CN115099988A - Model training method, data processing method, device and computer medium - Google Patents
Model training method, data processing method, device and computer medium Download PDFInfo
- Publication number
- CN115099988A CN115099988A CN202210753985.XA CN202210753985A CN115099988A CN 115099988 A CN115099988 A CN 115099988A CN 202210753985 A CN202210753985 A CN 202210753985A CN 115099988 A CN115099988 A CN 115099988A
- Authority
- CN
- China
- Prior art keywords
- model
- sample data
- evaluation index
- target
- meta
- 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
- 238000012549 training Methods 0.000 title claims abstract description 144
- 238000000034 method Methods 0.000 title claims abstract description 106
- 238000003672 processing method Methods 0.000 title abstract description 17
- 238000011156 evaluation Methods 0.000 claims abstract description 316
- 239000013598 vector Substances 0.000 claims description 57
- 230000006870 function Effects 0.000 claims description 41
- 230000015654 memory Effects 0.000 claims description 32
- 238000012545 processing Methods 0.000 claims description 21
- 238000003860 storage Methods 0.000 claims description 17
- 238000004590 computer program Methods 0.000 claims description 15
- 238000013473 artificial intelligence Methods 0.000 abstract description 7
- 238000010586 diagram Methods 0.000 description 14
- 238000005516 engineering process Methods 0.000 description 12
- 230000008569 process Effects 0.000 description 11
- 238000004891 communication Methods 0.000 description 10
- 230000000694 effects Effects 0.000 description 8
- 230000001360 synchronised effect Effects 0.000 description 7
- 230000006399 behavior Effects 0.000 description 4
- 238000001514 detection method Methods 0.000 description 4
- 238000011161 development Methods 0.000 description 4
- 238000004821 distillation Methods 0.000 description 4
- 230000014509 gene expression Effects 0.000 description 4
- 238000012954 risk control Methods 0.000 description 4
- 230000002159 abnormal effect Effects 0.000 description 3
- 230000008878 coupling Effects 0.000 description 3
- 238000010168 coupling process Methods 0.000 description 3
- 238000005859 coupling reaction Methods 0.000 description 3
- 238000013140 knowledge distillation Methods 0.000 description 3
- 238000010801 machine learning Methods 0.000 description 3
- 238000003062 neural network model Methods 0.000 description 3
- 230000003068 static effect Effects 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000004422 calculation algorithm Methods 0.000 description 2
- 238000013461 design Methods 0.000 description 2
- 239000000126 substance Substances 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000013526 transfer learning Methods 0.000 description 2
- 230000001133 acceleration Effects 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 238000013475 authorization Methods 0.000 description 1
- 238000013529 biological neural network Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000001149 cognitive effect Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 230000001186 cumulative effect Effects 0.000 description 1
- 238000005520 cutting process Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000013136 deep learning model Methods 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 230000010354 integration Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 230000008447 perception Effects 0.000 description 1
- 238000007637 random forest analysis Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012502 risk assessment Methods 0.000 description 1
- 238000013058 risk prediction model Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 238000012795 verification 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
- G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
- G06Q40/08—Insurance
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/20—Ensemble learning
Abstract
The application discloses a model training method, a data processing method, equipment and a computer medium, which can be applied to various scenes such as financial wind control, artificial intelligence and the like. The model training method comprises the following steps: acquiring a first sample set; acquiring an auxiliary model, a plurality of reference models and a plurality of unmarked sample characteristic information; training the initial meta-teacher model based on the first sample set, the auxiliary models and the plurality of reference models to obtain a target meta-teacher model; determining a plurality of sample evaluation index marks corresponding to the plurality of unmarked sample characteristic information by utilizing the plurality of unmarked sample characteristic information, the plurality of reference models and the target meta-teacher model; and training the initial student model according to the characteristic information of the plurality of unmarked samples and the evaluation index marks of the plurality of samples to obtain a corresponding target student model. The target student model trained by the model training method provided by the application can be fused with the wind control capability of a plurality of reference models, and the accuracy of wind control is improved.
Description
Technical Field
The present application relates to the field of artificial intelligence technologies, and in particular, to a model training method, a data processing method, a device, and a computer medium.
Background
In the initial stage of financial product business, because the overdue expression period of the loan of the user is long (generally 1-6 months, a financial institution is difficult to accumulate enough labeled sample data with the post-loan expression).
The financial wind-controlled cold start problem generally occurs in the pre-loan wind control stage at the early stage of the online financial business. In the related art, the wind control in the cold start scene is generally realized based on an expert strategy rule or an unsupervised method based on abnormal detection.
The wind control under the cold start scene is realized through expert strategy rules, a wind control rule engine suitable for a target business is specified through understanding of the business by the expert strategy rules mainly depending on the business in the field of financial wind control or the expert experience of technicians, or an existing wind control model or strategy is selected and reused from similar stock financial business based on the expert experience as a cold start model or strategy of the target scene, but the maximum risk control effect is difficult to achieve due to the fact that the expert strategy needs to be strongly dependent on the experience knowledge of the professionals, and the wind control efficiency is low.
An unsupervised method based on anomaly detection is characterized in that a high-risk user in a wind control scene is modeled into an anomaly point, and data points of the high-risk user in the mode deviate from features corresponding to most evaluation index data points. The effect of anomaly detection is to identify rare data from a large amount of multidimensional data that has a large difference in behavior from most data. The mainstream anomaly detection algorithm in the financial scenario is an isolated Forest (iForest). The isolated forest belongs to an integrated learning method, and the main idea is as follows: in the training stage, the multi-dimensional feature space of the training set samples is randomly segmented, and a decision stump is constructed according to the multi-dimensional feature space until all the samples are isolated to a subspace. In this process, normal samples need to be cut multiple times to stop cutting, while abnormal samples easily stop in a subspace very early. The isolated forest represents the abnormal evaluation index of the leaf node by the quality. In the testing stage, the quality is calculated for each test sample using the isolated tree, and the smaller the quality, the smaller the number of cuts, i.e., the easier it is to be an outlier. However, in a financial wind control scene, an unsupervised method similar to an isolated forest cannot help other related business scene experiences, and the problems of high classification contingency and low wind control accuracy exist.
Disclosure of Invention
The embodiment of the application provides a model training method, which can generate a sample evaluation index mark corresponding to a label-free sample based on a plurality of reference models, and further train a corresponding wind control model, wherein the wind control model can integrate the wind control capability of the plurality of reference models, the wind control accuracy is higher, and the wind control efficiency is improved.
In one aspect, an embodiment of the present application provides a model training method, where the method includes: acquiring a first sample set, wherein the first sample set comprises a plurality of first sub-sets, each first sub-set comprises a plurality of first sample data groups, each first sample data group comprises first object characteristic information and a first evaluation index mark corresponding to the first object characteristic information; the method comprises the steps of obtaining an auxiliary model, a plurality of reference models and a plurality of unmarked sample characteristic information, wherein the reference models are models used for determining a prediction evaluation index corresponding to characteristic information of an object to be analyzed; training an initial meta-teacher model based on the first sample set, the auxiliary models and the plurality of reference models to obtain a target meta-teacher model, wherein the auxiliary models are used for assisting in training the initial meta-teacher model; determining a plurality of sample evaluation index marks corresponding to the plurality of unmarked sample characteristic information by using the plurality of unmarked sample characteristic information, the plurality of reference models and the target meta-teacher model; training an initial student model according to the plurality of label-free sample characteristic information and the plurality of sample evaluation index labels to obtain a corresponding target student model, wherein the target student model is used for determining a first target evaluation index of the input object characteristic information according to the input object characteristic information.
In another aspect, an embodiment of the present application provides a data processing method, where the method includes: acquiring characteristic information of an object to be processed; inputting the object characteristic information to be processed into a target student model to obtain a first target evaluation index corresponding to the object characteristic information to be processed; the target student model is trained by the model training method.
In one aspect, an embodiment of the present application provides a model training apparatus, where the apparatus includes: the device comprises an acquisition unit, a judgment unit and a processing unit, wherein the acquisition unit is used for acquiring a first sample set, the first sample set comprises a plurality of first subsets, each first subset comprises a plurality of first sample data groups, each first sample data group comprises first object characteristic information and a first evaluation index mark corresponding to the first object characteristic information; the acquisition unit is further used for acquiring an auxiliary model, a plurality of reference models and a plurality of unmarked sample characteristic information, wherein the reference models are models used for determining the prediction evaluation indexes corresponding to the characteristic information of the object to be analyzed; a first training unit, configured to train an initial meta-teacher model based on the first sample set, the auxiliary models, and the multiple reference models to obtain a target meta-teacher model, where the auxiliary models are used to assist in training the initial meta-teacher model; a determination unit, configured to determine, by using the plurality of unmarked sample feature information, the plurality of reference models, and the target meta-teacher model, a plurality of sample evaluation index marks corresponding to the plurality of unmarked sample feature information; and the second training unit is used for training the initial student model according to the plurality of unlabeled sample characteristic information and the plurality of sample evaluation index labels to obtain a corresponding target student model, and the target student model is used for determining a first target evaluation index of the input object characteristic information according to the input object characteristic information.
In one aspect, an embodiment of the present application provides a data processing apparatus, where the apparatus includes: the device comprises an acquisition unit, a processing unit and a processing unit, wherein the acquisition unit is used for acquiring characteristic information of an object to be processed; the input unit is used for inputting the characteristic information of the object to be processed into a target student model to obtain a first target evaluation index corresponding to the characteristic information of the object to be processed; the target student model is trained by the model training method.
In another aspect, the present application provides a computer-readable storage medium, where a computer program is stored, where the computer program is suitable for being loaded by a processor to perform the method described in any one of the above embodiments.
In another aspect, the present application provides a computer device, where the computer device includes a processor and a memory, where the memory stores a computer program, and the processor is configured to execute the model training method or the data processing method according to any one of the above embodiments by calling the computer program stored in the memory.
In another aspect, the present application provides a computer program product, which includes computer instructions, and when executed by a processor, the computer instructions implement the model training method or the data processing method according to any one of the above embodiments.
According to the embodiment of the application, a first sample set is obtained, wherein the first sample set comprises a plurality of first sub-sets, each first sub-set comprises a plurality of first sample data groups, each first sample data group comprises first object characteristic information and a first evaluation index mark corresponding to the first object characteristic information; the method comprises the steps of obtaining an auxiliary model, a plurality of reference models and a plurality of unmarked sample characteristic information, wherein the reference models are models used for determining a prediction evaluation index corresponding to characteristic information of an object to be analyzed; training an initial meta-teacher model based on the first sample set, the auxiliary models and the plurality of reference models to obtain a target meta-teacher model, wherein the auxiliary models are used for assisting in training the initial meta-teacher model; determining a plurality of sample evaluation index marks corresponding to the plurality of unmarked sample characteristic information by using the plurality of unmarked sample characteristic information, the plurality of reference models and the target meta-teacher model; training an initial student model according to the plurality of unmarked sample characteristic information and the plurality of sample evaluation index marks to obtain a corresponding target student model, wherein the target student model is used for determining a mode of a first target evaluation index of the input object characteristic information according to the input object characteristic information, fusing the wind control capability of a plurality of reference models, generating the sample evaluation index marks corresponding to the unmarked samples, and further training the corresponding wind control model, and the wind control model can fuse the wind control capability of a plurality of reference models, so that the accuracy of wind control is improved.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the description below are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings based on these drawings without creative efforts.
Fig. 1 is a schematic structural diagram of a data processing system according to an embodiment of the present application.
Fig. 2a is a schematic flow chart of a model training method according to an embodiment of the present disclosure.
Fig. 2b is a scene schematic diagram of a model training method provided in the embodiment of the present application.
Fig. 2c is a schematic structural diagram of a model training system according to an embodiment of the present application.
Fig. 2d is a schematic structural diagram of a model training system provided in the embodiment of the present application.
Fig. 3 is a schematic flowchart of a data processing method according to an embodiment of the present application.
Fig. 4 is a schematic structural diagram of a model training device according to an embodiment of the present application.
Fig. 5 is a schematic structural diagram of a data processing apparatus according to an embodiment of the present application.
Fig. 6 is a schematic structural diagram of a computer device according to an embodiment of the present application.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The embodiment of the application can be applied to various scenes such as financial wind control and artificial intelligence.
The embodiment of the application provides a model training method, a data processing method, a device and a computer medium, and particularly, the model training method or the data processing method of the embodiment of the application can be executed by a computer device, wherein the computer device can be a terminal or a server and other devices. The terminal can be a smart phone, a tablet computer, a notebook computer, an intelligent voice interaction device, an intelligent household appliance, a wearable intelligent device, an aircraft, an intelligent vehicle-mounted terminal and other devices, and can further comprise a client, wherein the client can be a video client, a browser client or an instant messaging client and the like. The server may be an independent physical server, a server cluster or a distributed system formed by a plurality of physical servers, or a cloud server providing basic cloud computing services such as a cloud service, a cloud database, cloud computing, a cloud function, cloud storage, a Network service, cloud communication, middleware service, a domain name service, a security service, a Content Delivery Network (CDN), a big data and artificial intelligence platform, and the like.
For example, when the aforementioned model training method or data processing method is run on a terminal, the terminal may download and install a corresponding application program, and when the terminal actually runs the aforementioned method, the terminal is used to display a graphical user interface and interact with a user through the graphical user interface. In particular, the manner in which the terminal presents the graphical user interface to the user may include a variety of ways, for example, the graphical user interface may be rendered for display on a display screen of the terminal or presented by holographic projection. For example, the terminal may include a touch display screen for presenting a graphical user interface and receiving operation instructions generated by a user acting on the graphical user interface, and a processor for executing the aforementioned model training method or data processing method, generating the graphical user interface, responding to the operation instructions, and controlling the display of the graphical user interface on the touch display screen.
First, some terms or expressions appearing in the course of describing the embodiments of the present application are explained as follows:
artificial Intelligence (AI) is a theory, method, technique and application system that uses a digital computer or a machine controlled by a digital computer to simulate, extend and expand human Intelligence, perceive the environment, acquire knowledge and use the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.
Cloud technology refers to a hosting technology for unifying serial resources such as hardware, software, network and the like in a wide area network or a local area network to realize calculation, storage, processing and sharing of data. The cloud technology is based on the general names of network technology, information technology, integration technology, management platform technology, application technology and the like applied in the cloud computing business model, can form a resource pool, is used as required, and is flexible and convenient. Cloud computing technology will become an important support. Background services of the technical network system require a large amount of computing and storage resources, such as video websites, picture-like websites and more web portals. With the high development and application of the internet industry, each article may have its own identification mark and needs to be transmitted to a background system for logic processing, data in different levels are processed separately, and various industrial data need strong system background support and can only be realized through cloud computing.
The blockchain system may be a distributed system formed by connecting clients, a plurality of nodes (any form of computing device in an access network, such as a server and a user terminal) through a network communication mode. A Peer-To-Peer (P2P, Peer To Peer) network is formed among nodes, a P2P Protocol is an application layer Protocol operating on a Transmission Control Protocol (TCP), in a distributed system, any machine such as a server and a terminal can be added To become a node, and the node includes a hardware layer, an intermediate layer, an operating system layer and an application layer.
Neural Networks (NN): a deep learning model simulating the structure and function of a biological neural network in the field of machine learning and cognitive science.
The wind control model is a short name for a risk control model, is commonly found in credit guarantee companies, and is used for performing risk control on businesses.
Knowledge distillation: generally, a large model is often a single complex network or a collection of networks, and has good performance and generalization capability, while a small model has limited expression capability because of a small network size. Therefore, the knowledge learned by the large model can be used for guiding the training of the small model, so that the small model has the performance equivalent to that of the large model, but the number of parameters is greatly reduced, and the compression and acceleration of the model are realized, namely the application of knowledge distillation and transfer learning in model optimization.
Knowledge distillation adopts a teacher-student mode: the complex and large model is used as the meta-teacher model, the structure of the student model is simple, the meta-teacher model is used for assisting the training of the student model, the learning capacity of the meta-teacher model is strong, and the knowledge learned by the meta-teacher model can be transferred to the student model with relatively weak learning capacity, so that the generalization capacity of the student model is enhanced. The complex, heavy and effective meta-teacher model is not on-line, but is simply in the role of a teacher, and the flexible and light student small model is really deployed on-line to execute a prediction task.
Zero sample/no label sample: the unmarked sample refers to a sample which generates the loan application but has no overdue performance in the financial wind control scene.
An expert model: a certain number of historical risk control models accumulated during business development are known, which are all sophisticated models with supervised training on labeled samples.
Pseudo label: the prediction of fraud risk for the unlabeled sample is scored by a single or a combination of multiple expert models, with greater values corresponding to greater fraud risk.
Expert weight: and (3) the relevance weight of the unmarked sample and the expert model, and the risk prediction of each unmarked sample is scored as the first pseudo label of the unmarked sample by fusing a plurality of expert models based on the weight.
A student model: the wind control model trained on the characteristic X of the unmarked sample and the first pseudo label predicted by the expert model can be used as the wind control model of the financial institution in the cold start stage of the business.
And the meta-teacher model adaptively calculates the relevance weight of each unmarked sample and the expert model, selectively (such as reserving positive correlation and removing negative correlation) weights and fuses a plurality of the expert models to score the risk prediction of each unmarked sample, and the fused risk prediction scores can be used for guiding the learning of the student model, so that the learning of the student model is more effective.
KS (Kolmogorov-Smirnov, Momoglov Simmilnov) values measure the difference between the cumulative distributions of good and bad samples. The KS value range is [0,1], the larger the accumulated difference of good and bad samples is, the larger the KS index is, and the stronger the risk distinguishing capability of the model is.
The embodiment of the application can be realized by combining cloud technology or a block chain system. The corresponding data involved in the model training method disclosed in the embodiment of the present application may be saved on the blockchain. For example: a first sample set, a plurality of reference models, a target meta-teacher model, a plurality of unmarked sample characteristic information, a plurality of sample evaluation index marks, a target student model, etc. When the corresponding data are needed to be used, the corresponding data can be directly and quickly acquired from the block chain, so that the data processing efficiency and the data acquisition efficiency are improved.
The inventor finds that in the early stage of financial product business, due to the fact that the overdue performance period of the loan of the user is long, a financial institution has difficulty in accumulating a sufficient amount of marked samples with the post-loan performance. And the wind control model or risk prediction model based on machine learning needs to be established on the basis of a large number of marked samples, and the lack of the samples causes the financial wind control modeling to face the cold start problem. On the other hand, financial institutions accumulate a lot of stock wind control models from related financial businesses in the business development process. The risk prediction capability of the stock models can be used as an expert knowledge base to assist the construction of the wind control model of the financial business in the cold start stage. The application provides one kind can be under the no mark sample scene, and through fusing many expert's models in a self-adaptation way and distilling the study in order to train the wind control model, the wind control model that the model training method based on in this application confirmed is higher, the effect is better in financial wind control field precision, can promote financial institution's wind control ability in the cold start-up stage of business effectively.
The scheme of the present application is described in detail below.
Referring to fig. 1, fig. 1 is a schematic structural diagram of a data processing system according to an embodiment of the present disclosure. The data processing system may include the terminal 10 and the server 20, etc.; the terminal 10 and the server 20 are connected via a network, such as a wired or wireless network connection.
The terminal 10, among other things, may be used to display a graphical user interface. The terminal is used for interacting with a user through a graphical user interface, for example, downloading and installing a corresponding client through the terminal and running the client, for example, calling a corresponding applet and running the applet, for example, displaying a corresponding graphical user interface through logging in a website, and the like. In the embodiment of the present application, the terminal 10 may be configured to execute a data processing method, specifically, allow a user to upload object feature information to be processed; and inputting the object characteristic information to be processed into a target student model to obtain a first target evaluation index corresponding to the object characteristic information to be processed. Also, a first target evaluation index may be displayed.
The object in the present application may be a user, and the object characteristic information may include behavior information of the user, for example, one or more of loan times, loan platforms, loan amount, payment date, overdue days, and the like.
Optionally, the object feature information may further include corresponding basic information: age information, real estate conditions, vehicle estate conditions, and the like.
The aforementioned first target evaluation index may be a numerical value indicating a credit score of the subject, or a numerical value indicating a default probability. Optionally, the larger the first target evaluation index is, the higher the default probability representing the object is.
It is understood that, in the specific implementation manner of the present application, the data related to the object feature information and the like need to be approved or agreed by the user when the above embodiments of the present application are applied to specific products or technologies, and the collection, use and processing of the related data need to comply with the related laws and regulations and standards of the related countries and regions.
Alternatively, the aforementioned target student model may be trained by the server 20, or by the terminal 10.
When the server 20 or the terminal 10 is used for training the aforementioned target student model, i.e. performing a model training method, it can be specifically used for:
acquiring a first sample set, wherein the first sample set comprises a plurality of first sub-sets, each first sub-set comprises a plurality of first sample data groups, each first sample data group comprises first object characteristic information and a first evaluation index mark corresponding to the first object characteristic information;
the method comprises the steps of obtaining an auxiliary model, a plurality of reference models and a plurality of unmarked sample characteristic information, wherein the reference models are models used for determining a prediction evaluation index corresponding to characteristic information of an object to be analyzed;
training an initial meta-teacher model based on the first sample set, the auxiliary models and the plurality of reference models to obtain a target meta-teacher model, wherein the auxiliary models are used for assisting in training the initial meta-teacher model;
determining a plurality of sample evaluation index marks corresponding to the plurality of unmarked sample characteristic information by using the plurality of unmarked sample characteristic information, the plurality of reference models and the target meta-teacher model;
training an initial student model according to the plurality of label-free sample characteristic information and the plurality of sample evaluation index labels to obtain a corresponding target student model, wherein the target student model is used for determining a first target evaluation index of the input object characteristic information according to the input object characteristic information.
Wherein the reference model may be an expert model, and the evaluation index flag may be a numerical value indicating a credit score of the object or a numerical value of a default probability.
The following describes the specific implementation of the model training method and the data processing method in detail. It should be noted that the description sequence of the following embodiments is not intended to limit the priority sequence of the embodiments.
The embodiments of the present application provide a model training method, which may be executed by a terminal or a server, or may be executed by both the terminal and the server; the embodiment of the present application is described as an example in which the model training method is executed by a server.
Fig. 2a is a schematic flowchart of a model training method provided in an embodiment of the present application, where the method includes the following steps S201 to S205:
s201, obtaining a first sample set, wherein the first sample set comprises a plurality of first sub-sets, each first sub-set comprises a plurality of first sample data groups, each first sample data group comprises first object characteristic information and a first evaluation index mark corresponding to the first object characteristic information;
the first sample set is a marked sample set, and the first object characteristic information may be behavior information of the object, for example, one or more of loan times, loan platforms, loan amount, repayment date, overdue days, and the like.
Optionally, the first object feature information may further include corresponding basic information: age information, real estate status, vehicle property status, etc.
It should be noted that the object feature information referred to in this application is obtained after authorization of the object.
The first evaluation index mark can mark the evaluation score corresponding to the first object characteristic information of the related personnel, and the evaluation score can be used for indicating the credit score value of the object or the default probability value.
For example, when the first object feature information includes one or more overdue days and the one or more overdue days are 0, the first evaluation index flag corresponding to the first object feature information is 0.
When the first object feature information includes a plurality of overdue days, wherein half of the plurality of overdue days are overdue and half of the plurality of overdue days are not overdue before payment, the first evaluation index flag corresponding to the first object feature information may be 0.5.
In some optional embodiments of the present application, the first set of samples may be a set:
S＝{s 1 ,s 2 ,...,s k in which s j ＝{(x 1j ,Y 1j ),(x 2j ,Y 2j ),...,(x nj ,Y nj ) J is 1,2, k, k is the number of the first subset contained in the first sample set, n is s j The number of the first sample data group contained in (1); s j Is a first subset, s j In (x) contained in 1j ,Y 1j )、(x 2j ,Y 2j ) A nj ,Y nj ) Is as s j A plurality of first sample data groups contained therein. Optionally, the number of the first sample data groups contained in the different first subsets is the same.
With a first sample data set (x) 1j ,Y 1j ) For example, where x 1j Characterizing information for a first object, and Y 1j And marking the first evaluation index corresponding to the first object characteristic information.
S202, obtaining an auxiliary model, a plurality of reference models and a plurality of unmarked sample characteristic information, wherein the reference models are models for determining a prediction evaluation index corresponding to the characteristic information of an object to be analyzed;
optionally, the reference model is an expert model, and is a known historical wind control model accumulated during the business development process, and the wind control models are fine models which are supervised trained on labeled samples.
Optionally, the auxiliary model is used for assisting in training the initial meta-teacher model, and the auxiliary model may be a neural network model.
The evaluation index in the present application can be used for a credit score value or a default probability value of an object corresponding to the object feature information to be analyzed.
Alternatively, the set of multiple reference models may be represented as: t ═ T 1 ,T 2 ,...T m Where T represents a number of reference models: t is 1 、T 2 、...、T m A set of reference models is formed, and m represents the number of reference models contained in the set of reference models.
The unmarked sample feature information may be object feature information for which the corresponding evaluation index is not determined.
S203, training an initial meta-teacher model based on the first sample set, the auxiliary models and the plurality of reference models to obtain a target meta-teacher model, wherein the auxiliary models are used for assisting in training the initial meta-teacher model;
wherein the target meta teacher model may be a neural network model.
In some optional embodiments of the present application, in S203, training the initial meta-teacher model based on the first set of samples, the auxiliary models, and the plurality of reference models, and obtaining the target meta-teacher model includes the following S21-S25 (not shown in the figure):
s21, aiming at each first subset in the first sample set, acquiring a plurality of first sample data groups in the first subset;
s22, for each first sample data group in the plurality of first sample data groups, inputting first object feature information in the first sample data group into each reference model in the plurality of reference models to obtain a first prediction evaluation index vector corresponding to the first sample data group, where the first prediction evaluation index vector includes a plurality of first prediction evaluation indexes, and the first prediction evaluation indexes are in one-to-one correspondence with the reference models;
the result output by the reference model may be a row vector, that is, the first prediction evaluation index vector may be a row vector, and the first prediction evaluation index is also referred to as a first pseudo tag.
For example, for each first subset s of the aforementioned first sample sets j Can obtain s j A plurality of first sample data sets contained: (x) 1j ,Y 1j )、(x 2j ,Y 2j ) A nj ,Y nj )。
For a first sample data set (x) 1j ,Y 1j ) Its first object characteristic information x 1j Inputting the plurality of reference models: t is 1 、T 2 、...、T m In this way, x is obtained 1j Corresponding first prediction evaluation index vector
Wherein the first prediction evaluation index vector
A plurality of first predictive rating indices are included:
s23, determining a second sample data set corresponding to the first sample data set according to the first sample data set and a first prediction evaluation index vector corresponding to the first sample data set to obtain a second subset corresponding to the first subset, wherein the second subset comprises a plurality of second sample data sets, and the second sample data set comprises the first sample data set and a first prediction evaluation index vector corresponding to the first sample data set;
specifically, determining the second sample data set corresponding to the first sample data set according to the first prediction evaluation index vector corresponding to the first sample data set may include:
and taking the first sample data group and a data group formed by the first prediction evaluation index vector corresponding to the first sample data group as a second sample data group.
For example, the first sample data group (x) 1j ,Y 1j ) The corresponding second sample data set is: (x) 1j ,Y 1j ,Y 1j exp )。
Accordingly, the first subset s j The corresponding second subset may be D j A plurality of second sample data sets contained in the second subset and the first subset s j The plurality of first sample data groups contained in (1) are in one-to-one correspondence.
S24, determining a second sample set corresponding to the first sample set according to a second subset corresponding to each first subset, wherein the second sample set comprises a plurality of second subsets;
for example, the first set of samples S ═ { S ═ S 1 ,s 2 ,...,s k The corresponding second set of samples is:
D＝{D 1 ,D 2 ,...,D k wherein the second set of samples D comprises a plurality of second subsets: d 1 、D 2 、...、D k ；
k is the number of the second subset included in the second sample set, and n is the second subset D j The number of the second sample data group contained in the sample data; d j Including a plurality of second sample data sets:
and S25, training the initial meta-teacher model by using the second sample set and the auxiliary model to obtain a target meta-teacher model.
In some optional embodiments of the present application, in S25, training the initial meta-teacher model with the second sample set and the auxiliary models to obtain a target meta-teacher model, including the following S251-S256 (not shown in the figure):
s251, selecting a target subset from a plurality of second subsets included in the second sample set;
the second subset may be selected from a plurality of second subsets included in the second sample set in a front-to-back order as the target subset, or may be selected randomly from a plurality of second subsets included in the second sample set as the target subset.
S252, grouping at least some second sample data groups in the second sample data groups included in the target subset according to a first preset rule, to obtain a target number of task groups corresponding to the first preset rule, where each task group includes a plurality of second sample data groups;
the first preset rule may include a number of the grouped second sample data groups in the plurality of second sample data groups included in the target subset, and the target number.
Optionally, the plurality of second sample data groups contained in the target subset may all participate in the grouping, and the target number may be 5.
Optionally, the number of the second sample data groups included in each task group is the same.
S253, selecting a target task group from the target number of task groups;
the task group may be selected from the target number of task groups in the order from front to back as the target task group, or may be randomly selected from the target number of task groups as the target task group.
S254, selecting a first part of second sample data group and a second part of second sample data group from a plurality of second sample data groups included in the target task group according to a second preset rule;
the second preset rule may include the number of the second sample data groups included in the first part of the second sample data groups, or the ratio of the number of the first part of the second sample data groups in a plurality of second sample data groups included in the target task group; and the number of the second sample data groups contained in the second part of the second sample data groups, or the number proportion of the second part of the second sample data groups in a plurality of second sample data groups included in the target task group.
Optionally, a sum of the first part of the second sample data set and the second part of the second sample data set is a total number of the second sample data sets included in the target task group.
Optionally, the number of second sample data groups contained in the first portion of the second sample data group is the same as the number of second sample data groups contained in the second portion of the second sample data group.
S255, determining a support sample data set based on the first part of the second sample data set, and determining a query sample data set based on the second part of the second sample data set;
alternatively, the first portion of the second set of sample data may be taken directly as the support set of sample data and the second portion of the second set of sample data as the query set of sample data.
And S256, training the initial meta-teacher model according to the support sample data set, the query sample data set and the auxiliary model to obtain a target meta-teacher model.
Optionally, the model parameters of the auxiliary model are first parameter information, and in the foregoing S256, training the initial meta-teacher model according to the support sample data set, the query sample data set, and the auxiliary model to obtain the target meta-teacher model includes the following S2561-S2566 (not shown in the figure):
s2561, obtaining an initial meta-teacher model with model parameters being second parameter information;
the number of the second parameter information may be multiple, the number of the second parameter information is the same as the number of the reference models, and the second parameter information is the same as the number of the reference modelsThe initial value of the information may be set by the associated personnel and the initial meta-teacher model may be composed of a plurality of fully connected layers. When the number of the second parameter information is multiple, the initial values of the multiple second parameter information may be:
s2562, determining a plurality of second evaluation index marks corresponding to the support sample data group on the basis of the support sample data group through the initial meta teacher model;
optionally, in S2562, determining, by the initial meta teacher model, a plurality of second evaluation index markers corresponding to the support sample data set based on the support sample data set may include the following S001-S002 (not shown in the figure):
s001, aiming at each third sample data group in a plurality of third sample data groups contained in the support sample data group, obtaining second object characteristic information in the third sample data group and a second prediction evaluation index vector corresponding to the second object characteristic information to obtain a plurality of second object characteristic information and a plurality of second prediction evaluation index vectors corresponding to the support sample data group;
the first partial second sample data group in the support sample data group is determined to be a plurality of third sample data groups.
And S002, determining a plurality of second evaluation index marks corresponding to the support sample data group based on the plurality of second object feature information and the plurality of second prediction evaluation index vectors through the initial meta-teacher model.
In the aforementioned S002, the determining, by the initial meta-teacher model, a plurality of second evaluation index markers corresponding to the support sample data set based on the plurality of second object feature information and the plurality of second prediction evaluation index vectors may specifically include:
and inputting the plurality of second object feature information and the plurality of second prediction evaluation index vectors into the initial meta-teacher model to obtain a plurality of second evaluation index marks corresponding to the support sample data set. And the second evaluation index mark corresponds to the second object characteristic information one by one.
In some alternative embodiments of the present application, see fig. 2b, where fig. 2b is a scene diagram of a model training method, and X in fig. 2b s Representing a plurality of second object feature information, and Y s exp A plurality of second prediction evaluation index vectors representing a plurality of second object feature information;
representing a plurality of second evaluation index markers.
Optionally, inputting the plurality of second object feature information and the plurality of second prediction evaluation index vectors into the initial meta-teacher model, and obtaining a plurality of second evaluation index labels corresponding to the support sample data set may be implemented by the following formula:
wherein, X s Representing a plurality of second object feature information, and Y s exp A plurality of second prediction evaluation index vectors representing a plurality of second object feature information correspondences;
a plurality of second evaluation index markers are represented,
the information of the second parameter is represented,
representing the corresponding function of the initial meta teacher model.
Specifically, the initial meta-teacher model includes a weight determining unit and an evaluation index determining unit, and specifically, in the foregoing S002, the determining, by the initial meta-teacher model, a plurality of second evaluation index markers corresponding to the support sample data set based on the plurality of second object feature information and the plurality of second prediction evaluation index vectors may specifically include the following S01-S03 (not shown in the figure):
s01, for each second object feature information of the plurality of second object feature information, determining, by the weight determination unit, a plurality of corresponding weight information (also called expert weights) based on the second object feature information, where the number of the weight information is the same as the number of the reference models, the weight information corresponds to the reference models one-to-one, and the weight information corresponds to the second parameter information one-to-one;
wherein, the plurality of weight information are shown in fig. 2 b: w1, w2, w 3.
S02, determining, by the evaluation index flag determination unit, a second evaluation index flag corresponding to the second object feature information based on the plurality of pieces of weight information and a second predicted evaluation index vector corresponding to the second object feature information;
optionally, based on the plurality of weight information, determining a second evaluation index flag corresponding to the second object feature information by using a second predicted evaluation index vector corresponding to the second object feature information may include:
for each piece of weight information in the plurality of pieces of weight information, multiplying the piece of weight information by a target second prediction evaluation index corresponding to the piece of weight information in a plurality of second prediction evaluation indexes included in the second prediction evaluation index vector to obtain a to-be-processed index corresponding to the piece of weight information, and further obtaining a plurality of to-be-processed indexes corresponding to the pieces of weight information;
and summing the plurality of indexes to be processed to obtain a second evaluation index mark corresponding to the second object characteristic information.
And S03, determining a plurality of second evaluation index marks corresponding to the support sample data set according to the second evaluation index marks corresponding to the second object characteristic information.
Specifically, the second evaluation index markers corresponding to the second object feature information form a plurality of second evaluation index markers corresponding to the support sample data set.
S2563, determining a plurality of third evaluation index marks corresponding to the support sample data group on the basis of the support sample data group through the auxiliary model;
optionally, in S2563, determining, by the auxiliary model, a plurality of third evaluation index markers corresponding to the support sample data set based on the support sample data set may include the following S3001 to S3003 (not shown in the figure):
s3001, for each third sample data group in a plurality of third sample data groups included in the support sample data group, obtaining second object feature information in the third sample data group;
s3002, inputting the second object characteristic information into the auxiliary model to obtain a third evaluation index mark corresponding to the second object characteristic information;
and S3003, determining a plurality of third evaluation index marks corresponding to the support sample data group according to the third evaluation index marks corresponding to the second object feature information in each third sample data group.
And the third evaluation index marks corresponding to the second object characteristic information in the third sample data groups are the third evaluation index marks corresponding to the support sample data groups.
Optionally, when a plurality of second object feature information is input into the auxiliary model to obtain a plurality of third evaluation index marks, the plurality of second object feature information is input into the auxiliary model to obtain a plurality of third evaluation index marks, which may be implemented by the following formula:
wherein the content of the first and second substances,
the information of the first parameter is represented,
representing the corresponding function of the auxiliary model.
S2564, determining first parameter information to be updated of the auxiliary model according to the support sample data set and the plurality of second evaluation index marks, and obtaining an updated auxiliary model with model parameters of the first parameter information to be updated;
specifically, in S2564, the first parameter information to be updated of the auxiliary model is determined according to the set of support sample data and the plurality of second evaluation index markers, which includes the following S41-S43 (not shown in the figure):
s41, aiming at each third sample data group in a plurality of third sample data groups contained in the support sample data group, obtaining second object characteristic information in the third sample data group;
s42, inputting the second object characteristic information into the auxiliary model to obtain a fifth evaluation index mark corresponding to the second object characteristic information;
s43, determining a plurality of fifth evaluation index marks corresponding to the support sample data group according to the fifth evaluation index marks corresponding to the second object feature information in each third sample data group;
and S44, determining first parameter information to be updated of the auxiliary model according to a fifth preset loss function, the fifth evaluation index marks and the second evaluation index marks.
Specifically, the fifth preset loss function may be a cross entropy function, and specifically, a plurality of fifth evaluation index markers and the plurality of second evaluation index markers may be input to the fifth preset loss function to obtain a fifth loss value, and the first parameter information to be updated and the updated auxiliary model are determined based on the fifth loss value.
See FIG. 2b, FIG. 2b
Representing the first parameter information to be updated, of FIG. 2b
Indicating the first parameter information.
Optionally, after the first parameter information to be updated is determined, the first parameter information to be updated may be used as new first parameter information (i.e., a model parameter of the auxiliary model), and the step S42 is further executed again until the number of times of updating the auxiliary model reaches a certain number, and the updating is stopped, so as to obtain an updated auxiliary model, so that the difference between the fifth evaluation index markers and the second evaluation index markers is smaller. And within a certain updating frequency, the more updating frequency is, the smaller the difference between the plurality of fifth evaluation index marks and the plurality of second evaluation index marks is.
S2565, determining a plurality of fourth evaluation index marks corresponding to the query sample data set on the basis of the query sample data set through the updated auxiliary model;
specifically, in S2565, a plurality of fourth evaluation index markers corresponding to the query sample data set are determined based on the query sample data set through the updated auxiliary model, including the following S51-S53 (not shown in the figure):
s51, aiming at each fourth sample data group in a plurality of fourth sample data groups contained in the query sample data group, obtaining third object characteristic information in the fourth sample data group;
wherein the second sample data set is a plurality of fourth sample data sets for determining the second part of the query sample data sets.
S52, inputting the third object characteristic information into the updated auxiliary model to obtain a fourth evaluation index mark corresponding to the third object characteristic information;
and S53, determining a plurality of fourth evaluation index marks corresponding to the query sample data set according to the fourth evaluation index marks corresponding to the third object feature information in each fourth sample data set.
And if the plurality of fourth evaluation index marks corresponding to the third object characteristic information in the plurality of fourth sample data groups are the plurality of fourth evaluation index marks corresponding to the query sample data groups.
Optionally, when the plurality of third object feature information is input into the updated auxiliary model to obtain a plurality of fourth evaluation index marks corresponding to the plurality of third object feature information, which may be implemented by the following formula:
wherein, X q A plurality of third object characteristic information is represented,
indicates the first parameter information to be updated,
and representing the function corresponding to the updated auxiliary model.
S2566, training the initial meta-teacher model by using the support sample data set, the query sample data set, the second evaluation index marks, the third evaluation index marks, the fourth evaluation index marks, the first parameter information to be updated and the second parameter information to obtain a target meta-teacher model.
Selecting a target subset from a plurality of second subsets contained in the second sample set in the process of training a target meta-teacher model; and grouping a plurality of second sample data groups contained in the target subset, and simultaneously training a target meta-teacher model by means of a basic model, namely an auxiliary model, and adopting a training thought of meta-learning.
Optionally, in the foregoing S2566, training the initial meta-teacher model by using the support sample data set, the query sample data set, the second evaluation index markers, the third evaluation index markers, the fourth evaluation index markers, the first parameter information to be updated, and the second parameter information to obtain the target meta-teacher model, including the following S61-S64 (not shown in the figure):
s61, determining a first loss value according to a first preset loss function, the support sample data set and the plurality of second evaluation index marks;
optionally, in S61, determining a first loss value according to the first preset loss function, the set of support sample data, and the plurality of second evaluation index markers may include:
aiming at each third sample data group in a plurality of third sample data groups contained in the support sample data group, acquiring a sixth evaluation index mark corresponding to second object characteristic information in the third sample data group to obtain a plurality of sixth evaluation index marks corresponding to the support sample data group;
determining a first loss value according to a first preset loss function, the sixth evaluation index markers, and the second evaluation index markers.
The sixth evaluation index marks correspond to the second evaluation index marks one to one, the first preset loss function is a cross entropy function, and the determination of the first loss value according to the first preset loss function, the sixth evaluation index marks and the second evaluation index marks can be realized by the following formula:
loss 1 ＝crossentropy 1 (Y s soft_label ,Y S )
therein, loss 1 Cross to the first loss value 1 Is a first predetermined loss function, Y S For a plurality of sixth evaluation index markers, Y s soft_label A plurality of second evaluation index markers.
Wherein the first loss value may be used to determine an approximation of the plurality of second evaluation index markers to the plurality of sixth evaluation index markers in the set of support sample data in order to make the plurality of second evaluation index markers and the plurality of sixth evaluation index markers in the set of support sample data approach one another.
S62, determining a second loss value according to a second preset loss function, the support sample data set and the third evaluation index marks;
optionally, in S62, determining a second loss value according to a second preset loss function, the set of support sample data, and the third evaluation index markers may include:
aiming at each third sample data group in a plurality of third sample data groups contained in the support sample data group, acquiring a sixth evaluation index mark corresponding to second object characteristic information in the third sample data group to obtain a plurality of sixth evaluation index marks corresponding to the support sample data group;
determining a second loss value according to a second preset loss function, the sixth evaluation index markers, and the third evaluation index markers.
The sixth evaluation index marks correspond to the third evaluation index marks one by one, the second preset loss function is a cross entropy function, and the determination of the second loss value according to the second preset loss function, the sixth evaluation index marks and the third evaluation index marks can be realized by the following formula:
loss 2 ＝crossentropy 2 (Y s predict1 ,Y s )
therein, loss 2 At the second loss value, crossntropy 2 Is a second predetermined loss function, Y S For a plurality of sixth evaluation index markers, Y s predict1 A plurality of third evaluation indices are labeled.
S63, determining a third loss value according to a third preset loss function, the query sample data set and the fourth evaluation index marks;
optionally, in S63, determining a third loss value according to a third preset loss function, the query sample data set, and the fourth evaluation index markers, including;
acquiring a seventh evaluation index mark corresponding to third object characteristic information in a fourth sample data group aiming at each fourth sample data group in a plurality of fourth sample data groups contained in the query sample data group, and acquiring a plurality of seventh evaluation index marks corresponding to the query sample data group;
determining a third loss value according to a fourth preset loss function, the seventh evaluation index markers, and the fourth evaluation index markers.
The seventh evaluation index marks correspond to the fourth evaluation index marks one to one, the third preset loss function is a cross entropy function, and the determination of the third loss value according to the fourth preset loss function, the plurality of seventh evaluation index marks and the plurality of fourth evaluation index marks can be realized by the following formula:
among them, loss 3 Cross-sensitivity as a third loss value 3 Is a third predetermined loss function, Y q For a plurality of seventh evaluation index markings,
a plurality of fourth evaluation index markers.
The third evaluation index markers are output results of the auxiliary model of which the model parameters are the first parameter information, and the fourth evaluation index markers are output results of the updated auxiliary model of which the model parameters are the first parameter information to be updated; a plurality of third evaluation index marks correspond to the second loss values, and a plurality of fourth evaluation index marks correspond to the third loss values; the first parameter information to be updated is determined by the initial teacher model according to a plurality of second object feature information in the input support sample data set and a plurality of second evaluation index marks determined by the plurality of second prediction evaluation index vectors. The more accurate the plurality of second evaluation index markers determined by the initial teacher model, the smaller the third loss value is compared to the second loss value. And updating parameters of the target element teacher model based on an output result of the auxiliary model so as to achieve the aim of assisting the training of the target element teacher model according to the auxiliary model.
And S64, training the initial meta-teacher model according to the first loss value, the second loss value, the third loss value, the first parameter information to be updated, the second parameter information and the plurality of second evaluation index marks to obtain a target meta-teacher model.
Optionally, in S64, training an initial meta-teacher model according to the first loss value, the second loss value, the third loss value, the first parameter information to be updated, the second parameter information, and the second evaluation index markers to obtain a target meta-teacher model, including S641-S642 (not shown in the figure):
s641, determining second parameter information to be updated corresponding to the initial meta-teacher model according to the first loss value, the second loss value, the third loss value, the first parameter information to be updated, the second parameter information, and the second evaluation index marks;
alternatively, the determination of the aforementioned first loss value, second loss value, third loss value, and second parameter information to be updated may be implemented by the second parameter information to be updated determination unit in fig. 2 b.
Specifically, determining second parameter information to be updated corresponding to the initial meta-teacher model according to the first loss value, the second loss value, the third loss value, the first parameter information to be updated, the second parameter information, and the plurality of second evaluation index markers may include:
determining a first result according to the first loss value and the second parameter information, specifically, calculating a gradient value of the first loss value to the second parameter information to obtain a first result;
determining a second result according to the second loss value and the first parameter information, and specifically, calculating a gradient value of the second loss value to the first parameter information; obtaining a second result;
determining a third result according to the third loss value and the first parameter information to be updated, specifically, calculating a gradient value of the third loss value to the first parameter information to be updated to obtain a third result;
taking the plurality of second evaluation index marks as the input of a preset cross entropy function, and executing the preset cross entropy function to obtain a fourth result;
calculating the gradient value of the fourth result to the second parameter information to obtain a fifth result;
calculating the product of the transpose of the second result, the third result and the fifth result to obtain a sixth result;
summing the sixth result and the first result to obtain a summation result;
and determining the second parameter information to be updated according to the summation result, the preset learning rate of the initial meta-teacher model and the second parameter information.
Specifically, a product result of a preset learning rate of the initial meta-teacher model and the summation result may be calculated, and a difference between the second parameter information and the product result is used as second parameter information to be updated.
Specifically, determining second parameter information to be updated corresponding to the initial meta-teacher model according to the first loss value, the second loss value, the third loss value, the first parameter information to be updated, the second parameter information, and the plurality of second evaluation index markers may be implemented by the following formula:
wherein the content of the first and second substances,
the information of the second parameter is represented,
indicating the second parameter information to be updated,
representing first parameter information, an
Represents the first parameter information to be updated, loss 1 Representing a first loss value, loss 2 Represents the second loss value, loss 3 A third value of the loss is represented,
represents solving gradient, mu represents preset learning rate of initial meta teacher model, Y s soft_lable Representing a plurality of second evaluation index markers.
S642, training the initial meta teacher model by using the second to-be-updated parameter information and the second parameter information to obtain a target meta teacher model.
Optionally, in S642, training the initial meta-teacher model by using the second to-be-updated parameter information and the second parameter information to obtain a target meta-teacher model, including:
determining whether the absolute value of the difference value between the second parameter information to be updated and the second parameter information is smaller than a first preset value, if so, taking the initial meta teacher model as a target meta teacher model, and outputting the target meta teacher model; if not, the second to-be-updated parameter information is used as second parameter information of the initial meta-teacher model to obtain an updated initial meta-teacher model, an unselected task group is selected from the target number of task groups to serve as a target task group, the first part of second sample data group and the second part of second sample data group are selected from a plurality of second sample data groups included in the target task group according to a second preset rule, and the initial meta-teacher model is used as a target meta-teacher model until the absolute value of the difference value between the second to-be-updated parameter information and the second parameter information is smaller than a first preset value, and the target meta-teacher model is output.
Optionally, if there is no unselected task group in the target number of task groups, selecting an unselected second subset from a plurality of second subsets included in the second sample set as a target subset, and returning to perform grouping on at least part of second sample data groups in the plurality of second sample data groups included in the target subset according to a first preset rule, to obtain a target number of task groups corresponding to the first preset rule, until it is determined that an absolute value of a difference between the second parameter information to be updated and the second parameter information is smaller than a first preset value, taking the initial meta-teacher model as a target meta-teacher model, and outputting the target meta-teacher model.
In other alternative embodiments of the present application, the method further comprises:
obtaining the updating times of the model parameters of the initial meta-teacher model;
determining whether the updating times are larger than preset times, if so, taking the initial meta-teacher model as a target meta-teacher model, and outputting the target meta-teacher model; if not, starting to execute and determining whether the absolute value of the difference value between the second parameter information to be updated and the second parameter information is smaller than a first preset value or not.
And S204, determining a plurality of sample evaluation index marks corresponding to the plurality of unmarked sample characteristic information by using the plurality of unmarked sample characteristic information, the plurality of reference models and the target meta-teacher model.
Optionally, in S204, determining a plurality of sample evaluation index labels corresponding to the plurality of unlabeled sample feature information by using the plurality of unlabeled sample feature information, the plurality of reference models, and the target meta-teacher model, including the following S31-S33 (not shown in the figure):
s31, inputting the unmarked sample characteristic information into the plurality of reference models according to the unmarked sample characteristic information in the unmarked sample characteristic information to obtain a target evaluation index vector corresponding to the unmarked sample characteristic information;
in some optional embodiments of the present application, if the set of the plurality of unlabeled sample feature information is:
D new ＝{x 1 ,x 2 ,...,x n1 }∈R n1×d the characteristic information of the plurality of unmarked samples is as follows: x is the number of 1 、x 2 、...、x n1 (ii) a n1 represents the number of the unlabeled sample feature information, d represents the dimension information of each unlabeled sample feature information, and the dimension information may be specifically the number or the kind.
S32, determining a plurality of target evaluation index vectors corresponding to the plurality of unmarked sample characteristic information according to the target evaluation index vectors corresponding to the unmarked sample characteristic information;
optionally, the label-free sample characteristic information x is used 1 Inputting the plurality of reference models to obtain the unmarked sample characteristic information x 1 The corresponding target evaluation index vector is Y 1 exp And further obtaining a plurality of target evaluation index vectors corresponding to the characteristic information of the plurality of unmarked samples:
each target evaluation index vector comprises a plurality of second target evaluation indexes, and the second target evaluation indexes correspond to the reference model one to one. Wherein, the second target evaluation index can also be called as a second pseudo label.
And S33, inputting the plurality of unlabeled sample characteristic information and the plurality of target evaluation index vectors into the target meta-teacher model to obtain a plurality of sample evaluation index labels corresponding to the plurality of unlabeled sample characteristic information, so that the initial student model can perform distillation learning.
S205, training an initial student model according to the multiple unlabeled sample characteristic information and the multiple sample evaluation index labels to obtain a corresponding target student model, wherein the target student model is used for determining a first target evaluation index of the input object characteristic information according to the input object characteristic information. The initial student model can be a neural network model, and can also be a Bayes, decision tree, random forest model and the like.
According to the embodiment of the application, a first sample set is obtained, wherein the first sample set comprises a plurality of first sub-sets, each first sub-set comprises a plurality of first sample data groups, each first sample data group comprises first object characteristic information and a first evaluation index mark corresponding to the first object characteristic information; the method comprises the steps of obtaining an auxiliary model, a plurality of reference models and a plurality of unmarked sample characteristic information, wherein the reference models are models used for determining a prediction evaluation index corresponding to characteristic information of an object to be analyzed; training an initial meta-teacher model based on the first sample set, the auxiliary models and the plurality of reference models to obtain a target meta-teacher model, wherein the auxiliary models are used for assisting in training the initial meta-teacher model; determining a plurality of sample evaluation index marks corresponding to the plurality of unmarked sample characteristic information by using the plurality of unmarked sample characteristic information, the plurality of reference models and the target meta-teacher model; training an initial student model according to the plurality of unmarked sample characteristic information and the plurality of sample evaluation index marks to obtain a corresponding target student model, wherein the target student model is used for determining a mode of a first target evaluation index of the input object characteristic information according to the input object characteristic information, fusing the wind control capability of a plurality of reference models, generating the sample evaluation index marks corresponding to the unmarked samples, and further training the corresponding wind control model, and the wind control model can fuse the wind control capability of a plurality of reference models, so that the accuracy of wind control is improved.
In addition, the distillation learning-based target student model training method, namely the trained target meta-teacher model output data-based target student model training method, can migrate wind control or risk prediction knowledge most relevant to preset business types in a plurality of reference models to the target student model, so that accuracy of wind control is improved.
Moreover, because the multiple reference models are generally from related services, and different reference models can have different reference values when aiming at the same characteristic information of the object to be analyzed, the determination of the prediction evaluation index corresponding to the characteristic information of the object to be analyzed can be performed, and the target meta-teacher model related to the application can adaptively determine the weight information related to the reference values of different reference models, and further determine the sample evaluation index mark corresponding to the characteristic information of the unmarked sample based on the weight information when acquiring the characteristic information of the unmarked sample, so that the wind control capability of the multiple reference models can be fused, and the matching degree of the determined sample evaluation index mark and the characteristic information of the unmarked sample is improved. Therefore, the problem that only the characteristic information of the marked sample exists and the corresponding sample evaluation index mark is lacked in the cold start stage is solved.
Optionally, the aforementioned multiple reference models may belong to only one service type, that is, may only process services of the same type, or may belong to multiple service types at the same time, so as to enhance the generalization of the target meta-teacher model. The service types may include: fraud prediction, breach prediction, etc.
Optionally, the two sets of reference models belong to different service types, each set of reference model includes a plurality of reference models, and each set of reference model may correspond to one target meta-teacher model, that is, the service type corresponding to the target meta-teacher model is the same as or similar to the service type to which the plurality of reference models corresponding thereto belong. Correspondingly, the service types of the services which can be processed by the trained target student model are consistent with the service types corresponding to the target meta-teacher model by using the multiple sample evaluation index marks corresponding to the multiple unmarked sample characteristic information determined based on the target meta-teacher model.
Optionally, in the foregoing S205, training an initial student model according to the feature information of the multiple unlabeled samples and the multiple sample evaluation index labels to obtain a corresponding objective student model, includes:
inputting the feature information of the plurality of unmarked samples into an initial student model with model parameters as third parameter information to obtain a plurality of prediction evaluation indexes;
determining third parameter information to be updated based on the plurality of sample evaluation index marks, the plurality of predicted evaluation indexes and a preset objective function;
determining a target absolute value of a difference value between the third parameter information to be updated and the third parameter information;
and determining whether the target absolute value is smaller than a second preset value, if so, taking the initial student model as a target student model, if not, taking the third to-be-updated parameter information as third parameter information (namely, as a model parameter of the initial student model), returning to execute the initial student model with the plurality of unmarked sample characteristic information input model parameters as the third parameter information, and obtaining a plurality of prediction evaluation indexes, and outputting the target student model until the target absolute value is smaller than the second preset value.
The target absolute value of the difference between the third parameter information to be updated and the third parameter information is related to the similarity between the sample evaluation index markers and the prediction evaluation indexes, and the greater the similarity between the sample evaluation index markers and the prediction evaluation indexes, the smaller the target absolute value of the difference between the third parameter information to be updated and the third parameter information is, that is, the closer the sample evaluation index markers and the prediction evaluation indexes are, the smaller the target absolute value of the difference between the third parameter information to be updated and the third parameter information is.
The scheme of the application can be suitable for cold start scenes of financial wind control, and can be used for user risk assessment of credit approval links.
Fig. 2c is a schematic structural diagram of a model training system corresponding to the model training method provided in the embodiment of the present application, where the model training system may include: the first target meta teacher training module and the target student training module.
Optionally, the first target meta-teacher model training module may involve data including: the first sample set, the plurality of reference models, the initial meta-teacher model, and the auxiliary model are specifically used for performing the obtaining of the plurality of reference models and the auxiliary models in the foregoing S201-S202, and S203.
Optionally, the target student model training module, the related data include: a plurality of unlabeled sample feature information, a plurality of reference models, an initial student model, and a target meta-teacher model. Specifically, the method is used for performing the aforementioned steps of obtaining a plurality of unmarked sample feature information and obtaining a reference model in S202, and S204-S205.
For the specific implementation principle, reference may be made to the foregoing contents, which are not described herein again. After the training of the target student model is completed, the model can be used as a wind control model to be on-line for related personnel to use.
Fig. 2d is another schematic structural diagram of a model training system corresponding to the model training method provided in the embodiment of the present application, where the model training system may include: the system comprises a pseudo label generation module, a second target meta teacher model training module and a target student model training module.
The data related to the pseudo tag generation module comprises the following data: the output data of the pseudo label generating module is a plurality of first prediction evaluation index vectors, wherein each first prediction evaluation index vector comprises a plurality of first prediction evaluation indexes, and the first prediction evaluation indexes are also called first pseudo labels.
The data involved by the second target meta teacher model training module includes: a plurality of first prediction evaluation index vectors, a first set of samples, an auxiliary model, and an initial meta-teacher model. The output data of the second target meta-teacher model training module includes: and training the initial meta teacher model, namely the target meta teacher model.
The target student model training module relates to data including: the system comprises a plurality of unlabeled sample feature information, a plurality of sample evaluation index labels, a plurality of reference models, a target meta-teacher model and an initial student model, wherein the output data of a target student model training module is a trained student model, namely a target student model.
Specifically, the pseudo tag generation module may be configured to execute the obtaining of the reference model in the foregoing steps S201 to S202, and the foregoing steps S21 to S22. Specifically, the second target meta-teacher model training module may be configured to execute the acquiring auxiliary model and the plurality of reference models in S202, and the aforementioned S203. Specifically, the target student model training module may be configured to perform the aforementioned acquiring a plurality of unlabeled sample feature information in S202, S204 and S205. The pseudo label generating module is further configured to execute the foregoing S31-S32, and the second target evaluation index in each target evaluation index vector is also referred to as a second pseudo label.
For the specific implementation principle, reference may be made to the foregoing contents, which are not described herein again. It should be noted that the numerical range of the first pseudo label and the second pseudo label is 0-1, and the training of the objective student model may adopt any machine learning method such as regression.
The training method (also called as a modeling scheme) of the wind control model (namely the target student model) based on the meta learning and the distillation learning, which is provided by the scheme, can be used for training the cold start wind control model based on the self-adaptive transfer learning of the trained target teacher model under the condition of only providing a plurality of unmarked sample characteristic information, and effectively improving the fraud risk prediction capability of the financial business at the cold start stage. Table 1 shows verification results of KS values corresponding to 12 sample data sets, and the KS values corresponding to the following wind control models are compared in table 1 at the same time: and DZ, training the wind control model by using the real characteristic information of the marked samples and the evaluation indexes corresponding to the characteristic information of each marked sample. The Mean refers to a plurality of unmarked sample characteristic information, and distills the trained wind control model based on the corresponding average prediction evaluation index determined by 4 reference models; and a Meta teacher, namely a wind control model determined by the model training method based on 4 reference models.
TABLE-Cold Start modeling scheme Effect display based on multiple reference models
| Sample data set | DZ | MEAN | Meta teacher |
| S1 | 0.330 | 0.336 | 0.347 |
| S2 | 0.216 | 0.250 | 0.257 |
| S3 | 0.498 | 0.450 | 0.439 |
| S4 | 0.259 | 0.159 | 0.225 |
| S5 | 0.229 | 0.225 | 0.237 |
| S6 | 0.338 | 0.302 | 0.297 |
| S7 | 0.124 | 0.108 | 0.107 |
| S8 | 0.240 | 0.215 | 0.216 |
| S9 | 0.244 | 0.243 | 0.234 |
| S10 | 0.228 | 0.240 | 0.269 |
| S11 | 0.258 | 0.172 | 0.219 |
| S12 | 0.223 | 0.214 | 0.230 |
| Mean value | 0.266 | 0.243 | 0.257 |
According to the content of the table 1, it can be seen that the effect of the wind control model trained by the model training method provided by the scheme is better than that of the wind control model determined by the Mean method on the whole, the target meta-teacher model trained by the model training method provided by the scheme can learn knowledge of different reference models, so that a pseudo label with higher quality is generated, the wind control model with higher wind control accuracy can be trained by distillation, and the KS value of one point can be improved compared with the wind control model determined by the Mean method on the whole. In addition, the comparison result of DZ and Meta teachers shows that the scheme can achieve a prediction effect close to a wind control model trained based on real marks in a cold start scene, and the effect of the scheme even exceeds the real training effect on partial evaluation finger data sets (such as S1, S2, S5, S10 and S12). The scheme is proved to be capable of effectively improving the risk prediction capability in the service cold start stage.
The embodiments of the present application provide a data processing method, which may be executed by a terminal or a server, or may be executed by both the terminal and the server; the embodiment of the present application is described as an example of terminal implementation.
Fig. 3 is a schematic flowchart of a data processing method according to an embodiment of the present application, where the method includes:
s301, obtaining characteristic information of an object to be processed;
the object feature information may include behavior information of the user, such as one or more of loan times, loan platforms, loan amounts, repayment dates, overdue days, and the like.
The object feature information may further include corresponding basic information: age information, real estate conditions, vehicle estate conditions, and the like.
S302, inputting the characteristic information of the object to be processed into a target student model to obtain a first target evaluation index corresponding to the characteristic information of the object to be processed;
the target student model is a target student model trained by the model training method in the embodiment corresponding to fig. 2 a. Optionally, the embodiment of the application may be applied to a credit approval link, and the first target evaluation index is determined according to the to-be-processed object characteristic information of the object applying for loan.
Optionally, the embodiment of the present application is also applicable to other scenarios, for example, human body gesture recognition, speech recognition, simultaneous interpretation, and the like.
When the method is applied to human body posture recognition, the object characteristic information may be object specific posture data, wherein the posture data may be position information of one or more specific joints, and accordingly, the first target evaluation index may be an accuracy score corresponding to a preset action.
When applied to speech recognition, the object feature information may be speech data specific to the object, and accordingly, the first target evaluation index may be a pronunciation accuracy score compared to a preset standard pronunciation.
The training process of the target student model in this embodiment can refer to the foregoing contents, and is not described here again.
All the above technical solutions can be combined arbitrarily to form the optional embodiments of the present application, and are not described herein again.
In order to better implement the model training method according to the embodiment of the present application, an embodiment of the present application further provides a model training device, please refer to fig. 4, and fig. 4 is a schematic structural diagram of the model training device according to the embodiment of the present application. The model training device 40 may include:
an obtaining unit 41, configured to obtain a first sample set, where the first sample set includes a plurality of first subsets, each first subset includes a plurality of first sample data sets, each first sample data set includes first object feature information, and a first evaluation index mark corresponding to the first object feature information;
the obtaining unit 41 is further configured to obtain an auxiliary model, a plurality of reference models, and a plurality of unmarked sample feature information, where the reference models are models used to determine a prediction evaluation index corresponding to feature information of an object to be analyzed;
a first training unit 42, configured to train an initial meta-teacher model based on the first sample set, the auxiliary models, and the multiple reference models to obtain a target meta-teacher model, where the auxiliary models are used to assist in training the initial meta-teacher model;
a determining unit 43, configured to determine, by using the plurality of unlabeled sample feature information, the plurality of reference models, and the target meta-teacher model, a plurality of sample evaluation index labels corresponding to the plurality of unlabeled sample feature information;
and the second training unit 44 is configured to train the initial student model according to the multiple unlabeled sample feature information and the multiple sample evaluation index labels to obtain a corresponding target student model, where the target student model is configured to determine a first target evaluation index of the input object feature information according to the input object feature information.
Optionally, when the apparatus is configured to train an initial meta-teacher model based on the first sample set, the auxiliary models, and the plurality of reference models to obtain a target meta-teacher model, the apparatus is specifically configured to:
for each first subset in the first sample set, acquiring a plurality of first sample data groups in the first subset;
for each first sample data group in the plurality of first sample data groups, inputting first object feature information in the first sample data group into each reference model in the plurality of reference models to obtain a first prediction evaluation index vector corresponding to the first sample data group, wherein the first prediction evaluation index vector comprises a plurality of first prediction evaluation indexes, and the first prediction evaluation indexes are in one-to-one correspondence with the reference models;
determining a second sample data group corresponding to the first sample data group according to the first sample data group and a first prediction evaluation index vector corresponding to the first sample data group to obtain a second subset corresponding to the first subset, wherein the second subset comprises a plurality of second sample data groups, and the second sample data group comprises the first sample data group and a first prediction evaluation index vector corresponding to the first sample data group;
determining a second sample set corresponding to the first sample set according to a second subset corresponding to each first subset, wherein the second sample set comprises a plurality of second subsets;
and training the initial meta-teacher model by using the second sample set and the auxiliary model to obtain a target meta-teacher model.
Optionally, when the apparatus is configured to train the initial meta-teacher model by using the second sample set and the auxiliary model to obtain the target meta-teacher model, the apparatus is specifically configured to:
selecting a target subset from a plurality of second subsets included in the second sample set;
grouping at least part of second sample data groups in a plurality of second sample data groups contained in the target subset according to a first preset rule to obtain a target number of task groups corresponding to the first preset rule, wherein each task group comprises a plurality of second sample data groups;
selecting a target task group from the target number of task groups;
selecting a first part of second sample data groups and a second part of second sample data groups from a plurality of second sample data groups included in the target task group according to a second preset rule;
determining a support sample data set based on the first portion of the second sample data set and a query sample data set based on the second portion of the second sample data set;
and training the initial meta-teacher model according to the support sample data set, the query sample data set and the auxiliary model to obtain a target meta-teacher model.
Optionally, the model parameter of the assisted model is first parameter information, and when the apparatus is configured to train the initial meta-teacher model according to the support sample data set, the query sample data set, and the assisted model to obtain the target meta-teacher model, the apparatus is specifically configured to:
acquiring an initial meta-teacher model with model parameters being second parameter information;
determining, by the initial meta-teacher model, a plurality of second evaluation index markers corresponding to the support sample data set based on the support sample data set;
determining, by the auxiliary model, a plurality of third evaluation index markers corresponding to the support sample data set based on the support sample data set;
determining first parameter information to be updated of the auxiliary model according to the support sample data set and the plurality of second evaluation index marks, and obtaining an updated auxiliary model with model parameters of the first parameter information to be updated;
determining a plurality of fourth evaluation index markers corresponding to the query sample data set based on the query sample data set through the updated auxiliary model;
and training an initial meta-teacher model by using the support sample data set, the query sample data set, the plurality of second evaluation index marks, the plurality of third evaluation index marks, the plurality of fourth evaluation index marks, the first parameter information to be updated and the second parameter information to obtain a target meta-teacher model.
Optionally, when the apparatus is configured to train the initial meta-teacher model by using the support sample data set, the query sample data set, the second evaluation index tags, the third evaluation index tags, the fourth evaluation index tags, the first parameter information to be updated, and the second parameter information to obtain the target meta-teacher model, the apparatus is specifically configured to:
determining a first loss value according to a first preset loss function, the support sample data set and the plurality of second evaluation index marks;
determining a second loss value according to a second preset loss function, the support sample data set and the third evaluation index marks;
determining a third loss value according to a third preset loss function, the query sample data set and the plurality of fourth evaluation index marks;
and training an initial meta-teacher model according to the first loss value, the second loss value, the third loss value, the first parameter information to be updated, the second parameter information and the plurality of second evaluation index marks to obtain a target meta-teacher model.
Optionally, when the apparatus is configured to train an initial meta-teacher model according to the first loss value, the second loss value, the third loss value, the first parameter information, the first to-be-updated parameter information, the second parameter information, and the plurality of second evaluation index markers to obtain a target meta-teacher model, the apparatus is specifically configured to:
determining second parameter information to be updated corresponding to the initial meta-teacher model according to the first loss value, the second loss value, the third loss value, the first parameter information to be updated, the second parameter information and the plurality of second evaluation index marks;
and training the initial meta-teacher model by using the second parameter information to be updated and the second parameter information to obtain a target meta-teacher model.
Optionally, when the device is configured to train the initial meta-teacher model by using the second to-be-updated parameter information and the second parameter information to obtain a target meta-teacher model, the device is specifically configured to:
determining whether the absolute value of the difference value between the second parameter information to be updated and the second parameter information is smaller than a first preset value, if so, taking the initial meta teacher model as a target meta teacher model, and outputting the target meta teacher model; if not, the second to-be-updated parameter information is used as second parameter information of the initial meta-teacher model to obtain an updated initial meta-teacher model, an unselected task group is selected from the target number of task groups to serve as a target task group, the first part of second sample data group and the second part of second sample data group are selected from a plurality of second sample data groups included in the target task group according to a second preset rule, and the initial meta-teacher model is used as a target meta-teacher model until the absolute value of the difference value between the second to-be-updated parameter information and the second parameter information is smaller than a first preset value, and the target meta-teacher model is output.
Optionally, the aforementioned apparatus is further configured to: if the task groups which are not selected do not exist in the target number of task groups, selecting an unselected second subset from a plurality of second subsets contained in the second sample set as a target subset, returning and executing grouping of at least part of second sample data groups in the plurality of second sample data groups contained in the target subset according to a first preset rule to obtain the target number of task groups corresponding to the first preset rule, and taking the initial meta-teacher model as a target meta-teacher model and outputting the target meta-teacher model after determining that the absolute value of the difference value between the second parameter information to be updated and the second parameter information is smaller than a first preset value.
Optionally, the aforementioned apparatus is further configured to:
obtaining the updating times of the model parameters of the initial meta-teacher model;
determining whether the updating times are larger than preset times, if so, taking the initial meta-teacher model as a target meta-teacher model, and outputting the target meta-teacher model; if not, starting execution to determine whether the absolute value of the difference value between the second parameter information to be updated and the second parameter information is smaller than a first preset value.
Optionally, when the apparatus is configured to determine, by using the multiple unlabeled sample feature information, the multiple reference models, and the target meta-teacher model, multiple sample evaluation index labels corresponding to the multiple unlabeled sample feature information, the apparatus is specifically configured to:
inputting the unmarked sample characteristic information into the plurality of reference models aiming at each unmarked sample characteristic information in the unmarked sample characteristic information to obtain a target evaluation index vector corresponding to the unmarked sample characteristic information;
determining a plurality of target evaluation index vectors corresponding to the characteristic information of the plurality of unmarked samples according to the target evaluation index vectors corresponding to the characteristic information of the unmarked samples;
and inputting the plurality of unmarked sample characteristic information and the plurality of target evaluation index vectors into the target meta-teacher model to obtain a plurality of sample evaluation index marks corresponding to the plurality of unmarked sample characteristic information.
Optionally, the initial meta-teacher model includes: the device is specifically configured to, when configured to determine, by the initial meta-teacher model, a plurality of second evaluation index markers corresponding to the support sample data set based on the support sample data set,:
for each third sample data group in a plurality of third sample data groups contained in the support sample data group, obtaining second object characteristic information in the third sample data group and a second prediction evaluation index vector corresponding to the second object characteristic information to obtain a plurality of second object characteristic information and a plurality of second prediction evaluation index vectors corresponding to the support sample data group;
determining, by the initial meta-teacher model, a plurality of second evaluation index markers corresponding to the support sample data set based on the plurality of second object feature information and the plurality of second prediction evaluation index vectors.
To better implement the data processing method according to the embodiment of the present application, a data processing apparatus is further provided in the embodiment of the present application, please refer to fig. 5, and fig. 5 is a schematic structural diagram of the data processing apparatus according to the embodiment of the present application. The data processing apparatus 50 may include:
an acquisition unit 51 configured to acquire object feature information to be processed;
the input unit 52 is configured to input the to-be-processed object feature information into a target student model, so as to obtain a first target evaluation index corresponding to the to-be-processed object feature information;
the target student model is a target student model trained by the model training method.
The various elements of the model training device 40, and the data processing device 50 described above may be implemented in whole or in part by software, hardware, and combinations thereof. The units may be embedded in hardware or independent from a processor in the computer device, or may be stored in a memory in the computer device in software, so that the processor can call and execute operations corresponding to the units.
The model training device 40 and/or the data processing device 50 may be integrated in a terminal or a server having a memory and a processor and having an arithmetic capability, or the model training device 40 and/or the data processing device 50 may be the terminal or the server.
Optionally, the present application further provides a computer device, which includes a memory and a processor, where the memory stores a computer program, and the processor implements the steps in the foregoing method embodiments when executing the computer program.
Fig. 6 is a schematic structural diagram of a computer device according to an embodiment of the present application, where the computer device may be the terminal or the server shown in fig. 1. As shown in fig. 6, the computer device 600 may include: a communication interface 601, a memory 602, a processor 603, and a communication bus 604. The communication interface 601, the memory 602, and the processor 603 communicate with each other via a communication bus 604. The communication interface 601 is used for data communication between the computer apparatus 600 and external apparatuses. The memory 602 may be used for storing software programs and modules, and the processor 603 may operate by executing the software programs and modules stored in the memory 602, such as the software programs of the corresponding operations in the foregoing method embodiments.
Alternatively, the processor 603 may call the software programs and modules stored in the memory 602 to perform the following operations:
acquiring a first sample set, wherein the first sample set comprises a plurality of first sub-sets, each first sub-set comprises a plurality of first sample data groups, each first sample data group comprises first object characteristic information and a first evaluation index mark corresponding to the first object characteristic information;
the method comprises the steps of obtaining an auxiliary model, a plurality of reference models and a plurality of unmarked sample characteristic information, wherein the reference models are models used for determining a prediction evaluation index corresponding to characteristic information of an object to be analyzed;
training an initial meta-teacher model based on the first sample set, the auxiliary models and the plurality of reference models to obtain a target meta-teacher model, wherein the auxiliary models are used for assisting in training the initial meta-teacher model;
determining a plurality of sample evaluation index marks corresponding to the plurality of unmarked sample characteristic information by using the plurality of unmarked sample characteristic information, the plurality of reference models and the target meta-teacher model;
training an initial student model according to the plurality of label-free sample characteristic information and the plurality of sample evaluation index labels to obtain a corresponding target student model, wherein the target student model is used for determining a first target evaluation index of the input object characteristic information according to the input object characteristic information.
Optionally, the processor 603 may also call the software programs and modules stored in the memory 602 to perform the following operations: acquiring characteristic information of an object to be processed; inputting the object characteristic information to be processed into a target student model to obtain a first target evaluation index corresponding to the object characteristic information to be processed; the target student model is trained by the model training method.
The present application also provides a computer-readable storage medium for storing a computer program. The computer-readable storage medium can be applied to a computer device, and the computer program enables the computer device to execute corresponding processes in the methods in the embodiments of the present application, which are not described herein again for brevity.
The present application also provides a computer program product comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and executes the computer instruction, so that the computer device executes corresponding processes in the methods in the embodiments of the present application, which is not described herein again for brevity.
The present application also provides a computer program comprising computer instructions stored in a computer readable storage medium. The processor of the computer device reads the computer instruction from the computer-readable storage medium, and executes the computer instruction, so that the computer device executes corresponding processes in the methods in the embodiments of the present application, which is not described herein again for brevity.
It should be understood that the processor of the embodiments of the present application may be an integrated circuit chip having signal processing capabilities. In implementation, the steps of the above method embodiments may be performed by integrated logic circuits of hardware in a processor or instructions in the form of software. The Processor may be a general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), an off the shelf Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components. The various methods, steps, and logic blocks disclosed in the embodiments of the present application may be implemented or performed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in connection with the embodiments of the present application may be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software module may be located in ram, flash memory, rom, prom, or eprom, registers, etc. storage media as is well known in the art. The storage medium is located in a memory, and a processor reads information in the memory and completes the steps of the method in combination with hardware of the processor.
It will be appreciated that the memory in the embodiments of the subject application can be either volatile memory or nonvolatile memory, or can include both volatile and nonvolatile memory. The non-volatile Memory may be a Read-Only Memory (ROM), a Programmable ROM (PROM), an Erasable PROM (EPROM), an Electrically Erasable PROM (EEPROM), or a flash Memory. The volatile Memory may be a Random Access Memory (RAM) which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as Static random access memory (Static RAM, SRAM), Dynamic Random Access Memory (DRAM), Synchronous Dynamic random access memory (Synchronous DRAM, SDRAM), Double Data Rate Synchronous Dynamic random access memory (DDR SDRAM), Enhanced Synchronous SDRAM (ESDRAM), Synchronous link SDRAM (SLDRAM), and Direct Rambus RAM (DR RAM). It should be noted that the memory of the systems and methods described herein is intended to comprise, without being limited to, these and any other suitable types of memory.
It should be understood that the above memories are exemplary but not limiting illustrations, for example, the memories in the embodiments of the present application may also be Static Random Access Memory (SRAM), dynamic random access memory (dynamic RAM, DRAM), Synchronous Dynamic Random Access Memory (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (enhanced SDRAM, ESDRAM), Synchronous Link DRAM (SLDRAM), Direct Rambus RAM (DR RAM), and the like. That is, the memory in the embodiments of the present application is intended to comprise, without being limited to, these and any other suitable types of memory.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present application.
It is clear to those skilled in the art that, for convenience and brevity of description, the specific working processes of the above-described systems, apparatuses and units may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the several embodiments provided in the present application, it should be understood that the disclosed system, apparatus and method may be implemented in other ways. For example, the above-described apparatus embodiments are merely illustrative, and for example, the division of the units is only one type of logical functional division, and other divisions may be realized in practice, for example, multiple units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application or portions thereof that substantially contribute to the prior art may be embodied in the form of a software product stored in a storage medium and including instructions for causing a computer device (which may be a personal computer or a server) to execute all or part of the steps of the method according to the embodiments of the present application. And the aforementioned storage medium includes: a U disk, a removable hard disk, a ROM, a RAM, a magnetic disk or an optical disk, and various media capable of storing program codes.
The above description is only for the specific embodiments of the present application, but the scope of the present application is not limited thereto, and any person skilled in the art can easily conceive of the changes or substitutions within the technical scope of the present application, and shall be covered by the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.
Claims (15)
1. A method of model training, the method comprising:
acquiring a first sample set, wherein the first sample set comprises a plurality of first sub-sets, each first sub-set comprises a plurality of first sample data groups, each first sample data group comprises first object characteristic information and a first evaluation index mark corresponding to the first object characteristic information;
the method comprises the steps of obtaining an auxiliary model, a plurality of reference models and a plurality of unmarked sample characteristic information, wherein the reference models are models used for determining a prediction evaluation index corresponding to characteristic information of an object to be analyzed;
training an initial meta-teacher model based on the first sample set, the auxiliary models and the plurality of reference models to obtain a target meta-teacher model, wherein the auxiliary models are used for assisting in training the initial meta-teacher model;
determining a plurality of sample evaluation index marks corresponding to the plurality of unmarked sample characteristic information by using the plurality of unmarked sample characteristic information, the plurality of reference models and the target meta-teacher model;
training an initial student model according to the plurality of label-free sample characteristic information and the plurality of sample evaluation index labels to obtain a corresponding target student model, wherein the target student model is used for determining a first target evaluation index of the input object characteristic information according to the input object characteristic information.
2. The method of claim 1, wherein training an initial meta-teacher model based on the first set of samples, the auxiliary models, and the plurality of reference models to obtain a target meta-teacher model comprises:
for each first subset in the first sample set, acquiring a plurality of first sample data groups in the first subset;
for each first sample data group in the plurality of first sample data groups, inputting first object feature information in the first sample data group into each reference model in the plurality of reference models to obtain a first prediction evaluation index vector corresponding to the first sample data group, wherein the first prediction evaluation index vector comprises a plurality of first prediction evaluation indexes, and the first prediction evaluation indexes are in one-to-one correspondence with the reference models;
determining a second sample data group corresponding to the first sample data group according to the first sample data group and a first prediction evaluation index vector corresponding to the first sample data group to obtain a second subset corresponding to the first subset, wherein the second subset comprises a plurality of second sample data groups, and the second sample data group comprises the first sample data group and a first prediction evaluation index vector corresponding to the first sample data group;
determining a second sample set corresponding to the first sample set according to a second subset corresponding to each first subset, wherein the second sample set comprises a plurality of second subsets;
and training the initial meta-teacher model by using the second sample set and the auxiliary model to obtain a target meta-teacher model.
3. The method of claim 2, wherein training an initial meta-teacher model using the second set of samples and the auxiliary models to obtain a target meta-teacher model comprises:
selecting a target subset from a plurality of second subsets included in the second sample set;
grouping at least part of second sample data groups in a plurality of second sample data groups contained in the target subset according to a first preset rule to obtain a target number of task groups corresponding to the first preset rule, wherein each task group comprises a plurality of second sample data groups;
selecting a target task group from the target number of task groups;
selecting a first part of second sample data groups and a second part of second sample data groups from a plurality of second sample data groups included in the target task group according to a second preset rule;
determining a set of support sample data based on the first portion of the second set of sample data and a set of query sample data based on the second portion of the second set of sample data;
and training the initial meta-teacher model according to the support sample data set, the query sample data set and the auxiliary model to obtain a target meta-teacher model.
4. The method of claim 3, wherein the model parameters of the auxiliary model are first parameter information, and the training of the initial meta-teacher model according to the support sample data set, the query sample data set, and the auxiliary model to obtain the target meta-teacher model comprises:
acquiring an initial meta-teacher model with model parameters being second parameter information;
determining, by the initial meta teacher model, a plurality of second evaluation index markers corresponding to the support sample data group based on the support sample data group;
determining, by the auxiliary model, a plurality of third evaluation index markers corresponding to the support sample data set based on the support sample data set;
determining first parameter information to be updated of the auxiliary model according to the support sample data set and the plurality of second evaluation index marks, and obtaining an updated auxiliary model with model parameters of the first parameter information to be updated;
determining a plurality of fourth evaluation index markers corresponding to the query sample data set based on the query sample data set through the updated auxiliary model;
and training an initial meta-teacher model by using the support sample data set, the query sample data set, the plurality of second evaluation index marks, the plurality of third evaluation index marks, the plurality of fourth evaluation index marks, the first parameter information to be updated and the second parameter information to obtain a target meta-teacher model.
5. The method of claim 4, wherein training an initial meta-teacher model with the support sample data set, the query sample data set, the second evaluation index flags, the third evaluation index flags, the fourth evaluation index flags, the first parameter information to be updated, and the second parameter information to obtain a target meta-teacher model comprises:
determining a first loss value according to a first preset loss function, the support sample data set and the plurality of second evaluation index marks;
determining a second loss value according to a second preset loss function, the support sample data set and the third evaluation index marks;
determining a third loss value according to a third preset loss function, the query sample data set and the plurality of fourth evaluation index marks;
and training an initial meta-teacher model according to the first loss value, the second loss value, the third loss value, the first parameter information, the first to-be-updated parameter information, the second parameter information and the plurality of second evaluation index marks to obtain a target meta-teacher model.
6. The method of claim 5, wherein training an initial meta-teacher model according to the first loss value, the second loss value, the third loss value, the first parameter information to be updated, the second parameter information, and the plurality of second evaluation index tags to obtain a target meta-teacher model comprises:
determining second parameter information to be updated corresponding to the initial meta-teacher model according to the first loss value, the second loss value, the third loss value, the first parameter information to be updated, the second parameter information and the plurality of second evaluation index marks;
and training the initial meta-teacher model by using the second parameter information to be updated and the second parameter information to obtain a target meta-teacher model.
7. The method of claim 6, wherein training the initial meta-teacher model with the second parameter information to be updated and the second parameter information to obtain a target meta-teacher model comprises:
determining whether the absolute value of the difference value between the second parameter information to be updated and the second parameter information is smaller than a first preset value, if so, taking the initial meta teacher model as a target meta teacher model, and outputting the target meta teacher model; if not, the second to-be-updated parameter information is used as second parameter information of the initial meta-teacher model to obtain an updated initial meta-teacher model, an unselected task group is selected from the target number of task groups to serve as a target task group, the first part of second sample data group and the second part of second sample data group are selected from a plurality of second sample data groups included in the target task group according to a second preset rule, and the initial meta-teacher model is used as a target meta-teacher model until the absolute value of the difference value between the second to-be-updated parameter information and the second parameter information is smaller than a first preset value, and the target meta-teacher model is output.
8. The method of claim 7, further comprising:
if the task groups which are not selected do not exist in the target number of task groups, selecting an unselected second subset from a plurality of second subsets contained in the second sample set as a target subset, returning and executing grouping of at least part of second sample data groups in the plurality of second sample data groups contained in the target subset according to a first preset rule to obtain the target number of task groups corresponding to the first preset rule, and taking the initial meta-teacher model as a target meta-teacher model and outputting the target meta-teacher model after determining that the absolute value of the difference value between the second parameter information to be updated and the second parameter information is smaller than a first preset value.
9. The method according to claim 7 or 8, characterized in that the method further comprises:
obtaining the updating times of the model parameters of the initial meta-teacher model;
determining whether the updating times are larger than preset times, if so, taking the initial meta-teacher model as a target meta-teacher model, and outputting the target meta-teacher model; if not, starting to execute and determining whether the absolute value of the difference value between the second parameter information to be updated and the second parameter information is smaller than a first preset value or not.
10. The method of any of claims 1-8, wherein determining a plurality of sample evaluation index labels to which the plurality of unlabeled sample feature information corresponds using the plurality of unlabeled sample feature information, the plurality of reference models, and the target meta-teacher model comprises:
inputting the unmarked sample characteristic information into the plurality of reference models aiming at each unmarked sample characteristic information in the unmarked sample characteristic information to obtain a target evaluation index vector corresponding to the unmarked sample characteristic information;
determining a plurality of target evaluation index vectors corresponding to the characteristic information of the plurality of unmarked samples according to the target evaluation index vectors corresponding to the characteristic information of the unmarked samples;
and inputting the plurality of unmarked sample characteristic information and the plurality of target evaluation index vectors into the target meta-teacher model to obtain a plurality of sample evaluation index marks corresponding to the plurality of unmarked sample characteristic information.
11. The method of any of claims 4-8, the initial meta teacher model comprising: a weight determination unit and an evaluation index flag determination unit, wherein the determining, by the initial meta-teacher model, a plurality of second evaluation index flags corresponding to the support sample data set based on the support sample data set includes:
for each third sample data group in a plurality of third sample data groups contained in the support sample data group, obtaining second object characteristic information in the third sample data group and a second prediction evaluation index vector corresponding to the second object characteristic information to obtain a plurality of second object characteristic information and a plurality of second prediction evaluation index vectors corresponding to the support sample data group;
determining, by the initial meta-teacher model, a plurality of second evaluation index markers corresponding to the support sample data set based on the plurality of second object feature information and the plurality of second prediction evaluation index vectors.
12. A method of data processing, the method comprising:
acquiring characteristic information of an object to be processed;
inputting the object characteristic information to be processed into a target student model to obtain a first target evaluation index corresponding to the object characteristic information to be processed;
wherein the objective student model is an objective student model trained by the model training method according to any one of claims 1 to 11.
13. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program adapted to be loaded by a processor to perform the method according to any of claims 1-12.
14. A computer device, characterized in that the computer device comprises a processor and a memory, in which a computer program is stored, the processor being adapted to perform the method according to any of claims 1-12 by calling the computer program stored in the memory.
15. A computer program product comprising computer instructions, characterized in that the computer instructions, when executed by a processor, implement the method according to any of claims 1-12.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210753985.XA CN115099988A (en) | 2022-06-28 | 2022-06-28 | Model training method, data processing method, device and computer medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210753985.XA CN115099988A (en) | 2022-06-28 | 2022-06-28 | Model training method, data processing method, device and computer medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115099988A true CN115099988A (en) | 2022-09-23 |
Family
ID=83294212
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210753985.XA Pending CN115099988A (en) | 2022-06-28 | 2022-06-28 | Model training method, data processing method, device and computer medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115099988A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116579651A (en) * | 2023-05-11 | 2023-08-11 | 中国矿业报社 | Mining project evaluation method based on semi-supervised learning |
Citations (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111027870A (en) * | 2019-12-14 | 2020-04-17 | 支付宝(杭州)信息技术有限公司 | User risk assessment method and device, electronic equipment and storage medium |
| CN111080123A (en) * | 2019-12-14 | 2020-04-28 | 支付宝(杭州)信息技术有限公司 | User risk assessment method and device, electronic equipment and storage medium |
| CN111105008A (en) * | 2018-10-29 | 2020-05-05 | 富士通株式会社 | Model training method, data recognition method and data recognition device |
| CN111126592A (en) * | 2018-10-30 | 2020-05-08 | 三星电子株式会社 | Method and apparatus for outputting prediction result, method and apparatus for generating neural network, and storage medium |
| CN112182362A (en) * | 2020-08-31 | 2021-01-05 | 华为技术有限公司 | Method and device for training model for online click rate prediction and recommendation system |
| CN112749728A (en) * | 2020-08-13 | 2021-05-04 | 腾讯科技（深圳）有限公司 | Student model training method and device, computer equipment and storage medium |
| CN113139500A (en) * | 2021-05-10 | 2021-07-20 | 重庆中科云从科技有限公司 | Smoke detection method, system, medium and device |
| 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 |
| CN114169392A (en) * | 2021-10-29 | 2022-03-11 | 阿里巴巴（中国）有限公司 | Model training method and device, task processing method, storage medium and processor |
| CN114330510A (en) * | 2021-12-06 | 2022-04-12 | 北京大学 | Model training method and device, electronic equipment and storage medium |
| CN114519416A (en) * | 2021-12-30 | 2022-05-20 | 深圳云天励飞技术股份有限公司 | Model distillation method and device and electronic equipment |
| WO2022105173A1 (en) * | 2020-11-20 | 2022-05-27 | 平安科技（深圳）有限公司 | Model distillation method and apparatus, and storage medium and device |
| CN114611672A (en) * | 2022-03-16 | 2022-06-10 | 腾讯科技（深圳）有限公司 | Model training method, face recognition method and device |
-
2022
- 2022-06-28 CN CN202210753985.XA patent/CN115099988A/en active Pending
Patent Citations (14)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111105008A (en) * | 2018-10-29 | 2020-05-05 | 富士通株式会社 | Model training method, data recognition method and data recognition device |
| CN111126592A (en) * | 2018-10-30 | 2020-05-08 | 三星电子株式会社 | Method and apparatus for outputting prediction result, method and apparatus for generating neural network, and storage medium |
| CN111080123A (en) * | 2019-12-14 | 2020-04-28 | 支付宝(杭州)信息技术有限公司 | User risk assessment method and device, electronic equipment and storage medium |
| CN111027870A (en) * | 2019-12-14 | 2020-04-17 | 支付宝(杭州)信息技术有限公司 | User risk assessment method and device, electronic equipment and storage medium |
| CN112749728A (en) * | 2020-08-13 | 2021-05-04 | 腾讯科技（深圳）有限公司 | Student model training method and device, computer equipment and storage medium |
| CN112182362A (en) * | 2020-08-31 | 2021-01-05 | 华为技术有限公司 | Method and device for training model for online click rate prediction and recommendation system |
| WO2022105173A1 (en) * | 2020-11-20 | 2022-05-27 | 平安科技（深圳）有限公司 | Model distillation method and apparatus, and storage medium and device |
| CN113139500A (en) * | 2021-05-10 | 2021-07-20 | 重庆中科云从科技有限公司 | Smoke detection method, system, medium and device |
| US20220129731A1 (en) * | 2021-05-27 | 2022-04-28 | Beijing Baidu Netcom Science Technology Co., Ltd. | Method and apparatus for training image recognition model, and method and apparatus for recognizing image |
| 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 |
| CN114169392A (en) * | 2021-10-29 | 2022-03-11 | 阿里巴巴（中国）有限公司 | Model training method and device, task processing method, storage medium and processor |
| CN114330510A (en) * | 2021-12-06 | 2022-04-12 | 北京大学 | Model training method and device, electronic equipment and storage medium |
| CN114519416A (en) * | 2021-12-30 | 2022-05-20 | 深圳云天励飞技术股份有限公司 | Model distillation method and device and electronic equipment |
| CN114611672A (en) * | 2022-03-16 | 2022-06-10 | 腾讯科技（深圳）有限公司 | Model training method, face recognition method and device |
Non-Patent Citations (1)
| Title |
|---|
| 葛仕明;赵胜伟;刘文瑜;李晨钰;: "基于深度特征蒸馏的人脸识别", 北京交通大学学报, no. 06 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116579651A (en) * | 2023-05-11 | 2023-08-11 | 中国矿业报社 | Mining project evaluation method based on semi-supervised learning |
| CN116579651B (en) * | 2023-05-11 | 2023-11-10 | 中国矿业报社 | Mining project evaluation method based on semi-supervised learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110070391B (en) | Data processing method and device, computer readable medium and electronic equipment | |
| CN104781837B (en) | System and method for forming predictions using event-based sentiment analysis | |
| CN109165840A (en) | Risk profile processing method, device, computer equipment and medium | |
| CN109766454A (en) | A kind of investor's classification method, device, equipment and medium | |
| Winter et al. | Trusted artificial intelligence: Towards certification of machine learning applications | |
| CN112182384B (en) | Content recommendation method and device based on countermeasure learning and computer equipment | |
| CN112819024A (en) | Model processing method, user data processing method and device and computer equipment | |
| CN113313479A (en) | Payment service big data processing method and system based on artificial intelligence | |
| CN113987182A (en) | Fraud entity identification method, device and related equipment based on security intelligence | |
| CN111709225A (en) | Event cause and effect relationship judging method and device and computer readable storage medium | |
| CN115099988A (en) | Model training method, data processing method, device and computer medium | |
| Kavikondala et al. | Automated retraining of machine learning models | |
| CN114580794B (en) | Data processing method, apparatus, program product, computer device and medium | |
| CN116523622A (en) | Object risk prediction method and device, electronic equipment and storage medium | |
| CN114692785A (en) | Behavior classification method, device, equipment and storage medium | |
| CN116205700A (en) | Recommendation method and device for target product, computer equipment and storage medium | |
| Kalia | Potential Impact of Generative Artificial Intelligence (AI) on the Financial Industry | |
| CN115496638B (en) | Student course score analysis management method and system based on smart campus | |
| Abdullah et al. | A decision making method with triangular fuzzy numbers for unraveling the criteria of e-commerce | |
| CN114328797B (en) | Content search method, device, electronic apparatus, storage medium, and program product | |
| CN116628236B (en) | Method and device for delivering multimedia information, electronic equipment and storage medium | |
| Dehuri et al. | Advances in Machine Learning for Big Data Analysis | |
| CN117036022A (en) | Financing risk prediction method and device, storage medium and electronic equipment | |
| Kumar et al. | Role of Explainable Edge AI to Resolve Real Time Problem | |
| Abuzayeda et al. | Advancing Student Guidance Using Classification Data Mining Techniques |
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 | ||
| REG | Reference to a national code |
Ref country code: HK Ref legal event code: DE Ref document number: 40074455 Country of ref document: HK |