | 杭州金智塔科技有限公司 | Data processing system and method |
| CN113204443A (en) * | 2021-06-03 | 2021-08-03 | 京东科技控股股份有限公司 | Data processing method, equipment, medium and product based on federal learning framework |
| CN113392101A (en) * | 2020-03-13 | 2021-09-14 | 京东城市(北京)数字科技有限公司 | Method, main server, service platform and system for constructing horizontal federated tree |
| CN113408747A (en) * | 2021-06-28 | 2021-09-17 | 淮安集略科技有限公司 | Model parameter updating method and device, computer readable medium and electronic equipment |
| CN113449880A (en) * | 2021-08-30 | 2021-09-28 | 深圳致星科技有限公司 | Heterogeneous acceleration system and method for longitudinal federated learning decision tree model |
| CN113554476A (en) * | 2020-04-23 | 2021-10-26 | 京东数字科技控股有限公司 | Training method and system of credit prediction model, electronic device and storage medium |
| CN113674843A (en) * | 2021-07-08 | 2021-11-19 | 浙江一山智慧医疗研究有限公司 | Method, device, system, electronic device and storage medium for medical expense prediction |
| CN113723477A (en) * | 2021-08-16 | 2021-11-30 | 同盾科技有限公司 | Cross-feature federal abnormal data detection method based on isolated forest |
| CN113822311A (en) * | 2020-12-31 | 2021-12-21 | 京东科技控股股份有限公司 | Method and device for training federated learning model and electronic equipment |
| CN113824546A (en) * | 2020-06-19 | 2021-12-21 | 百度在线网络技术（北京）有限公司 | Method and apparatus for generating information |
| WO2022088606A1 (en) * | 2020-10-29 | 2022-05-05 | 平安科技（深圳）有限公司 | Gbdt and lr fusion method and apparatus based on federated learning, device, and storage medium |
| WO2022151654A1 (en) * | 2021-01-14 | 2022-07-21 | 新智数字科技有限公司 | Random greedy algorithm-based horizontal federated gradient boosted tree optimization method |
| US11588621B2 (en) | 2019-12-06 | 2023-02-21 | International Business Machines Corporation | Efficient private vertical federated learning |
| CN115936112A (en) * | 2023-01-06 | 2023-04-07 | 北京国际大数据交易有限公司 | Client portrait model training method and system based on federal learning |
| CN116757286A (en) * | 2023-08-16 | 2023-09-15 | 杭州金智塔科技有限公司 | Multi-party joint causal tree model construction system and method based on federal learning |
| CN117034000A (en) * | 2023-03-22 | 2023-11-10 | 浙江明日数据智能有限公司 | Modeling method and device for longitudinal federal learning, storage medium and electronic equipment |
| CN117251805A (en) * | 2023-11-20 | 2023-12-19 | 杭州金智塔科技有限公司 | Federal gradient lifting decision tree model updating system based on breadth-first algorithm |
| CN110210233B (en) * | 2019-04-19 | 2024-05-24 | 平安科技（深圳）有限公司 | Combined construction method and device of prediction model, storage medium and computer equipment |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106204103A (en) * | 2016-06-24 | 2016-12-07 | 有米科技股份有限公司 | The method of similar users found by a kind of moving advertising platform |
| TW201732591A (en) * | 2016-01-29 | 2017-09-16 | Alibaba Group Services Ltd | Disk failure prediction method and apparatus |
| CN107563429A (en) * | 2017-07-27 | 2018-01-09 | 国家计算机网络与信息安全管理中心 | A kind of sorting technique and device of network user colony |
| CN108256052A (en) * | 2018-01-15 | 2018-07-06 | 成都初联创智软件有限公司 | Automobile industry potential customers' recognition methods based on tri-training |
| CN108364018A (en) * | 2018-01-25 | 2018-08-03 | 北京墨丘科技有限公司 | A kind of guard method of labeled data, terminal device and system |
-
2018
- 2018-08-10 CN CN201810918868.8A patent/CN109299728B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TW201732591A (en) * | 2016-01-29 | 2017-09-16 | Alibaba Group Services Ltd | Disk failure prediction method and apparatus |
| CN106204103A (en) * | 2016-06-24 | 2016-12-07 | 有米科技股份有限公司 | The method of similar users found by a kind of moving advertising platform |
| CN107563429A (en) * | 2017-07-27 | 2018-01-09 | 国家计算机网络与信息安全管理中心 | A kind of sorting technique and device of network user colony |
| CN108256052A (en) * | 2018-01-15 | 2018-07-06 | 成都初联创智软件有限公司 | Automobile industry potential customers' recognition methods based on tri-training |
| CN108364018A (en) * | 2018-01-25 | 2018-08-03 | 北京墨丘科技有限公司 | A kind of guard method of labeled data, terminal device and system |
Non-Patent Citations (3)
| Title |
|---|
| 陈启伟 等: ""基于Ext_GBDT集成的类别不平衡信用评分模型"", 《计算机应用研究》 * |
| 陈启伟 等: ""基于Ext_GBDT集成的类别不平衡信用评分模型"", 《计算机应用研究》, 28 February 2018 (2018-02-28) * |
| 陈启伟 等: ""基于Ext-GBDT集成的类别不平衡信用评分模型"", 《计算机应用研究》, pages 421 * |
Cited By (82)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110210233B (en) * | 2019-04-19 | 2024-05-24 | 平安科技（深圳）有限公司 | Combined construction method and device of prediction model, storage medium and computer equipment |
| CN110210233A (en) * | 2019-04-19 | 2019-09-06 | 平安科技（深圳）有限公司 | Joint mapping method, apparatus, storage medium and the computer equipment of prediction model |
| WO2020211240A1 (en) * | 2019-04-19 | 2020-10-22 | 平安科技（深圳）有限公司 | Joint construction method and apparatus for prediction model, and computer device |
| CN110569659A (en) * | 2019-07-01 | 2019-12-13 | 阿里巴巴集团控股有限公司 | data processing method and device and electronic equipment |
| CN112183759A (en) * | 2019-07-04 | 2021-01-05 | 创新先进技术有限公司 | Model training method, device and system |
| CN110443408A (en) * | 2019-07-04 | 2019-11-12 | 特斯联（北京）科技有限公司 | Travel forecasting approaches and device |
| CN112183759B (en) * | 2019-07-04 | 2024-02-13 | 创新先进技术有限公司 | Model training method, device and system |
| WO2021008017A1 (en) * | 2019-07-17 | 2021-01-21 | 深圳前海微众银行股份有限公司 | Federation learning method, system, terminal device and storage medium |
| CN110443378A (en) * | 2019-08-02 | 2019-11-12 | 深圳前海微众银行股份有限公司 | Feature correlation analysis method, device and readable storage medium storing program for executing in federation's study |
| CN110443378B (en) * | 2019-08-02 | 2023-11-03 | 深圳前海微众银行股份有限公司 | Feature correlation analysis method and device in federal learning and readable storage medium |
| CN110598870A (en) * | 2019-09-02 | 2019-12-20 | 深圳前海微众银行股份有限公司 | Method and device for federated learning |
| CN110598870B (en) * | 2019-09-02 | 2024-04-30 | 深圳前海微众银行股份有限公司 | Federal learning method and device |
| CN110728317A (en) * | 2019-09-30 | 2020-01-24 | 腾讯科技（深圳）有限公司 | Training method and system of decision tree model, storage medium and prediction method |
| CN110782340A (en) * | 2019-10-25 | 2020-02-11 | 深圳前海微众银行股份有限公司 | Interactive modeling method, device and equipment of decision tree model and storage medium |
| CN110851785B (en) * | 2019-11-14 | 2023-06-06 | 深圳前海微众银行股份有限公司 | Longitudinal federal learning optimization method, device, equipment and storage medium |
| CN110851786A (en) * | 2019-11-14 | 2020-02-28 | 深圳前海微众银行股份有限公司 | Longitudinal federated learning optimization method, device, equipment and storage medium |
| CN110851785A (en) * | 2019-11-14 | 2020-02-28 | 深圳前海微众银行股份有限公司 | Longitudinal federated learning optimization method, device, equipment and storage medium |
| CN110990829A (en) * | 2019-11-21 | 2020-04-10 | 支付宝(杭州)信息技术有限公司 | Method, device and equipment for training GBDT model in trusted execution environment |
| CN111091197A (en) * | 2019-11-21 | 2020-05-01 | 支付宝(杭州)信息技术有限公司 | Method, device and equipment for training GBDT model in trusted execution environment |
| CN111091197B (en) * | 2019-11-21 | 2022-03-01 | 支付宝(杭州)信息技术有限公司 | Method, device and equipment for training GBDT model in trusted execution environment |
| CN111079939A (en) * | 2019-11-28 | 2020-04-28 | 支付宝(杭州)信息技术有限公司 | Machine learning model feature screening method and device based on data privacy protection |
| US11588621B2 (en) | 2019-12-06 | 2023-02-21 | International Business Machines Corporation | Efficient private vertical federated learning |
| CN111046425A (en) * | 2019-12-12 | 2020-04-21 | 支付宝(杭州)信息技术有限公司 | Method and device for risk identification by combining multiple parties |
| CN110991552A (en) * | 2019-12-12 | 2020-04-10 | 支付宝(杭州)信息技术有限公司 | Isolated forest model construction and prediction method and device based on federal learning |
| CN111046425B (en) * | 2019-12-12 | 2021-07-13 | 支付宝(杭州)信息技术有限公司 | Method and device for risk identification by combining multiple parties |
| WO2021114821A1 (en) * | 2019-12-12 | 2021-06-17 | 支付宝(杭州)信息技术有限公司 | Isolation forest model construction and prediction method and device based on federated learning |
| CN111144576A (en) * | 2019-12-13 | 2020-05-12 | 支付宝(杭州)信息技术有限公司 | Model training method and device and electronic equipment |
| CN111178408A (en) * | 2019-12-19 | 2020-05-19 | 中国科学院计算技术研究所 | Health monitoring model construction method and system based on federal random forest learning |
| CN110968886A (en) * | 2019-12-20 | 2020-04-07 | 支付宝(杭州)信息技术有限公司 | Method and system for screening training samples of machine learning model |
| CN111242385A (en) * | 2020-01-19 | 2020-06-05 | 苏宁云计算有限公司 | Prediction method, device and system of gradient lifting tree model |
| CN111310933B (en) * | 2020-02-11 | 2024-02-02 | 深圳前海微众银行股份有限公司 | Feature dependency graph calculation optimization method, device, equipment and readable storage medium |
| CN111310933A (en) * | 2020-02-11 | 2020-06-19 | 深圳前海微众银行股份有限公司 | Feature dependence graph calculation optimization method, device and equipment and readable storage medium |
| CN111339275B (en) * | 2020-02-27 | 2023-05-12 | 深圳大学 | Answer information matching method, device, server and storage medium |
| CN111339275A (en) * | 2020-02-27 | 2020-06-26 | 深圳大学 | Method and device for matching answer information, server and storage medium |
| CN111340614A (en) * | 2020-02-28 | 2020-06-26 | 深圳前海微众银行股份有限公司 | Sample sampling method and device based on federal learning and readable storage medium |
| CN113392101A (en) * | 2020-03-13 | 2021-09-14 | 京东城市(北京)数字科技有限公司 | Method, main server, service platform and system for constructing horizontal federated tree |
| CN113554476B (en) * | 2020-04-23 | 2024-04-19 | 京东科技控股股份有限公司 | Training method and system of credit prediction model, electronic equipment and storage medium |
| CN113554476A (en) * | 2020-04-23 | 2021-10-26 | 京东数字科技控股有限公司 | Training method and system of credit prediction model, electronic device and storage medium |
| CN111353554A (en) * | 2020-05-09 | 2020-06-30 | 支付宝(杭州)信息技术有限公司 | Method and device for predicting missing user service attributes |
| CN111340150A (en) * | 2020-05-22 | 2020-06-26 | 支付宝(杭州)信息技术有限公司 | Method and device for training first classification model |
| CN111598186A (en) * | 2020-06-05 | 2020-08-28 | 腾讯科技（深圳）有限公司 | Decision model training method, prediction method and device based on longitudinal federal learning |
| CN111695697B (en) * | 2020-06-12 | 2023-09-08 | 深圳前海微众银行股份有限公司 | Multiparty joint decision tree construction method, equipment and readable storage medium |
| CN111695697A (en) * | 2020-06-12 | 2020-09-22 | 深圳前海微众银行股份有限公司 | Multi-party combined decision tree construction method and device and readable storage medium |
| WO2021249086A1 (en) * | 2020-06-12 | 2021-12-16 | 深圳前海微众银行股份有限公司 | Multi-party joint decision tree construction method, device and readable storage medium |
| CN113824546B (en) * | 2020-06-19 | 2024-04-02 | 百度在线网络技术（北京）有限公司 | Method and device for generating information |
| CN113824546A (en) * | 2020-06-19 | 2021-12-21 | 百度在线网络技术（北京）有限公司 | Method and apparatus for generating information |
| CN111783139A (en) * | 2020-06-29 | 2020-10-16 | 京东数字科技控股有限公司 | Federal learning classification tree construction method, model construction method and terminal equipment |
| CN112001500A (en) * | 2020-08-13 | 2020-11-27 | 星环信息科技（上海）有限公司 | Model training method, device and storage medium based on longitudinal federated learning system |
| CN112085758A (en) * | 2020-09-04 | 2020-12-15 | 西北工业大学 | Edge-end fused terminal context adaptive model segmentation method |
| CN112085758B (en) * | 2020-09-04 | 2022-06-24 | 西北工业大学 | Edge-end fused terminal context adaptive model segmentation method |
| CN111967615B (en) * | 2020-09-25 | 2024-05-28 | 北京百度网讯科技有限公司 | Multi-model training method and device based on feature extraction, electronic equipment and 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 |
| CN112199706B (en) * | 2020-10-26 | 2022-11-22 | 支付宝(杭州)信息技术有限公司 | Tree model training method and business prediction method based on multi-party safety calculation |
| CN112199706A (en) * | 2020-10-26 | 2021-01-08 | 支付宝(杭州)信息技术有限公司 | Tree model training method and business prediction method based on multi-party safety calculation |
| CN112464174A (en) * | 2020-10-27 | 2021-03-09 | 华控清交信息科技(北京)有限公司 | Method and device for verifying multi-party secure computing software and device for verifying |
| CN112464174B (en) * | 2020-10-27 | 2023-09-29 | 华控清交信息科技(北京)有限公司 | Method and device for verifying multi-party security computing software and device for verification |
| WO2022088606A1 (en) * | 2020-10-29 | 2022-05-05 | 平安科技（深圳）有限公司 | Gbdt and lr fusion method and apparatus based on federated learning, device, and storage medium |
| CN112308157A (en) * | 2020-11-05 | 2021-02-02 | 浙江大学 | Decision tree-oriented transverse federated learning method |
| CN113822311A (en) * | 2020-12-31 | 2021-12-21 | 京东科技控股股份有限公司 | Method and device for training federated learning model and electronic equipment |
| CN113822311B (en) * | 2020-12-31 | 2023-09-01 | 京东科技控股股份有限公司 | Training method and device of federal learning model and electronic equipment |
| WO2022151654A1 (en) * | 2021-01-14 | 2022-07-21 | 新智数字科技有限公司 | Random greedy algorithm-based horizontal federated gradient boosted tree optimization method |
| CN112836830A (en) * | 2021-02-01 | 2021-05-25 | 广西师范大学 | Method for voting and training in parallel by using federated gradient boosting decision tree |
| CN112836830B (en) * | 2021-02-01 | 2022-05-06 | 广西师范大学 | Method for voting and training in parallel by using federated gradient boosting decision tree |
| CN112884164B (en) * | 2021-03-18 | 2023-06-23 | 中国地质大学（北京） | Federal machine learning migration method and system for intelligent mobile terminal |
| CN112884164A (en) * | 2021-03-18 | 2021-06-01 | 中国地质大学（北京） | Federal machine learning migration method and system for intelligent mobile terminal |
| CN112712182B (en) * | 2021-03-29 | 2021-06-01 | 腾讯科技（深圳）有限公司 | Model training method and device based on federal learning and storage medium |
| CN112712182A (en) * | 2021-03-29 | 2021-04-27 | 腾讯科技（深圳）有限公司 | Model training method and device based on federal learning and storage medium |
| CN112989399A (en) * | 2021-05-18 | 2021-06-18 | 杭州金智塔科技有限公司 | Data processing system and method |
| CN113204443A (en) * | 2021-06-03 | 2021-08-03 | 京东科技控股股份有限公司 | Data processing method, equipment, medium and product based on federal learning framework |
| CN113204443B (en) * | 2021-06-03 | 2024-04-16 | 京东科技控股股份有限公司 | Data processing method, device, medium and product based on federal learning framework |
| CN113408747A (en) * | 2021-06-28 | 2021-09-17 | 淮安集略科技有限公司 | Model parameter updating method and device, computer readable medium and electronic equipment |
| CN113674843A (en) * | 2021-07-08 | 2021-11-19 | 浙江一山智慧医疗研究有限公司 | Method, device, system, electronic device and storage medium for medical expense prediction |
| CN113723477B (en) * | 2021-08-16 | 2024-04-30 | 同盾科技有限公司 | Cross-feature federal abnormal data detection method based on isolated forest |
| CN113723477A (en) * | 2021-08-16 | 2021-11-30 | 同盾科技有限公司 | Cross-feature federal abnormal data detection method based on isolated forest |
| CN113449880A (en) * | 2021-08-30 | 2021-09-28 | 深圳致星科技有限公司 | Heterogeneous acceleration system and method for longitudinal federated learning decision tree model |
| CN113449880B (en) * | 2021-08-30 | 2021-11-30 | 深圳致星科技有限公司 | Heterogeneous acceleration system and method for longitudinal federated learning decision tree model |
| CN115936112A (en) * | 2023-01-06 | 2023-04-07 | 北京国际大数据交易有限公司 | Client portrait model training method and system based on federal learning |
| CN117034000A (en) * | 2023-03-22 | 2023-11-10 | 浙江明日数据智能有限公司 | Modeling method and device for longitudinal federal learning, storage medium and electronic equipment |
| CN116757286B (en) * | 2023-08-16 | 2024-01-19 | 杭州金智塔科技有限公司 | Multi-party joint causal tree model construction system and method based on federal learning |
| CN116757286A (en) * | 2023-08-16 | 2023-09-15 | 杭州金智塔科技有限公司 | Multi-party joint causal tree model construction system and method based on federal learning |
| CN117251805B (en) * | 2023-11-20 | 2024-04-16 | 杭州金智塔科技有限公司 | Federal gradient lifting decision tree model updating system based on breadth-first algorithm |
| CN117251805A (en) * | 2023-11-20 | 2023-12-19 | 杭州金智塔科技有限公司 | Federal gradient lifting decision tree model updating system based on breadth-first algorithm |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109299728B (en) | 2023-06-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109299728A (en) | Federal learning method, system and readable storage medium storing program for executing | |
| Wang et al. | Fast image encryption algorithm based on parallel computing system | |
| Hesamifard et al. | Cryptodl: Deep neural networks over encrypted data | |
| CN110189192B (en) | Information recommendation model generation method and device | |
| Wan et al. | Privacy-preservation for gradient descent methods | |
| Talarposhti et al. | A secure image encryption method based on dynamic harmony search (DHS) combined with chaotic map | |
| Song et al. | Protection of image ROI using chaos-based encryption and DCNN-based object detection | |
| Cutello et al. | A hybrid immune algorithm with information gain for the graph coloring problem | |
| CN108171663B (en) | Image filling system of convolutional neural network based on feature map nearest neighbor replacement | |
| CN110443378A (en) | Feature correlation analysis method, device and readable storage medium storing program for executing in federation's study | |
| CN111428887B (en) | Model training control method, device and system based on multiple computing nodes | |
| CN107124276A (en) | A kind of safe data outsourcing machine learning data analysis method | |
| CN109687952A (en) | Data processing method and its device, electronic device and storage medium | |
| CN111143865B (en) | User behavior analysis system and method for automatically generating label on ciphertext data | |
| CN111144576A (en) | Model training method and device and electronic equipment | |
| Viard et al. | Enumerating maximal cliques in link streams with durations | |
| Qadir et al. | Digital image scrambling based on two dimensional cellular automata | |
| Hu et al. | Quantum image encryption scheme based on 2d s ine 2-l ogistic chaotic map | |
| CN114448598A (en) | Ciphertext compression method, ciphertext decompression method, device, equipment and storage medium | |
| Sanon et al. | Secure Federated Learning: An Evaluation of Homomorphic Encrypted Network Traffic Prediction | |
| Zur et al. | Comparison of two methods of adding jitter to artificial neural network training | |
| Kiran et al. | Lightweight encryption mechanism with discrete-time chaotic maps for Internet of Robotic Things | |
| US10333697B2 (en) | Nondecreasing sequence determining device, method and program | |
| CN115134078B (en) | Secret sharing-based statistical method, device and storage medium | |
| Zefreh et al. | Image security system using recursive cellular automata substitution and its parallelization |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |