CN111402095A - Method for detecting student behaviors and psychology based on homomorphic encrypted federated learning - Google Patents
Method for detecting student behaviors and psychology based on homomorphic encrypted federated learning Download PDFInfo
- Publication number
- CN111402095A CN111402095A CN202010209355.7A CN202010209355A CN111402095A CN 111402095 A CN111402095 A CN 111402095A CN 202010209355 A CN202010209355 A CN 202010209355A CN 111402095 A CN111402095 A CN 111402095A
- Authority
- CN
- China
- Prior art keywords
- data
- data set
- student
- expressing
- feature
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000006399 behavior Effects 0.000 title claims abstract description 67
- 238000000034 method Methods 0.000 title claims abstract description 49
- 238000013528 artificial neural network Methods 0.000 claims abstract description 28
- 238000005516 engineering process Methods 0.000 claims abstract description 13
- 238000012549 training Methods 0.000 claims description 24
- 230000008569 process Effects 0.000 claims description 22
- 230000006870 function Effects 0.000 claims description 13
- 238000000926 separation method Methods 0.000 claims description 11
- 230000007246 mechanism Effects 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 7
- 230000002159 abnormal effect Effects 0.000 claims description 6
- 238000004364 calculation method Methods 0.000 claims description 5
- 230000006996 mental state Effects 0.000 claims description 5
- 238000013135 deep learning Methods 0.000 claims description 3
- 230000009467 reduction Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 230000000306 recurrent effect Effects 0.000 claims description 2
- 125000004122 cyclic group Chemical group 0.000 abstract description 4
- 238000001514 detection method Methods 0.000 abstract description 4
- 238000004458 analytical method Methods 0.000 description 8
- 238000010801 machine learning Methods 0.000 description 5
- 238000007405 data analysis Methods 0.000 description 3
- 238000007418 data mining Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000011160 research Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000013527 convolutional neural network Methods 0.000 description 2
- 238000013480 data collection Methods 0.000 description 2
- 238000000354 decomposition reaction Methods 0.000 description 2
- 230000007786 learning performance Effects 0.000 description 2
- 230000004630 mental health Effects 0.000 description 2
- 230000009323 psychological health Effects 0.000 description 2
- 230000009897 systematic effect Effects 0.000 description 2
- 230000005540 biological transmission Effects 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 238000004140 cleaning Methods 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 230000008451 emotion Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- 238000013467 fragmentation Methods 0.000 description 1
- 238000006062 fragmentation reaction Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000002955 isolation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005065 mining Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000000547 structure data Methods 0.000 description 1
- 238000012546 transfer Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/10—Services
- G06Q50/20—Education
- G06Q50/205—Education administration or guidance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/602—Providing cryptographic facilities or services
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules
- G06F21/6218—Protecting access to data via a platform, e.g. using keys or access control rules to a system of files or objects, e.g. local or distributed file system or database
- G06F21/6245—Protecting personal data, e.g. for financial or medical purposes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H50/00—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics
- G16H50/30—ICT specially adapted for medical diagnosis, medical simulation or medical data mining; ICT specially adapted for detecting, monitoring or modelling epidemics or pandemics for calculating health indices; for individual health risk assessment
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L9/00—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols
- H04L9/008—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols involving homomorphic encryption
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
- H04L9/00—Cryptographic mechanisms or cryptographic arrangements for secret or secure communications; Network security protocols
- H04L9/30—Public key, i.e. encryption algorithm being computationally infeasible to invert or user's encryption keys not requiring secrecy
- H04L9/3006—Public key, i.e. encryption algorithm being computationally infeasible to invert or user's encryption keys not requiring secrecy underlying computational problems or public-key parameters
- H04L9/302—Public key, i.e. encryption algorithm being computationally infeasible to invert or user's encryption keys not requiring secrecy underlying computational problems or public-key parameters involving the integer factorization problem, e.g. RSA or quadratic sieve [QS] schemes
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Physics & Mathematics (AREA)
- Business, Economics & Management (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Computer Security & Cryptography (AREA)
- Biomedical Technology (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Bioethics (AREA)
- Mathematical Physics (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Databases & Information Systems (AREA)
- Primary Health Care (AREA)
- Signal Processing (AREA)
- Computer Hardware Design (AREA)
- Public Health (AREA)
- Computer Networks & Wireless Communication (AREA)
- Educational Administration (AREA)
- Educational Technology (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
- Pathology (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Economics (AREA)
- Epidemiology (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention provides a method for detecting student behaviors and psychology based on homomorphic encrypted federated learning, which comprises the steps of obtaining mutually independent data sets A and B; selecting intersection data through consistency of data corresponding to the same characteristic items between the data sets A and B by adopting an encryption-based user sample alignment technology, and distinguishing a data set B and a data set to be detected with the difference of the data set A; encrypting the selected intersection data of the data sets A and B by adopting a homomorphic encryption technology; a convolution cyclic neural network is constructed, intersection data of homomorphic encrypted data sets A and B are trained through federal learning, and a model for predicting the psychological state of students is obtained; and predicting the psychological state of each datum in the to-be-detected data set in a model for predicting the psychological state of the student. By implementing the invention, the requirements of student behavior and psychological detection are met on the premise of protecting data privacy, and the converged homomorphic encryption federated learning algorithm is adopted, so that the problems in the prior art are solved.
Description
Technical Field
The invention relates to the technical field of big data mining, in particular to a method for detecting student behaviors and psychology based on homomorphic encrypted federated learning.
Background
Many colleges and universities develop management and teacher-student service oriented applications by using campus behavior big data, and rely on big data mining methods to support campus management and decision making in the education field and analysis of student behavior rules. Although many data analysis methods have been proposed in recent years, the analysis of big data in colleges and universities still is a challenging research field, and there are still many problems that are worthy of further exploration and urgent solution.
At present, a large amount of teaching resources and management data are accumulated in many colleges and universities, so that a large-scale and complex-structure data set is formed, powerful support is provided for big data analysis of the colleges and universities, and the data set becomes an indispensable part of student psychological education of the colleges and universities. With the continuous advance of teaching innovation, the demand of colleges and universities on data has been shifted from the original simple transaction processing mode to information analysis processing, data mining, decision support and the like. Therefore, the system for establishing the correlation between the student behaviors and the psychological detection aiming at the existing big data set has important practical significance for the psychological education of the students in colleges and universities.
Today's big data analysis still faces two major challenges: in most industries, data exists in isolated islands; and secondly, data privacy and security are enhanced. Meanwhile, as the awareness of the compromise between data security and user privacy of large companies is increased, the importance of data privacy and security becomes a global main problem. However, we face a problem that data of college students is isolated under the premise of data security, so that student behavior data and psychological data cannot be shared, thereby resulting in prohibition of collection, fusion and use of college data for AI processing in different places. Therefore, how to legally solve the data fragmentation and isolation problems is a major challenge facing researchers.
The prior art presents one possible solution to these challenges: and (4) federal learning. Federated learning refers to a machine learning setup where multiple clients (e.g., mobile devices or entire organizations) cooperatively train models under a central server (e.g., a service provider), which setup ensures that training data is decentralized at the same time. Federal learning uses local data collection and minimization principles to reduce some of the systematic privacy risks and costs that traditional centralized machine learning approaches bring.
The technology provides a Mobile edge computing framework (MEC) capable of completing client model training while protecting client privacy, and effectively solving the problem of system heterogeneity in a real cellular network, so that the FedCS can manage client devices according to client resource conditions, thereby effectively dealing with client selection problems with resource constraints, which allows a server to aggregate as much client update information as possible and accelerate the improvement of the performance of a machine learning model, wherein the FedCS adopts classical deep neural network modeling in an experiment, and can subsequently explore more data training more complex models with, furthermore, another possible direction of future work is to deal with dynamic application scenarios, such as how to improve Federal learning performance under the conditions of average number of resources and time required for updating and uploading, and further, the FedCS also provides a Federal learning framework (Federal learning framework L) which aims to efficiently accomplish Federal learning in the framework of a heterogeneous client, and thus, the Federal learning algorithm is optimized for the dynamic learning of the client, and the Federal learning model is optimized for the dynamic learning of how to improve Federal learning performance under the conditions of the average number of resources and the dynamic fluctuation of time required for updating and uploading, and the Federal learning algorithm, thereby, and the Federal learning algorithm is optimized for the optimization of the Federal learning model under the assumption that the Federal learning model that the Federal learning algorithm is not considered to satisfy the global learning system of the dynamic learning model.
However, the two existing federal learning frameworks have disadvantages, which are: in the first federated learning framework, federated learning often faces statistical heterogeneity issues. As the global model training result of classical federated learning may tend to some update parameters uploaded by the client, the complexity calculation related to data is used for the target learning of the model, the requirement of the federated learning which is unknown in tasks is difficult to meet, and the convergence of the algorithm is uncertain under the conditions of a presumed convex loss function and a presumed set; in the second federated learning framework, on one hand, if no additional monitoring, data pool and other information exists and only one round of communication is performed, the effect of the generated global neural network model is not necessarily reliable, which leads to data increase accompanied with distributed storage, and the problem of protecting data privacy while ensuring enough data for modeling is very troublesome; on the other hand, in the federal learning problem, even if only the updated information of the model is transmitted between the client and the central server and the local data of the client is not transmitted, the risk of exposing the privacy of the user still exists, and especially under the condition that the traditional method for protecting the local privacy is too strict in practical application, the traditional method is often not applicable to the modern high-dimensional statistics and machine learning problems.
Therefore, a new federated learning framework is needed, which is suitable for student behavior and psychological detection, can overcome the problems existing in the prior art, meets the federated learning requirement of task agnostic on the premise of protecting data privacy, and enables the algorithm to converge under the condition of assuming a convex loss function and an assumption set.
Disclosure of Invention
The technical problem to be solved by the embodiments of the present invention is to provide a method for detecting student behaviors and psychology based on homomorphic encryption federal learning, which meets the requirements of student behavior and psychology detection on the premise of protecting data privacy, and the homomorphic encryption federal learning algorithm adopted is converged, thereby overcoming the problems existing in the prior art.
In order to solve the above technical problem, an embodiment of the present invention provides a method for detecting student behaviors and psychology based on homomorphic encrypted federal learning, including the following steps:
step S1, acquiring a data set A and a data set B which are independent of each other; the data set A consists of a plurality of pieces of data which are formed by adopting a first characteristic item set, and the data set B consists of a plurality of pieces of data which are formed by adopting a second characteristic item set; the first feature item set comprises one or more feature items for expressing the identity of the student and one feature item for expressing the psychological state of the student; the second characteristic item set comprises the same characteristic items corresponding to the characteristic items used for expressing the identity of the student in the first characteristic item set and at least one characteristic item used for expressing student behavior data;
step S2, selecting intersection data between the data set A and the data set B by adopting an encryption-based user sample alignment technology and according to the consistency of data corresponding to the same characteristic items between the data set A and the data set B, and distinguishing data which is different from the data set A in the data set B to form a data set to be detected;
step S3, adopting homomorphic encryption technology to encrypt the data corresponding to the feature items for expressing the psychological states of students in the selected intersection data of the data set A, and encrypt the data corresponding to the feature items for expressing the behavior data of students in the selected intersection data of the data set B;
step S4, constructing a long-time memory-based convolution cycle neural network, taking data corresponding to feature items for expressing student psychological states in intersection data selected from the homomorphic encrypted data set A as labels of the convolution cycle neural network, taking data corresponding to feature items for expressing student behavior data in intersection data selected from the homomorphic encrypted data set B as input data of the convolution cycle neural network, and training the convolution cycle neural network through federal learning to obtain a trained model for predicting student psychological states by student behaviors;
and step S5, taking each piece of data in the data set to be tested as data to be tested, sequentially inputting the data to be tested into the trained model for predicting the psychological states of students according to the behaviors of the students for calculation, and obtaining results which are respectively the psychological states predicted by the corresponding data in the data set to be tested.
In step S2, the step of using an encryption-based user sample alignment technique to sort out intersection data between the data set a and the data set B through consistency of data corresponding to the same feature items between the data set a and the data set B includes:
based on an RSA encryption mechanism, generating a public key by taking the data set B as a generator of the public key, and giving the generated public key to the data set A;
the data set A quotes a random number based on Hash, and then interactively transmits the random number to the data set B;
and the data set B is subjected to Hash at the same time and then interactively transmitted to the data set A, the data set A selects the data which are the same as the data corresponding to the same characteristic items between the data sets B and feeds the data back to the data set B, and the data set B further selects the data which are the same as the data corresponding to the same characteristic items between the data sets A, so that the intersection data between the data set A and the data set B is obtained.
In step S3, the encrypting, by using a homomorphic encryption technique, data corresponding to the feature item representing the psychological state of the student in the selected intersection data of the data set a and data corresponding to the feature item representing the behavioral data of the student in the selected intersection data of the data set B specifically includes:
on the data corresponding to the feature item for expressing the psychological state of the student in the selected intersection data of the data set A, normal data and abnormal data of the feature item for expressing the psychological state of the student are used as data of a label Y, and the label Y and the labels 1-Y are further subjected to homomorphic encryption and then are sent to the data set B;
performing box separation on data of feature items used for expressing student behavior data in the selected intersection data of the data set B, and further performing ciphertext summation operation in the box separation processing process;
after the summation operation of the ciphertext, the data set B sends the result to the data set A for decryption, and further calculates the evidence weight value and the information value of each data used for expressing the feature item of the student behavior data after the data is subjected to box separation.
In step S4, the step of training the convolutional recurrent neural network through federal learning to obtain a trained model for predicting student psychological states through student behaviors includes:
setting a collaborator C, generating a public key by taking the collaborator C as a public key generator, further distributing the public key generated by the collaborator C to the data set A and the data set B for homomorphic encryption of data to be exchanged in a training process, and performing dimension reduction and fixed input dimension on feature data after homomorphic encryption;
performing polynomial expansion on the deep learning loss function to realize the loss function in an addition form so as to encrypt corresponding data in a homomorphic manner;
taking data corresponding to feature items for expressing student psychological states in the intersection data selected from the homomorphic encrypted data set A as labels of the convolution cycle neural network and taking data corresponding to feature items for expressing student behavior data in the intersection data selected from the homomorphic encrypted data set B as input data of the convolution cycle neural network, and training the convolution cycle neural network through federal learning;
and after the training is finished, obtaining a trained model for predicting the psychological state of the student according to the behavior of the student.
Wherein the same characteristic items used for expressing the identity of the student between the first characteristic item set and the second characteristic item set comprise the school number, the gender and the age.
The second characteristic items are used for expressing student behavior data in a centralized manner, and comprise one-card consumption frequency, one-card balance, one-card consumption site, class, specialty, score, library borrowing frequency, library borrowing time and reading preference.
The embodiment of the invention has the following beneficial effects:
1. according to the invention, the fact that real data are not required to be copied to other departments when students' data are utilized to conduct behavior and psychological relevance research among college departments can be realized, so that the privacy of the students is prevented from being revealed;
2. the invention uses homomorphic encryption method and loss function polynomial decomposition, which can simplify the training process of homomorphic encryption model, and only one more encryption and decryption process is needed in the actual application process of the model;
3. compared with the existing federal learning algorithm which trains the same model in different places, the method of the invention collects the encryption gradient results and then decrypts the encryption gradient results, thereby updating the parameters of the combined model, and homomorphic federal learning of the invention only needs to train the encryption model once and then uses the encryption data for prediction, thereby greatly reducing the requirements of computing hardware in two places.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is within the scope of the present invention for those skilled in the art to obtain other drawings based on the drawings without inventive exercise.
Fig. 1 is a flowchart of a method for detecting student behavior and mind based on homomorphic encrypted federal learning according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention will be described in further detail with reference to the accompanying drawings.
As shown in fig. 1, in an embodiment of the present invention, a method for detecting behavior and mind of a student based on homomorphic encrypted federal learning includes the following steps:
step S1, acquiring a data set A and a data set B which are independent of each other; the data set A consists of a plurality of pieces of data which are formed by adopting a first characteristic item set, and the data set B consists of a plurality of pieces of data which are formed by adopting a second characteristic item set; the first feature item set comprises one or more feature items for expressing the identity of the student and one feature item for expressing the psychological state of the student; the second characteristic item set comprises the same characteristic items corresponding to the characteristic items used for expressing the identity of the student in the first characteristic item set and at least one characteristic item used for expressing student behavior data;
the specific process is that on the premise of protecting data privacy, a longitudinal federal learning algorithm is used to screen out suspected mental health abnormal groups of which the data set A and the data set B are independent of each other and have the same type of characteristics.
At this time, the first feature item set of the setting data set A comprises one or more feature items for expressing the identity of the student and one feature item for expressing the psychological state of the student; wherein, the same characteristic items for expressing the identity of the student comprise school number, name, sex, age and the like; the characteristic items used for expressing the psychological states of the students are labels;
setting a second characteristic item set of a data set B to comprise the same characteristic items corresponding to the characteristic items used for expressing the identity of the student in the first characteristic item set and at least one characteristic item used for expressing student behavior data; wherein, the same characteristic items for expressing the identity of the student comprise school number, name, sex, age and the like; the characteristic items used for expressing the student behavior data are multiple and comprise all-purpose card consumption frequency, all-purpose card balance, all-purpose card consumption site, class, specialty, score, library borrowing frequency, library borrowing time, reading hobbies and the like.
It is understood that if the same feature items in the first feature item set of the data set a for representing the identity of the student only have the school number, name, gender and age, the same feature items in the second feature item set of the data set B for representing the identity of the student also correspond to only the school number, name, gender and age. Of course, only the same academy number and name between datasets A and B can be used to achieve the correct mapping.
In one embodiment, two departments a and B in colleges and universities need to jointly establish a correlation model of student behaviors and student mental states, and the department a has a mental state label which needs to be predicted by the model, namely has a data set a; the department B has a large amount of student user data, namely a data set B, including all-purpose cards, libraries, educational administration and other relevant data which are closely related to students and represent the school behaviors of the students, and does not have corresponding psychological state labels of the students.
At this time, the department A can also collect some data related to the psychological states of students besides the basic information of the students, wherein part of information dimensions are not available for the department B, especially, the label Y value of the psychological states of the students obtained by the modeling analysis of the department A is not available for the department B, but the quantity of the samples collected by the department A is not rich enough due to the influence of various objective factors.
The information about students in department A is shown in the following table 1:
TABLE 1
At the moment, the department B can obtain information of students in various school systems after data cleaning due to the fact that a school data center is built, and the images of the students in the school can be relatively and stereoscopically depicted by means of the multidimensional information which can reflect the school behaviors of the students. For example, the one-card system is related to dimension information such as consumption frequency, money amount and places of students, the educational administration system is related to dimension information such as classes, specialties and scores of students, and the library is related to dimension information such as borrowing times, time and reading hobbies of students. On the basis, if the psychological state labels of students grasped by the department A can be matched, the three-dimensional image of the students in a school can be vivid, and suspected psychological health abnormal groups with similar characteristics can be automatically and quickly screened out, so that the department A can intervene as early as possible.
The data of students in department B in school are shown in the following table 2, wherein the dimension represented by the data characteristic Xj of department B is far more than the dimension represented by the data Xi of department A:
TABLE 2
Step S2, selecting intersection data between the data set A and the data set B by adopting an encryption-based user sample alignment technology and according to the consistency of data corresponding to the same characteristic items between the data set A and the data set B, and distinguishing data which is different from the data set A in the data set B to form a data set to be detected;
firstly, based on an RSA encryption mechanism, a public key is generated by taking a data set B as a generator of the public key, and the generated public key is given to a data set A;
then, the data set A quotes a random number based on Hash, and then interactively transmits the random number to the data set B;
and finally, the data set B is subjected to Hash simultaneously and then interactively transmitted to the data set A, the data which are the same as the data corresponding to the same characteristic items between the data set B are selected by the data set A and fed back to the data set B, the data which are the same as the data corresponding to the same characteristic items between the data set A are further selected by the data set B, intersection data between the data set A and the data set B are obtained, and meanwhile, data which are different from the data set A in the data set B are distinguished to form a data set to be detected.
It should be noted that, based on the encrypted user sample alignment technology, there is no plaintext data transmission in the whole encryption and alignment process, and even if a violent or collision manner is adopted, the original data cannot be analyzed, so that the difference set part of the two parties can be well protected by the set of mechanism.
In one embodiment, the student groups contained in the data of the two departments a and B are not completely coincident, such as some students not participating in psychological tests, or the loss of behavioral data. At the beginning of cooperation, the A department and the B department need to carry out user matching on the premise of not disclosing respective data, confirm the common student groups of the two departments and ensure that the respective student groups which are not overlapped with each other are not exposed.
Assuming that four students are available in department A [ u1, u2, u3 and u4] and four students are available in department B [ u1, u2, u3, u5 and u6], the three common student IDs [ u1, u2 and u3] are ensured to be known by both parties in the whole process, while the student B does not know that the student A has [ u4] and the student A does not know that the student B has [ u5 and u6 ]. Through the encryption mechanisms of RSA and Hash, the two parties are guaranteed to only use the intersection part finally, and the difference part is not revealed to the other party.
Through the encryption mechanisms of RSA and Hash, the two parties are guaranteed to only use the intersection part finally, and the difference part is not revealed to the other party. The method comprises the following specific steps:
step 1: based on the RSA encryption mechanism, the B department is used as a generator of the public key, and the generated public key is sent to the A department.
Step 2: and the department A quotes a random number based on the Hash and then interactively transmits the random number to the department B.
And step 3: and the department B simultaneously performs Hash and then interactively transmits the Hash to the department A, and the department A performs student ID intersection of a result.
And 4, step 4: and the obtained intersection is transmitted to the department B, and the two departments can confirm the student ID shared by the two parties.
However, there is no clear text data transfer during the whole process of student ID alignment, and even if violence or collision is adopted, the original student ID cannot be resolved.
Step S3, adopting homomorphic encryption technology to encrypt the data corresponding to the feature items for expressing the psychological states of students in the selected intersection data of the data set A, and encrypt the data corresponding to the feature items for expressing the behavior data of students in the selected intersection data of the data set B;
firstly, on data corresponding to a feature item for expressing the psychological state of a student in the selected intersection data of the data set A, normal data and abnormal data of the feature item for expressing the psychological state of the student are used as data of a label Y, and the label Y and 1-Y are subjected to homomorphic encryption and then are sent to a data set B;
secondly, performing box separation on data of feature items used for expressing student behavior data in the selected intersection data of the data set B, and further performing ciphertext summation operation in the box separation processing process;
and finally, after the ciphertext summation operation, the data set B sends the result to the data set A for decryption, and further calculates the evidence weight value and the information value of each data used for expressing the feature item of the student behavior data after the data is subjected to box separation.
In one embodiment, after the data is aligned by the user sample, some analysis and integration are often required to be performed on A, B department data characteristics due to the heterogeneity of databases among different departments. The department A can also collect some data related to the psychological states of students besides the basic information of the students, wherein part of information dimensions are not available to the department B, especially, label Y values of the psychological states of the students obtained by the modeling analysis of the department A are not available to the department B, but the quantity of the samples collected by the department A is not rich enough due to the influence of various objective factors, and the characteristic items of the samples expressed by the department B are more comprehensive, which is the reason why both the departments A, B need to model together.
After the A, B department data is cleaned and integrated, the characteristics of the union student need to be encrypted homomorphically. For example, two numbers are encrypted, after encryption, ciphertexts of the two numbers can be subjected to mathematical operation, such as addition, the result is still the ciphertexts, and the result obtained after decryption of the ciphertexts is the same as the result of addition of the ciphertexts in the plain texts. In the process of adopting the homomorphic encryption technology, the original data of each party and the data encryption state are not transmitted, so that the safety of the original data of the two parties is ensured.
Firstly, department A has Y label data representing the psychological state of students, department A encrypts Y and 1-Y homomorphically, and then gives the encrypted data to department B, department B carries out box separation processing on own characteristic data, and then department B carries out ciphertext summation operation in box separation, and then gives the result to department A for decryption, and then calculates the WOE Value (Evidence Weight) and the IV Value (Information Value or Information quantity) of each characteristic box of department B. In the process, no plaintext data is transmitted, the Y value of the A department is unknown to the B department, and meanwhile, the A department does not know what the value of each feature of the B department is, so that the calculation of the feature engineering is completed under the condition of safety and privacy protection.
Step S4, constructing a long-time memory-based convolution cycle neural network, taking data corresponding to feature items for expressing student psychological states in intersection data selected from the homomorphic encrypted data set A as labels of the convolution cycle neural network, taking data corresponding to feature items for expressing student behavior data in intersection data selected from the homomorphic encrypted data set B as input data of the convolution cycle neural network, and training the convolution cycle neural network through federal learning to obtain a trained model for predicting student psychological states by student behaviors;
the specific process comprises the following steps of firstly constructing a convolution cyclic neural network based on long-time memory, wherein L STM extraction time characteristics are added after CNN of space characteristics is extracted in consideration of the time sequence characteristics of student behavior data, and the specific structure of the convolution cyclic neural network is shown in the following table 3:
TABLE 3
First block | [(3×3)×16]×2 |
Second block | [(3×3)×32]×2 |
Third block | [(3×3)×64]×3 |
The fourth block | 256 LSTM |
Fifth piece | 256 LSTM |
The sixth block | 512 FC |
Seventh Block | 128 FC |
Output of | 1 FC |
In table 3, the three convolutions are explained as [ (number of convolution kernels × convolution kernels wide) × number of convolution kernels ] × number of times of convolution repetition, wherein a maximum pooling layer of 2 × 2 exists between every two convolution blocks, the fourth and fifth blocks are one-way L STM, a process of combining width features and channel features exists between CNN and L STM, only the features of a time dimension are reserved, and the sixth and seventh blocks are full connection layers and serve as nonlinear regressors of a network.
Secondly, training the convolution cyclic neural network through federal learning, and specifically comprising the following steps:
setting a collaborator C, generating a public key by taking the collaborator C as a public key generator, further distributing the public key generated by the collaborator C to the data set A and the data set B, performing homomorphic encryption on data to be exchanged in a training process, and performing dimension reduction and fixed input dimension on feature data subjected to homomorphic encryption;
performing polynomial expansion on the deep learning loss function to realize the loss function in an addition form so as to encrypt corresponding data in a homomorphic manner;
taking data corresponding to characteristic items for expressing the psychological states of students in the intersection data selected from the homomorphic encrypted data set A as labels of a convolution cycle neural network and taking data corresponding to characteristic items for expressing student behavior data in the intersection data selected from the homomorphic encrypted data set B as input data of the convolution cycle neural network, and training the convolution cycle neural network through federal learning;
and after the training is finished, obtaining a trained model for predicting the psychological state of the student according to the behavior of the student.
It should be noted that the encryption training by the third party collaborator C is performed to ensure the confidentiality of data during the training process, and the addition operation is satisfied by performing polynomial expansion on the loss function and the gradient, so that the homomorphic encryption technology can be applied to the loss function and the gradient.
In one embodiment, the data on two sides have the same ID, the characteristics are not identical, and the defects of the characteristics of the other side can be overcome through the characteristics of one side, the department A and the department B can train a student behavior and psychological correlation model based on homomorphic encryption federal learning under the condition that the respective data are kept in the local and the data privacy is not leaked due to data interaction in training, and the final model can achieve the following effects that the department A can utilize the data characteristics of the student behavior of the department B to analyze and evaluate the psychological state of students more comprehensively and more accurately and further optimize the prediction effect of the model, the department B can utilize labeled generalized student data collected by a large data platform to carry out data ET L work to carry out special analysis, increase the depiction of the psychological state of the students on the existing student behavior data structure to realize diversification of data dimension, and both sides can utilize the data of respective difference set part to predict the state of the student, enrich the data structure and fully play the common psychological prediction function of the model.
And step S5, taking each piece of data in the data set to be tested as data to be tested, sequentially inputting the data to be tested into the trained model for predicting the psychological states of students according to the behaviors of the students for calculation, and obtaining results which are respectively the psychological states predicted by the corresponding data in the data set to be tested.
The specific process is that the data containing behavior data but not containing the mental state characteristic items can be used as the data to be tested to predict the corresponding mental state. And selecting any data in the data set to be tested as the data to be tested, inputting the data to be tested into a trained model for predicting the psychological state of the student according to the behavior of the student, and calculating to obtain a result, wherein the result is the predicted psychological state of the data set to be tested.
Therefore, the embodiment of the invention aims to collect the mental health problems of students for clustering, screen suspected abnormal psychological health groups with similar characteristics by using a longitudinal federated learning algorithm on the premise of protecting data privacy, then carry out data ET L (Extract-Transform-L oad) work on the groups according to labeled generalized student data collected from a big data platform, and finally carry out thematic analysis and draw a conclusion.
In conclusion, the basic goal and the greatest advantage of the federal learning are protection of user privacy data, and parameters of a terminal model are transmitted to the cloud instead of terminal data information. Federal learning uses local data collection and minimization principles to reduce some of the systematic privacy risks and costs that traditional centralized machine learning approaches bring.
Meanwhile, the position-track information and the like generated by the data condition of the campus wireless network connected by students and the consumption record of the one-card are analyzed, the behavior patterns of student groups are classified and analyzed through a special group mining algorithm, the special characteristics of various students are researched, and the student management staff can conveniently carry out different and targeted education management work; through collecting postings and comments of students on social media, the emotion polarity classification is carried out, whether positive or negative influences exist is reasonably and accurately judged, thought dynamics of the students are concerned, network behavior patterns of college students are known, the students who frequently make negative evaluations for a long time pay attention in time, the thought dynamics is known, and correct guidance is carried out.
The embodiment of the invention has the following beneficial effects:
1. according to the invention, the fact that real data are not required to be copied to other departments when students' data are utilized to conduct behavior and psychological relevance research among college departments can be realized, so that the privacy of the students is prevented from being revealed;
2. the invention uses homomorphic encryption method and loss function polynomial decomposition, which can simplify the training process of homomorphic encryption model, and only one more encryption and decryption process is needed in the actual application process of the model;
3. compared with the existing federal learning algorithm which trains the same model in different places, the method of the invention collects the encryption gradient results and then decrypts the encryption gradient results, thereby updating the parameters of the combined model, and homomorphic federal learning of the invention only needs to train the encryption model once and then uses the encryption data for prediction, thereby greatly reducing the requirements of computing hardware in two places.
It will be understood by those skilled in the art that all or part of the steps in the method for implementing the above embodiments may be implemented by relevant hardware instructed by a program, and the program may be stored in a computer-readable storage medium, such as ROM/RAM, magnetic disk, optical disk, etc.
While the invention has been described in connection with what is presently considered to be the most practical and preferred embodiment, it is to be understood that the invention is not to be limited to the disclosed embodiment, but on the contrary, is intended to cover various modifications and equivalent arrangements included within the spirit and scope of the appended claims.
Claims (6)
1. A method for detecting student behaviors and psychology based on homomorphic encrypted federated learning is characterized by comprising the following steps:
step S1, acquiring a data set A and a data set B which are independent of each other; the data set A consists of a plurality of pieces of data which are formed by adopting a first characteristic item set, and the data set B consists of a plurality of pieces of data which are formed by adopting a second characteristic item set; the first feature item set comprises one or more feature items for expressing the identity of the student and one feature item for expressing the psychological state of the student; the second characteristic item set comprises the same characteristic items corresponding to the characteristic items used for expressing the identity of the student in the first characteristic item set and at least one characteristic item used for expressing student behavior data;
step S2, selecting intersection data between the data set A and the data set B by adopting an encryption-based user sample alignment technology and according to the consistency of data corresponding to the same characteristic items between the data set A and the data set B, and distinguishing data which is different from the data set A in the data set B to form a data set to be detected;
step S3, adopting homomorphic encryption technology to encrypt the data corresponding to the feature items for expressing the psychological states of students in the selected intersection data of the data set A, and encrypt the data corresponding to the feature items for expressing the behavior data of students in the selected intersection data of the data set B;
step S4, constructing a long-time memory-based convolution cycle neural network, taking data corresponding to feature items for expressing student psychological states in intersection data selected from the homomorphic encrypted data set A as labels of the convolution cycle neural network, taking data corresponding to feature items for expressing student behavior data in intersection data selected from the homomorphic encrypted data set B as input data of the convolution cycle neural network, and training the convolution cycle neural network through federal learning to obtain a trained model for predicting student psychological states by student behaviors;
and step S5, taking each piece of data in the data set to be tested as data to be tested, sequentially inputting the data to be tested into the trained model for predicting the psychological states of students according to the behaviors of the students for calculation, and obtaining results which are respectively the psychological states predicted by the corresponding data in the data set to be tested.
2. The method for detecting student behavior and mind based on homomorphic encryption federated learning of claim 1, wherein in the step S2, the step of sorting out intersection data between the data set a and the data set B by consistency of data corresponding to the same feature items between the data set a and the data set B by using the encryption-based user sample alignment technique includes:
based on an RSA encryption mechanism, generating a public key by taking the data set B as a generator of the public key, and giving the generated public key to the data set A;
the data set A quotes a random number based on Hash, and then interactively transmits the random number to the data set B;
and the data set B is subjected to Hash at the same time and then interactively transmitted to the data set A, the data set A selects the data which are the same as the data corresponding to the same characteristic items between the data sets B and feeds the data back to the data set B, and the data set B further selects the data which are the same as the data corresponding to the same characteristic items between the data sets A, so that the intersection data between the data set A and the data set B is obtained.
3. The method for detecting student behavior and mind based on homomorphic encryption federal learning as claimed in claim 1, wherein in step S3, the step of encrypting the data corresponding to the feature item for expressing student mental state in the selected intersection data of the data set a and encrypting the data corresponding to the feature item for expressing student behavior data in the selected intersection data of the data set B by using homomorphic encryption technology comprises the following specific steps:
on the data corresponding to the feature item for expressing the psychological state of the student in the selected intersection data of the data set A, normal data and abnormal data of the feature item for expressing the psychological state of the student are used as data of a label Y, and the label Y and the labels 1-Y are further subjected to homomorphic encryption and then are sent to the data set B;
performing box separation on data of feature items used for expressing student behavior data in the selected intersection data of the data set B, and further performing ciphertext summation operation in the box separation processing process;
after the summation operation of the ciphertext, the data set B sends the result to the data set A for decryption, and further calculates the evidence weight value and the information value of each data used for expressing the feature item of the student behavior data after the data is subjected to box separation.
4. The method according to claim 1, wherein in step S4, the step of training the convolutional recurrent neural network through federal learning to obtain the trained model for predicting the psychological state of the student based on the behavior of the student includes:
setting a collaborator C, generating a public key by taking the collaborator C as a public key generator, further distributing the public key generated by the collaborator C to the data set A and the data set B for homomorphic encryption of data to be exchanged in a training process, and performing dimension reduction and fixed input dimension on feature data after homomorphic encryption;
performing polynomial expansion on the deep learning loss function to realize the loss function in an addition form so as to encrypt corresponding data in a homomorphic manner;
taking data corresponding to feature items for expressing student psychological states in the intersection data selected from the homomorphic encrypted data set A as labels of the convolution cycle neural network and taking data corresponding to feature items for expressing student behavior data in the intersection data selected from the homomorphic encrypted data set B as input data of the convolution cycle neural network, and training the convolution cycle neural network through federal learning;
and after the training is finished, obtaining a trained model for predicting the psychological state of the student according to the behavior of the student.
5. The method for testing student behavior and psychology based on homomorphic encrypted federated learning of claim 1, wherein the same feature items used to express student identity between the first set of feature items and the second set of feature items include school number, gender, and age.
6. The method for testing student behaviors and psychology based on homomorphic encryption federal learning as claimed in claim 5, wherein the second feature item set comprises a plurality of feature items for expressing student behavior data, including consumption frequency of one-card, balance of one-card, consumption field of one-card, class, specialty, score, borrowing times of library, borrowing time of library and reading hobbies.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010209355.7A CN111402095A (en) | 2020-03-23 | 2020-03-23 | Method for detecting student behaviors and psychology based on homomorphic encrypted federated learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010209355.7A CN111402095A (en) | 2020-03-23 | 2020-03-23 | Method for detecting student behaviors and psychology based on homomorphic encrypted federated learning |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111402095A true CN111402095A (en) | 2020-07-10 |
Family
ID=71431095
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010209355.7A Pending CN111402095A (en) | 2020-03-23 | 2020-03-23 | Method for detecting student behaviors and psychology based on homomorphic encrypted federated learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111402095A (en) |
Cited By (22)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111914277A (en) * | 2020-08-07 | 2020-11-10 | 平安科技(深圳)有限公司 | Intersection data generation method and federal model training method based on intersection data |
CN111915019A (en) * | 2020-08-07 | 2020-11-10 | 平安科技(深圳)有限公司 | Federal learning method, system, computer device, and storage medium |
CN111935168A (en) * | 2020-08-19 | 2020-11-13 | 四川大学 | Industrial information physical system-oriented intrusion detection model establishing method |
CN112001452A (en) * | 2020-08-27 | 2020-11-27 | 深圳前海微众银行股份有限公司 | Feature selection method, device, equipment and readable storage medium |
CN112185558A (en) * | 2020-09-22 | 2021-01-05 | 珠海中科先进技术研究院有限公司 | Mental health and rehabilitation evaluation method, device and medium based on deep learning |
CN112182598A (en) * | 2020-09-14 | 2021-01-05 | 天冕信息技术(深圳)有限公司 | Public sample ID identification method, device, server and readable storage medium |
CN112201342A (en) * | 2020-09-27 | 2021-01-08 | 博雅正链(北京)科技有限公司 | Medical auxiliary diagnosis method, device, equipment and storage medium based on federal learning |
CN112289448A (en) * | 2020-11-06 | 2021-01-29 | 新智数字科技有限公司 | Health risk prediction method and device based on joint learning |
CN112348192A (en) * | 2020-09-18 | 2021-02-09 | 同盾控股有限公司 | Knowledge reasoning method, system, device and medium based on knowledge federation |
CN112364255A (en) * | 2020-11-05 | 2021-02-12 | 天津大学 | Student risk early warning model establishing technology based on social network |
CN112530546A (en) * | 2020-12-14 | 2021-03-19 | 重庆邮电大学 | Psychological pre-judging method and system based on K-means clustering and XGboost algorithm |
CN112668472A (en) * | 2020-12-28 | 2021-04-16 | 中国科学院自动化研究所 | Iris image feature extraction method, system and device based on federal learning |
CN112714221A (en) * | 2020-12-24 | 2021-04-27 | 平安科技(深圳)有限公司 | Method, system and related equipment for detecting intelligent voice of federated |
CN112910624A (en) * | 2021-01-14 | 2021-06-04 | 东北大学 | Ciphertext prediction method based on homomorphic encryption |
CN112951433A (en) * | 2021-04-06 | 2021-06-11 | 重庆邮电大学 | Mental health assessment method based on privacy calculation |
CN112966233A (en) * | 2021-02-23 | 2021-06-15 | 杭州安恒信息技术股份有限公司 | User risk operation detection method and device and computer equipment |
CN113326764A (en) * | 2021-05-27 | 2021-08-31 | 北京百度网讯科技有限公司 | Method and device for training image recognition model and image recognition |
CN113901501A (en) * | 2021-10-20 | 2022-01-07 | 苏州斐波那契信息技术有限公司 | Private domain user image expansion method based on federal learning |
CN114257412A (en) * | 2021-11-19 | 2022-03-29 | 西安电子科技大学 | Privacy protection multi-party data cooperation box-separating method, system, equipment and terminal |
CN115906162A (en) * | 2022-11-17 | 2023-04-04 | 重庆邮电大学 | Privacy protection method based on heterogeneous representation and federal factorization machine |
CN117236420A (en) * | 2023-11-14 | 2023-12-15 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Method and system for debugging vertical federation learning abnormal data based on data subset |
CN117278199A (en) * | 2023-10-18 | 2023-12-22 | 上海零数众合信息科技有限公司 | Federal learning feature screening method and system based on homomorphic encryption |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109256192A (en) * | 2018-07-24 | 2019-01-22 | 同济大学 | A kind of undergraduate psychological behavior unusual fluctuation monitoring and pre-alarming method neural network based |
CN109712718A (en) * | 2018-12-29 | 2019-05-03 | 深圳阿塔基科技有限公司 | Method, apparatus and storage medium based on artificial intelligence analysis's students psychology |
CN110111894A (en) * | 2019-05-15 | 2019-08-09 | 杭州华网信息技术有限公司 | The mental health method for early warning of behavior is serialized based on deep learning student |
CN110633806A (en) * | 2019-10-21 | 2019-12-31 | 深圳前海微众银行股份有限公司 | Longitudinal federated learning system optimization method, device, equipment and readable storage medium |
WO2020029590A1 (en) * | 2018-08-10 | 2020-02-13 | 深圳前海微众银行股份有限公司 | Sample prediction method and device based on federated training, and storage medium |
CN110874649A (en) * | 2020-01-16 | 2020-03-10 | 支付宝(杭州)信息技术有限公司 | State machine-based federal learning method, system, client and electronic equipment |
-
2020
- 2020-03-23 CN CN202010209355.7A patent/CN111402095A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109256192A (en) * | 2018-07-24 | 2019-01-22 | 同济大学 | A kind of undergraduate psychological behavior unusual fluctuation monitoring and pre-alarming method neural network based |
WO2020029590A1 (en) * | 2018-08-10 | 2020-02-13 | 深圳前海微众银行股份有限公司 | Sample prediction method and device based on federated training, and storage medium |
CN109712718A (en) * | 2018-12-29 | 2019-05-03 | 深圳阿塔基科技有限公司 | Method, apparatus and storage medium based on artificial intelligence analysis's students psychology |
CN110111894A (en) * | 2019-05-15 | 2019-08-09 | 杭州华网信息技术有限公司 | The mental health method for early warning of behavior is serialized based on deep learning student |
CN110633806A (en) * | 2019-10-21 | 2019-12-31 | 深圳前海微众银行股份有限公司 | Longitudinal federated learning system optimization method, device, equipment and readable storage medium |
CN110874649A (en) * | 2020-01-16 | 2020-03-10 | 支付宝(杭州)信息技术有限公司 | State machine-based federal learning method, system, client and electronic equipment |
Non-Patent Citations (1)
Title |
---|
FEDAI联邦学习: "只看这一篇就够了:快速了解联邦学习技术及应用实践", 《HTTPS://ZHUANLAN.ZHIHU.COM/P/81827262/》 * |
Cited By (31)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111915019B (en) * | 2020-08-07 | 2023-06-20 | 平安科技(深圳)有限公司 | Federal learning method, system, computer device, and storage medium |
CN111915019A (en) * | 2020-08-07 | 2020-11-10 | 平安科技(深圳)有限公司 | Federal learning method, system, computer device, and storage medium |
CN111914277A (en) * | 2020-08-07 | 2020-11-10 | 平安科技(深圳)有限公司 | Intersection data generation method and federal model training method based on intersection data |
CN111914277B (en) * | 2020-08-07 | 2023-09-01 | 平安科技(深圳)有限公司 | Intersection data generation method and federal model training method based on intersection data |
CN111935168A (en) * | 2020-08-19 | 2020-11-13 | 四川大学 | Industrial information physical system-oriented intrusion detection model establishing method |
CN112001452A (en) * | 2020-08-27 | 2020-11-27 | 深圳前海微众银行股份有限公司 | Feature selection method, device, equipment and readable storage medium |
CN112182598A (en) * | 2020-09-14 | 2021-01-05 | 天冕信息技术(深圳)有限公司 | Public sample ID identification method, device, server and readable storage medium |
CN112348192A (en) * | 2020-09-18 | 2021-02-09 | 同盾控股有限公司 | Knowledge reasoning method, system, device and medium based on knowledge federation |
CN112185558A (en) * | 2020-09-22 | 2021-01-05 | 珠海中科先进技术研究院有限公司 | Mental health and rehabilitation evaluation method, device and medium based on deep learning |
CN112201342A (en) * | 2020-09-27 | 2021-01-08 | 博雅正链(北京)科技有限公司 | Medical auxiliary diagnosis method, device, equipment and storage medium based on federal learning |
CN112201342B (en) * | 2020-09-27 | 2024-04-26 | 博雅正链(北京)科技有限公司 | Medical auxiliary diagnosis method, device, equipment and storage medium based on federal learning |
CN112364255A (en) * | 2020-11-05 | 2021-02-12 | 天津大学 | Student risk early warning model establishing technology based on social network |
CN112289448A (en) * | 2020-11-06 | 2021-01-29 | 新智数字科技有限公司 | Health risk prediction method and device based on joint learning |
CN112530546A (en) * | 2020-12-14 | 2021-03-19 | 重庆邮电大学 | Psychological pre-judging method and system based on K-means clustering and XGboost algorithm |
CN112530546B (en) * | 2020-12-14 | 2024-03-22 | 重庆邮电大学 | Psychological pre-judging method and system based on K-means clustering and XGBoost algorithm |
CN112714221A (en) * | 2020-12-24 | 2021-04-27 | 平安科技(深圳)有限公司 | Method, system and related equipment for detecting intelligent voice of federated |
CN112668472B (en) * | 2020-12-28 | 2021-08-31 | 中国科学院自动化研究所 | Iris image feature extraction method, system and device based on federal learning |
CN112668472A (en) * | 2020-12-28 | 2021-04-16 | 中国科学院自动化研究所 | Iris image feature extraction method, system and device based on federal learning |
CN112910624A (en) * | 2021-01-14 | 2021-06-04 | 东北大学 | Ciphertext prediction method based on homomorphic encryption |
CN112910624B (en) * | 2021-01-14 | 2022-05-10 | 东北大学 | Ciphertext prediction method based on homomorphic encryption |
CN112966233A (en) * | 2021-02-23 | 2021-06-15 | 杭州安恒信息技术股份有限公司 | User risk operation detection method and device and computer equipment |
CN112951433A (en) * | 2021-04-06 | 2021-06-11 | 重庆邮电大学 | Mental health assessment method based on privacy calculation |
CN113326764A (en) * | 2021-05-27 | 2021-08-31 | 北京百度网讯科技有限公司 | Method and device for training image recognition model and image recognition |
CN113901501B (en) * | 2021-10-20 | 2022-11-08 | 苏州斐波那契信息技术有限公司 | Private domain user image expansion method based on federal learning |
CN113901501A (en) * | 2021-10-20 | 2022-01-07 | 苏州斐波那契信息技术有限公司 | Private domain user image expansion method based on federal learning |
CN114257412B (en) * | 2021-11-19 | 2022-10-14 | 西安电子科技大学 | Privacy protection multi-party data cooperation box-separating method, system, equipment and terminal |
CN114257412A (en) * | 2021-11-19 | 2022-03-29 | 西安电子科技大学 | Privacy protection multi-party data cooperation box-separating method, system, equipment and terminal |
CN115906162A (en) * | 2022-11-17 | 2023-04-04 | 重庆邮电大学 | Privacy protection method based on heterogeneous representation and federal factorization machine |
CN117278199A (en) * | 2023-10-18 | 2023-12-22 | 上海零数众合信息科技有限公司 | Federal learning feature screening method and system based on homomorphic encryption |
CN117236420A (en) * | 2023-11-14 | 2023-12-15 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Method and system for debugging vertical federation learning abnormal data based on data subset |
CN117236420B (en) * | 2023-11-14 | 2024-03-26 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Method and system for debugging vertical federation learning abnormal data based on data subset |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111402095A (en) | Method for detecting student behaviors and psychology based on homomorphic encrypted federated learning | |
Kumar et al. | An empirical study of the applications of data mining techniques in higher education | |
Aguiar et al. | Engagement vs performance: using electronic portfolios to predict first semester engineering student retention | |
CN111081337B (en) | Collaborative task prediction method and computer readable storage medium | |
Hasan et al. | Modelling and predicting student's academic performance using classification data mining techniques | |
Brohi et al. | Accuracy comparison of machine learning algorithms for predictive analytics in higher education | |
Alhazmi et al. | Early predicting of students performance in higher education | |
CN112085051A (en) | Image classification method and system based on weighted voting and electronic equipment | |
Howe et al. | Integrative urban AI to expand coverage, access, and equity of urban data | |
Chu et al. | [Retracted] Optimal Allocation of Higher Education Resources Based on Data Mining and Cloud Computing | |
Xu et al. | Recommendation System for Privacy‐Preserving Education Technologies | |
Nicula et al. | Why Do Romanians Take to the Streets? A Spatial Analysis of Romania's 2016-2017 Protests | |
Sangeetha et al. | Modelling of E-governance framework for mining knowledge from massive grievance redressal data | |
Marivate et al. | Exploring data science for public good in South Africa: evaluating factors that lead to success | |
Bublyk et al. | Information system development for recording offenses in smart city based on cloud technologies and social networks | |
Honaker et al. | Small area estimation with multiple overimputation | |
Kumar et al. | Analysis of student performance based on classification and map reduce approach in big data | |
Patil et al. | AI Enabled Blockchain solution for the Indian Judicial System | |
Alnoukari | ASD-BI: An agile methodology for effective integration of data mining in business intelligence systems | |
Soliudeen et al. | Higher education governance of big data: A systematic literature review | |
Lu et al. | Influence of experiential teaching and itinerary assessment on the improvement of key competencies of students | |
NAIEM et al. | DDOS attacks defense approaches and mechanism in cloud enviroment | |
WO2015114658A2 (en) | An assessment system and method for assessing educational institutes | |
Du et al. | [Retracted] Construction and Analysis of School Moral Education System Based on Big Data Technology | |
Silva et al. | What’s New in Urban Data Analytics? |
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 |