CN114897158A

CN114897158A - Training method of data processing model, data processing method, device and equipment

Info

Publication number: CN114897158A
Application number: CN202210491871.2A
Authority: CN
Inventors: 王晓星; 初祥祥; 魏晓林
Original assignee: Beijing Sankuai Online Technology Co Ltd
Current assignee: Beijing Sankuai Online Technology Co Ltd
Priority date: 2022-05-07
Filing date: 2022-05-07
Publication date: 2022-08-12

Abstract

The application discloses a training method of a data processing model, a data processing method, a device and equipment, and belongs to the technical field of artificial intelligence. The method comprises the following steps: acquiring a first data enhancement parameter, a first network model and first training data corresponding to current training, wherein the first data enhancement parameter is used for representing data enhancement processing information; performing data enhancement processing on the first training data based on the first data enhancement parameters to obtain the first training data after the data enhancement processing; adjusting model parameters of the first network model based on the first training data after data enhancement processing to obtain a second network model; and in response to the training end condition being met, taking the second network model as a data processing model. Because the data processing model is obtained by training based on the first training data after data enhancement processing, the prediction capability of the model is improved, the robustness and the accuracy of the model are improved, and the accuracy of a data processing result is improved.

Description

Training method of data processing model, data processing method, device and equipment

Technical Field

The embodiment of the application relates to the technical field of artificial intelligence, in particular to a training method of a data processing model, a data processing method, a data processing device and equipment.

Background

With the continuous development of artificial intelligence technology, the types and functions of neural network models are more and more, and common neural network models include classification models, recognition models, recommendation models and the like. The trained neural network model can perform data processing on the input data, and thus the trained neural network model can also be called a data processing model.

In the related art, a neural network model is established, and meanwhile, training data and a labeling result of the training data are obtained, so that the neural network model is trained for multiple times by using the training data and the labeling result of the training data to obtain a data processing model. The method comprises the steps of carrying out one-time training on a neural network model, namely inputting training data into the neural network model, outputting a prediction result of the training data by the neural network model, and carrying out one-time adjustment on model parameters of the neural network model based on the prediction result of the training data and a labeling result of the training data.

The above training mode may cause the neural network model to have poor robustness, thereby affecting the accuracy of the data processing model.

Disclosure of Invention

The embodiment of the application provides a training method, a data processing method, a device and equipment of a data processing model, which can be used for solving the problems in the related technology.

In one aspect, an embodiment of the present application provides a method for training a data processing model, where the method includes:

acquiring a first data enhancement parameter, a first network model and first training data corresponding to current training, wherein the first data enhancement parameter is used for representing data enhancement processing information;

performing data enhancement processing on the first training data based on the first data enhancement parameter to obtain first training data subjected to data enhancement processing;

adjusting model parameters of the first network model based on the first training data after the data enhancement processing to obtain a second network model;

and in response to the training end condition being met, taking the second network model as a data processing model.

In one possible implementation, the first data enhancement parameter includes a probability of being selected for at least one candidate data enhancement process;

the data enhancement processing is performed on the first training data based on the first data enhancement parameter to obtain the first training data after the data enhancement processing, and the data enhancement processing includes:

determining a target data enhancement treatment from the at least one candidate data enhancement treatment based on the probability of being selected of the at least one candidate data enhancement treatment;

and performing target data enhancement processing on the first training data to obtain the first training data after the data enhancement processing.

In a possible implementation manner, the first data enhancement parameter further includes an executed probability of each sub-data enhancement processing in the target data enhancement processing;

the performing target data enhancement processing on the first training data to obtain the first training data after the data enhancement processing includes:

determining target sub-data enhancement processing from the respective sub-data enhancement processing based on the executed probability of the respective sub-data enhancement processing;

and performing target subdata enhancement processing on the first training data to obtain the first training data after the data enhancement processing.

In a possible implementation manner, the adjusting the model parameter of the first network model based on the first training data after the data enhancement processing to obtain a second network model includes:

inputting the first training data subjected to the data enhancement processing into the first network model to obtain a prediction result of the first training data subjected to the data enhancement processing;

acquiring a labeling result of the first training data;

and adjusting model parameters of the first network model based on the prediction result of the first training data after the data enhancement processing and the labeling result of the first training data to obtain a second network model.

In a possible implementation manner, after obtaining the second network model, the method further includes:

responding to the condition that the training end condition is not met, and acquiring first verification data of the current training;

determining a loss value of the second network model corresponding to the first verification data based on the first verification data and the second network model;

and determining a second data enhancement parameter of the next training based on the loss value of the second network model corresponding to the first verification data, wherein the second data enhancement parameter is used for representing information of data enhancement processing.

In a possible implementation manner, the determining, based on the first verification data and the second network model, a loss value of the second network model corresponding to the first verification data includes:

performing data enhancement processing on the first verification data based on the first data enhancement parameter to obtain first verification data subjected to data enhancement processing;

and determining a loss value of the second network model corresponding to the first verification data based on the first verification data and the second network model after the data enhancement processing.

In one possible implementation, obtaining the first network model includes:

acquiring a candidate network model and a first network structure parameter corresponding to the current training, wherein the first network structure parameter is used for representing the structure information of the first network model;

and sampling the candidate network model based on the first network structure parameter to obtain the first network model.

In a possible implementation manner, the candidate network model includes a plurality of candidate edges, the candidate edge characterization performs operation processing on the feature vector, and the first network structure parameter includes an importance parameter of each candidate edge;

the sampling processing is performed on the candidate network model based on the first network structure parameter to obtain the first network model, and the sampling processing includes:

sampling the candidate edges based on the importance parameters of the candidate edges to obtain a plurality of target edges;

a first network model is determined based on the plurality of target edges.

In a possible implementation manner, any one candidate edge corresponds to at least one candidate operation processing, and the first network structure parameter further includes an importance parameter of various candidate operation processing corresponding to any one target edge;

the determining a first network model based on the plurality of target edges includes:

for any one target edge, sampling various candidate operation processes corresponding to any one target edge based on the importance parameters of the various candidate operation processes corresponding to any one target edge to obtain a target operation process corresponding to any one target edge;

and determining the first network model based on target operation processing corresponding to the plurality of target edges.

In one possible implementation, the obtaining the first network structure parameter includes:

acquiring second verification data corresponding to the last training;

determining a loss value of the first network model corresponding to the second verification data based on the second verification data and the first network model;

and determining the first network structure parameter based on the loss value of the first network model corresponding to the second verification data.

In one possible implementation manner, the obtaining the first data enhancement parameter includes:

acquiring second verification data corresponding to the last training;

and determining the first data enhancement parameter based on the loss value of the first network model corresponding to the second verification data.

In another aspect, an embodiment of the present application provides a data processing method, where the method includes:

acquiring target data;

and inputting the target data into a data processing model to obtain a data processing result of the target data, wherein the data processing model is obtained by training according to the training method of the data processing model.

In another aspect, an embodiment of the present application provides a training apparatus for a data processing model, where the apparatus includes:

the acquisition module is used for acquiring a first data enhancement parameter, a first network model and first training data corresponding to current training, wherein the first data enhancement parameter is used for representing information of data enhancement processing;

the data enhancement module is used for carrying out data enhancement processing on the first training data based on the first data enhancement parameter to obtain first training data after the data enhancement processing;

the adjusting module is used for adjusting model parameters of the first network model based on the first training data after the data enhancement processing to obtain a second network model;

and the determining module is used for responding to the condition that the training is finished and taking the second network model as a data processing model.

the data enhancement module is used for determining target data enhancement processing from the at least one candidate data enhancement processing based on the selected probability of the at least one candidate data enhancement processing; and performing target data enhancement processing on the first training data to obtain the first training data after the data enhancement processing.

the data enhancement module is used for determining target sub-data enhancement processing from each sub-data enhancement processing based on the executed probability of each sub-data enhancement processing; and performing target subdata enhancement processing on the first training data to obtain the first training data after the data enhancement processing.

In a possible implementation manner, the adjusting module is configured to input the first training data after the data enhancement processing into the first network model to obtain a prediction result of the first training data after the data enhancement processing; acquiring a labeling result of the first training data; and adjusting model parameters of the first network model based on the prediction result of the first training data after the data enhancement processing and the labeling result of the first training data to obtain a second network model.

In a possible implementation manner, the obtaining module is further configured to obtain first verification data of the current training in response to that the training end condition is not satisfied;

the determining module is configured to determine, based on the first verification data and the second network model, a loss value of the second network model corresponding to the first verification data; and determining a second data enhancement parameter for next training based on the loss value of the second network model corresponding to the first verification data, wherein the second data enhancement parameter is used for representing information of data enhancement processing.

In a possible implementation manner, the determining module is configured to perform data enhancement processing on the first verification data based on the first data enhancement parameter to obtain first verification data after the data enhancement processing; and determining a loss value of the second network model corresponding to the first verification data based on the first verification data and the second network model after the data enhancement processing.

In a possible implementation manner, the obtaining module is configured to obtain a candidate network model and a first network structure parameter corresponding to the current training, where the first network structure parameter is used to represent structure information of the first network model; and sampling the candidate network model based on the first network structure parameter to obtain the first network model.

the acquisition module is used for sampling the candidate edges based on the importance parameters of the candidate edges to obtain a plurality of target edges; a first network model is determined based on the plurality of target edges.

In a possible implementation manner, any one of the candidate edges corresponds to at least one of the candidate arithmetic processings, and the first network structure parameter further includes an importance parameter of each of the candidate arithmetic processings corresponding to any one of the target edges;

the acquisition module is used for sampling various candidate operation processes corresponding to any target edge based on the importance parameters of the various candidate operation processes corresponding to any target edge to obtain the target operation process corresponding to any target edge; and determining the first network model based on target operation processing corresponding to the plurality of target edges.

In a possible implementation manner, the obtaining module is configured to obtain second verification data corresponding to a last training; determining a loss value of the first network model corresponding to the second verification data based on the second verification data and the first network model; and determining the first network structure parameter based on the loss value of the first network model corresponding to the second verification data.

In a possible implementation manner, the obtaining module is configured to obtain second verification data corresponding to a last training; determining a loss value of the first network model corresponding to the second verification data based on the second verification data and the first network model; and determining the first data enhancement parameter based on the loss value of the first network model corresponding to the second verification data.

In another aspect, an embodiment of the present application provides a data processing apparatus, where the apparatus includes:

the acquisition module is used for acquiring target data;

and the obtaining module is used for inputting the target data into a data processing model to obtain a data processing result of the target data, and the data processing model is obtained by training according to any one of the training methods of the data processing model.

In another aspect, an embodiment of the present application provides an electronic device, where the electronic device includes a processor and a memory, where the memory stores at least one program code, and the at least one program code is loaded and executed by the processor, so that the electronic device implements any one of the above-mentioned training methods for a data processing model or any one of the above-mentioned data processing methods.

In another aspect, a computer-readable storage medium is provided, in which at least one program code is stored, and the at least one program code is loaded and executed by a processor, so as to enable a computer to implement any one of the above-mentioned training methods for a data processing model or any one of the above-mentioned data processing methods.

In another aspect, a computer program or a computer program product is provided, in which at least one computer instruction is stored, and the at least one computer instruction is loaded and executed by a processor, so as to enable a computer to implement any one of the above-mentioned training methods for a data processing model or any one of the above-mentioned data processing methods.

The technical scheme provided by the embodiment of the application at least has the following beneficial effects:

according to the technical scheme, during each training, data enhancement processing is performed on first training data corresponding to the current training based on first data enhancement parameters corresponding to the current training, then model parameters of a first network model corresponding to the current training are adjusted based on the first training data after the data enhancement processing, the prediction capability of the model is improved, the robustness and the accuracy of the model are improved, and the accuracy of a data processing result is improved when the data processing is performed based on the data processing model.

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 without creative efforts.

Fig. 1 is a schematic diagram of an implementation environment of a training method of a data processing model or a data processing method provided in an embodiment of the present application;

FIG. 2 is a flow chart of a method for training a data processing model according to an embodiment of the present disclosure;

fig. 3 is a flowchart of a data processing method provided in an embodiment of the present application;

FIG. 4 is a schematic structural diagram of a training apparatus for a data processing model according to an embodiment of the present disclosure;

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 terminal device according to an embodiment of the present application;

fig. 7 is a schematic structural diagram of a server according to an embodiment of the present application.

Detailed Description

To make the objects, technical solutions and advantages of the present application more clear, embodiments of the present application will be described in further detail below with reference to the accompanying drawings.

Fig. 1 is a schematic diagram of an implementation environment of a training method of a data processing model or a data processing method provided in an embodiment of the present application, where the implementation environment includes an electronic device 11 as shown in fig. 1, and the training method or the data processing method of the data processing model in the embodiment of the present application may be executed by the electronic device 11. Illustratively, the electronic device 11 may include at least one of a terminal device or a server.

The terminal device may be at least one of a smartphone, a gaming console, a desktop computer, a tablet computer, and a laptop portable computer. The server may be one server, or a server cluster formed by multiple servers, or any one of a cloud computing platform and a virtualization center, which is not limited in this embodiment of the present application. The server can be in communication connection with the terminal device through a wired network or a wireless network. The server may have functions of data processing, data storage, data transceiving, and the like, and is not limited in the embodiment of the present application.

Based on the foregoing implementation environment, the present application provides a method for training a data processing model, which can be executed by the electronic device 11 in fig. 1, taking a flowchart of the method for training a data processing model provided in the present application as an example, as shown in fig. 2. As shown in fig. 2, the method includes steps 201 to 204.

Step 201, obtaining a first data enhancement parameter, a first network model and first training data corresponding to the current training, where the first data enhancement parameter is used to represent information of data enhancement processing.

The current training is any one of the first training to the last training. For the current training, a first data enhancement parameter corresponding to the current training, a first network model corresponding to the current training, and first training data corresponding to the current training need to be obtained. The number of the first training data is multiple, and the data type and the acquisition mode of the first training data are not limited in the embodiment of the application. Illustratively, the data type of the first training data is multimedia data (i.e. the first training data is first training multimedia data), and any multimedia data can be obtained as the first training multimedia data, wherein the multimedia data includes at least one of text data, image data, audio data and video data.

Illustratively, obtaining a first data enhancement parameter comprises: acquiring second verification data corresponding to the last training; determining a loss value of the first network model corresponding to the second verification data based on the second verification data and the first network model; and determining a first data enhancement parameter based on the loss value of the first network model corresponding to the second verification data.

For example, if the data type of the first training data is multimedia data, the data type of the second validation data is also multimedia data (i.e., the second validation data is second validation multimedia data). In addition, the second verification data is acquired in a manner similar to that of the first training data. The second verification data and the first training data may be the same data or different data, and are not limited herein.

In this embodiment, the second verification data may be input to the first network model, and the first network model outputs a prediction result of the second verification data, where the prediction result of the second verification data is a data processing result of the second verification data obtained through prediction. In addition, a labeling result of the second verification data may be obtained, where the labeling result of the second verification data is a data processing result of the second verification data obtained by labeling. And then, determining a loss value of the first network model corresponding to the second verification data based on the prediction result of the second verification data and the labeling result of the second verification data.

The first network model is a data processing result obtained by performing data processing on the input second verification data and outputting a prediction result of the second verification data, where the prediction result is a result of the data processing obtained by the prediction. The data processing is not limited in the embodiments of the present application, and may be any one of classification processing, translation processing, recognition processing, detection processing, and the like.

For example, if the second verification data is the second verification multimedia data, the first network model may output a prediction result of the second verification multimedia data after classifying the second verification multimedia data, where the prediction result is a classification result of the second verification multimedia data obtained through prediction.

Optionally, a third data enhancement parameter corresponding to the last training is obtained. And performing data enhancement processing on the second verification data based on the third data enhancement parameter to obtain the second verification data after the data enhancement processing. And determining a loss value of the first network model corresponding to the second verification data based on the second verification data and the first network model after the data enhancement processing. The obtaining manner of the third data enhancement parameter is similar to the obtaining manner of the first data enhancement parameter, and may be as described in the above "obtaining the first data enhancement parameter", and is not described herein again.

The second verification data after the data enhancement processing may be input to the first network model, and the first network model outputs a prediction result of the second verification data after the data enhancement processing, where the prediction result of the second verification data after the data enhancement processing is a data processing result of the second verification data after the data enhancement processing obtained through prediction. And then, determining a loss value of the first network model corresponding to the second verification data based on the prediction result of the second verification data after the data enhancement processing and the labeling result of the second verification data.

Next, a first data enhancement parameter is determined based on the loss value of the first network model corresponding to the second validation data. A random gradient descent algorithm may be employed to determine the first data enhancement parameter based on a loss value of the first network model corresponding to the second validation data. The manner of determining the first data enhancement parameter is similar to the manner of determining the second data enhancement parameter for the next training, and may be described in the following description about the second data enhancement parameter, which is not repeated herein.

It should be noted that, taking the second verification data as the second verification multimedia data as an example, the embodiment of the present application may obtain the third data enhancement parameter corresponding to the last training. And performing data enhancement processing on the second verification multimedia data based on the third data enhancement parameter to obtain the second verification multimedia data after the data enhancement processing. And inputting the second verification multimedia data subjected to the data enhancement processing into the first network model, and outputting a prediction result of the second verification multimedia data subjected to the data enhancement processing by the first network model, wherein the prediction result is a data processing result obtained through prediction. And meanwhile, acquiring a labeling result of the second verification multimedia data, wherein the labeling result is a data processing result obtained through labeling. And then, determining a loss value of the first network model corresponding to the second verification multimedia data based on the prediction result of the second verification multimedia data and the labeling result of the second verification multimedia data after the data enhancement processing, and determining a first data enhancement parameter based on the loss value. The data processing result is a processing result of any one of data processing such as classification processing, translation processing, recognition processing, and detection processing.

In one possible implementation, obtaining a first network model includes: acquiring a candidate network model and a first network structure parameter corresponding to the current training, wherein the first network structure parameter is used for representing the structure information of the first network model; and sampling the candidate network model based on the first network structure parameter to obtain a first network model.

The embodiment of the application does not limit the structure and the acquisition mode of the candidate network model. For example, a pre-built candidate network model may be directly obtained, where the candidate network model is formed by stacking at least one network layer, and any network layer may be referred to as a Cell (Cell). Each network layer comprises a plurality of (e.g. 4) nodes, any one of which characterizes a layer of feature vectors. A Super-Edge (Super-Edge) exists between any two nodes in a network layer, and the Super-Edge representation carries out operation processing on the feature vector. Any one of the super edges corresponds to at least one operation process, for example, any one of the super edges corresponds to 7 operation processes, which are respectively 3 × 3 separable convolution, 5 × 5 separable convolution, 3 × 3 separable hole convolution, 5 × 5 separable hole convolution, 3 × 3 maximum pooling, 3 × 3 average pooling, and direct Connection operation (Skip Connection). The super edge in the embodiment of the present application is a candidate edge mentioned below, and the operation processing in the embodiment of the present application is candidate operation processing mentioned below.

And if the current training is the first training, the first network structure parameter corresponding to the current training is an initial network structure parameter, and the initial network structure parameter is a manually set or randomly set network structure parameter. If the current training is any training except the first training, the first network structure parameter corresponding to the current training is determined based on the second verification data, and the following description focuses on the manner of determining the first network structure parameter based on the second verification data.

In one possible implementation, the obtaining the first network structure parameter includes: acquiring second verification data corresponding to the last training; determining a loss value of the first network model corresponding to the second verification data based on the second verification data and the first network model; and determining the first network structure parameter based on the loss value of the first network model corresponding to the second verification data.

The manner of obtaining the second verification data and the manner of determining the loss value of the first network model corresponding to the second verification data are described above, and are not described herein again. A random gradient descent algorithm may be employed to determine the first network structure parameter based on the loss value of the first network model corresponding to the second validation data. The method for determining the first network structure parameter is similar to the method for determining the second network structure parameter of the next training, and may be described in the following description about the second network structure parameter, which is not described herein again.

Since the first network structure parameter is used for representing the structure information of the first network model, the candidate network model can be sampled based on the first network structure parameter to obtain the first network model. The candidate network model comprises a plurality of candidate edges, and each candidate edge corresponds to at least one candidate operation, so that the candidate network model is sampled, namely, the candidate edges in the candidate network model and the candidate operation corresponding to the candidate edges are sampled.

Optionally, if the candidate network model is a, the candidate edge in the candidate network model is B, and the candidate operation process corresponding to the candidate edge in the candidate network model is a, the first network model may be represented as:

b to p (. beta.). Wherein a to p (alpha, beta) represent a first network model obtained by sampling the candidate network model based on alpha and beta,

for the equivalent notation, a to p (α) indicate that the candidate arithmetic processing corresponding to the candidate edge in the candidate network model is sampled based on α, and B to p (β) indicate that the candidate edge in the candidate network model is sampled based on β. Wherein α represents the importance parameter of each candidate operation corresponding to the target edge, β represents the importance parameter of each candidate edge, and α and β are described in detail below, which is not repeated herein.

In a possible implementation manner, the candidate network model includes a plurality of candidate edges, the candidate edge characterization performs operation processing on the feature vector, and the first network structure parameter includes an importance parameter of each candidate edge; based on the first network structure parameter, sampling the candidate network model to obtain a first network model, including: sampling a plurality of candidate edges based on the importance parameters of the candidate edges to obtain a plurality of target edges; a first network model is determined based on the plurality of target edges.

As mentioned above, a candidate edge exists between any two nodes in any network layer in the candidate network model, and the candidate edge characterizes the feature vector. The first network structure parameter includes an importance parameter of each candidate edge, and the plurality of candidate edges may be sampled based on the importance parameter of each candidate edge to obtain a plurality of target edges.

Optionally, a gunn bell (gunbel) re-parameterization manner is adopted, and a plurality of candidate edges are sampled based on the importance parameters of the candidate edges, which is described in detail below.

In the embodiment of the present application, for a node j inside a cell, a candidate edge of an input node j is denoted as e _i,j Where i ∈ {1,2, …, j-1 }. Candidate edge e _i,j Is recorded as beta _i,j Then, the candidate edge e is determined according to the following formula (1) _i,j Whether or not to be sampled.

Wherein, beta _i,j Characterizing candidate edges e _i,j Is an importance parameter of _k,j Characterizing candidate edges e _k,j Exp is the sign of the Exponential function (Exp), Σ is the summation sign, and τ is the hyperparameter.

Representing normalized candidate edges e _i,j Is determined by the importance parameter of the system,

representing normalized candidate edges e _k,j Log is logarithmic sign. g _i，j Indicating the opposite edge e of the needle _i，j Random variable, g, obtained by random sampling according to Gumbel distribution _k,j Indicating the opposite edge e of the needle _k,j Randomly sampled random variables are distributed according to Gumbel,

representing candidate edges e _i,j Sampling probability of (B) _i,j Representing candidate edges e _i,j Whether sampled or not, argtopp 2 is used to take the first two variable values that maximize the objective function, which is

The variable value is i.

As can be seen from the formula (1), B _i,j Is 0 or 1. Wherein, B _i,j Is 0 represents the candidate edge e _i,j Is not sampled, B _i,j A value of 1 indicates a candidate edge e _i,j And sampling, wherein the candidate edge to be sampled is the target edge. The number of the target edges is multiple, and the first network model can be determined based on the multiple target edges.

In a possible implementation manner, any one of the candidate edges corresponds to at least one of the candidate arithmetic processings, and the first network structure parameter further includes an importance parameter of each of the candidate arithmetic processings corresponding to any one of the target edges; determining a first network model based on a plurality of target edges, comprising: for any one target edge, sampling various candidate operation processes corresponding to any one target edge based on the importance parameters of the various candidate operation processes corresponding to any one target edge to obtain a target operation process corresponding to any one target edge; and determining a first network model based on target operation processing corresponding to a plurality of target edges.

As already mentioned above, any one of the candidate edges in the candidate network model corresponds to at least one of the candidate arithmetic processes, and the target edge is a sampled candidate edge. The first network configuration parameter includes an importance parameter of each candidate operation process corresponding to any one of the target edges, and the target operation process corresponding to any one of the target edges may be obtained by sampling each candidate operation process corresponding to any one of the target edges based on the importance parameter of each candidate operation process corresponding to any one of the target edges.

Optionally, a Gumbel-fold parameterization manner is adopted, and based on the importance parameters of various candidate operation processes corresponding to any one target edge, the sampling process is performed on the various candidate operation processes corresponding to any one target edge, which is described in detail below.

In the embodiment of the present application, if the edge e is candidate _i,j Is sampled, then edge e is candidate _i,j Is the target edge. Then the target edge e _i,j The importance parameter of the corresponding candidate operation process o is recorded as

The target edge e is determined according to the following formula (2) _i,j Whether the corresponding candidate arithmetic processing o is sampled.

Wherein,

characterizing a target edge e _i,j The importance parameter of the corresponding candidate arithmetic processing o,

characterizing a target edge e _i,j Importance parameter of corresponding candidate operation processing O', O characterizing target edge e _i,j For each of the candidate arithmetic processes, exp is the sign of the exponential function, and Σ is the sign of the summation.

Representing normalized object edge e _i,j The importance parameter of the corresponding candidate arithmetic processing o,

representing normalized object edge e _i,j The corresponding candidate operation processes the importance parameter of o', log being the logarithmic sign.

Representing target edges e _i,j The corresponding candidate operation processing o distributes the random variables obtained by random sampling according to Gumbel,

representing target edges e _i,j The corresponding candidate operation processing o' distributes random variables obtained by random sampling according to Gumbel,

representing the target edge e _i,j Corresponding toCandidate operation processing o sampling probability, A _i,j Representing the target edge e _i,j And one _ hot is a symbol of the one-hot coding corresponding to the one-hot coding of the target operation processing. argmax is used to take the value of the variable at which the target function is maximized, where the target function is

The value of the variable is o, that is,

representing the target edge e _i,j And (5) corresponding target operation processing.

In this embodiment, a formula (2) may be adopted to determine the target operation processing corresponding to each target edge, and then, based on the target operation processing corresponding to each target edge, the first network model may be determined.

Step 202, performing data enhancement processing on the first training data based on the first data enhancement parameter to obtain the first training data after the data enhancement processing.

In the embodiment of the application, because the first data enhancement parameter is used for representing the information of the data enhancement processing, the data enhancement processing can be performed on the first training data based on the first data enhancement parameter to obtain the first training data after the data enhancement processing.

For example, if the first training data is first training multimedia data, the first training multimedia data is subjected to data enhancement processing based on the first data enhancement parameter, so as to obtain the first training multimedia data subjected to data enhancement processing.

In one possible implementation, the first data enhancement parameter includes a probability of being selected of at least one candidate data enhancement process; based on the first data enhancement parameter, performing data enhancement processing on the first training data to obtain the first training data after the data enhancement processing, including: determining a target data enhancement treatment from the at least one candidate data enhancement treatment based on the probability of being selected of the at least one candidate data enhancement treatment; and performing target data enhancement processing on the first training data to obtain the first training data after data enhancement processing.

In the implementation of the present application, the probability of being selected, which is greater than the first reference probability, may be determined from the probabilities of being selected of the at least one candidate data enhancement process included in the first data enhancement parameter. And taking the candidate data enhancement processing corresponding to the selected probability larger than the first reference probability as target data enhancement processing. The first reference probability may be a set probability value, for example, the first reference probability is 0.75. The first reference probability may also be determined based on the selection probability of the at least one candidate data enhancement processing, for example, the selection probabilities of the at least one candidate data enhancement processing are sorted in descending order, and the nth selection probability after sorting is used as the first reference probability, where N is a positive integer.

For example, the first data enhancement parameter includes three candidate data enhancement processes having respective probabilities of being selected of 0.87, 0.03, and 0.1, and a first reference probability of 0.75. At this time, the candidate data enhancement processing corresponding to 0.87 is the target data enhancement processing.

And after determining the target data enhancement processing, performing the target data enhancement processing on the first training data to obtain the first training data after the data enhancement processing.

It should be noted that, taking the first training data as the first training multimedia data as an example, the probability of being selected, which is greater than the first reference probability, may be determined from the probability of being selected of the at least one candidate data enhancement process included in the first data enhancement parameter. And taking the candidate data enhancement processing corresponding to the selected probability which is greater than the first reference probability as target data enhancement processing. And performing target data enhancement processing on the first training multimedia data to obtain the first training multimedia data after data enhancement processing.

In a possible implementation manner, the first data enhancement parameter further includes an executed probability of each sub-data enhancement processing in the target data enhancement processing; performing target data enhancement processing on the first training data to obtain the first training data after data enhancement processing, including: determining target sub-data enhancement processing from each sub-data enhancement processing based on the executed probability of each sub-data enhancement processing; and performing target subdata enhancement processing on the first training data to obtain the first training data after data enhancement processing.

Any candidate data enhancement processing comprises at least one sub data enhancement processing, the number of the sub data enhancement processing is at least one, and any sub data enhancement processing can be any one of translation processing, rotation processing, histogram equalization processing and the like. For example, if any of the candidate data enhancement processes includes two types of sub-data enhancement processes, and 15 types of sub-data enhancement processes are included, 225 (15) is included in total ² ) Candidate data enhancement processing is performed.

In the implementation of the present application, the target data enhancement processing is determined from at least one candidate data enhancement processing, and therefore, the target data enhancement processing also includes at least one seed data enhancement processing. The executed probability that is greater than the second reference probability may be determined from the executed probabilities of the respective sub data enhancement processes in the target data enhancement process included in the first data enhancement parameter. And taking the sub-data enhancement processing in the target data enhancement processing corresponding to the executed probability which is greater than the second reference probability as the target sub-data enhancement processing. The second reference probability may be a set probability value, for example, the second reference probability is 0.6. The second reference probability may also be determined based on the executed probabilities of the respective sub-data enhancement processes in the target data enhancement process, for example, the executed probabilities of the respective sub-data enhancement processes in the target data enhancement process are sorted in descending order, and the mth executed probability after sorting is used as the second reference probability, where M is a positive integer.

For example, the target data enhancement processing includes two kinds of sub data enhancement processing, the first data enhancement parameter includes executed probabilities of the two kinds of sub data enhancement processing, the executed probabilities of the two kinds of sub data enhancement processing are 0.67, 0.33, respectively, and the second reference probability is 0.6. At this time, the sub-data enhancement processing corresponding to 0.67 is the target sub-data enhancement processing.

And after determining the target subdata enhancement processing, performing target subdata enhancement processing on the first training data to obtain the first training data after data enhancement processing. Optionally, the first data enhancement parameter further includes an executed amplitude value of each sub data enhancement processing in the target data enhancement processing (i.e., an amplitude at which the sub data enhancement processing is executed). In the embodiment of the application, target sub-data enhancement processing is performed on first training data based on the executed amplitude of each sub-data enhancement processing in the target data enhancement processing, so that the first training data after the data enhancement processing is obtained. For example, based on the executed amplitude of each sub-data enhancement processing in the target data enhancement processing, the target sub-data enhancement processing is performed on the first training multimedia data to obtain the first training multimedia data after the data enhancement processing.

Step 203, based on the first training data after the data enhancement, adjusting the model parameters of the first network model to obtain a second network model.

In this embodiment of the application, a loss value of the first network model corresponding to the first training data may be determined based on the first training data and the first network model after the data enhancement processing. And adjusting the first network model based on the loss value of the first network model corresponding to the first training data to obtain a second network model.

In a possible implementation manner, adjusting model parameters of the first network model based on the first training data after the data enhancement processing to obtain a second network model includes: inputting the first training data subjected to data enhancement processing into a first network model to obtain a prediction result of the first training data subjected to data enhancement processing; acquiring a labeling result of the first training data; and adjusting model parameters of the first network model based on the prediction result of the first training data and the labeling result of the first training data after the data enhancement processing to obtain a second network model.

Inputting the first training data after the data enhancement processing into a first network model, and performing data processing on the first training data after the data enhancement processing by using the first network model to obtain a prediction result of the first training data after the data enhancement processing, wherein the prediction result is a data processing result of the first training data after the data enhancement processing obtained through prediction. In addition, a labeling result of the first training data, which is a data processing result of the first training data obtained by labeling, may also be obtained. Next, determining a loss value of the first network model corresponding to the first training data based on the prediction result of the first training data after the data enhancement processing and the labeling result of the first training data, wherein the labeling result is the data processing result of the first training data obtained through labeling. And then, adjusting the first network model based on the loss value of the first network model corresponding to the first training data to obtain a second network model. The first training data may be first training multimedia data, and the data processing result is a processing result of any one of classification processing, translation processing, recognition processing, detection processing, and the like.

It should be noted that the adjustment of the first network model is to adjust the model parameters of the first network model. In this embodiment of the application, the model parameter of the first network model may be adjusted based on the loss value of the first network model corresponding to the first training data according to the following formula (3), so as to obtain the model parameter of the second network model.

Where s.t. is restricted To (Subject To) symbols, w ^* (γ, a) model parameters characterizing the second network model, which can be simply denoted as w ^* . Gamma is a first data enhancement parameter and a represents a candidate network model. argmin is used to take the value of the variable at which the minimum value of the objective function is taken, the objective function being E _Γ～p(γ) [L _train (w,α|Γ(D _train ))]The variable value is w. Where E represents the mathematically expected symbol. Γ to p (γ) indicate that the first training data is subjected to data enhancement processing based on the first data enhancement parameter γ. L is _train (w,a|Γ(D _train ) Representing a loss value of the first network model corresponding to the first training data, w being a model of the first network modelParameter, D _train Characterizing first training data, Γ (D) _train ) Representing the first training data after the data enhancement process.

And step 204, in response to the training end condition being met, taking the second network model as a data processing model.

And when the training end condition is met, taking the second network model as a data processing model. The training end condition is not limited in the embodiment of the application, and as an example, the training end condition is satisfied as the number of times of training, and the number of times of training may be set according to manual experience, for example, the number of times of training is 500.

It should be noted that different data processing corresponds to different data processing models. When the data processing is a classification processing, the data processing model may be referred to as a classification model, and the classification model is used for performing a classification processing on the multimedia data. When the data processing is a translation process, the data processing model may be referred to as a translation model, and the translation model is used to perform a translation process on the multimedia data. When the data processing is recognition processing, the data processing model may be referred to as a recognition model, and the recognition model is used to perform recognition processing on the multimedia data. When the data processing is detection processing, the data processing model may be referred to as a detection model, and the detection model is used for performing detection processing on the multimedia data.

In a possible implementation manner, after obtaining the second network model, the method further includes: responding to the condition that the training end condition is not met, and acquiring first verification data of the current training; determining a loss value of a second network model corresponding to the first verification data based on the first verification data and the second network model; and determining a second data enhancement parameter of the next training based on the loss value of the second network model corresponding to the first verification data, wherein the second data enhancement parameter is used for representing the information of data enhancement processing.

And when the training end condition is not met, acquiring first verification data of the current training. The number of the first verification data is plural, and the data type of the first verification data is the same as the data type of the first training data, for example, if the data type of the first training data is multimedia data, the data type of the first verification data is also multimedia data (i.e., the first verification data is first verification multimedia data). In addition, the first verification data is acquired in a manner similar to that of the first training data. The first verification data and the first training data may be the same data or different data, and are not limited herein.

In this embodiment of the application, the first verification data may be input to the second network model, and the second network model outputs a prediction result of the first verification data, where the prediction result of the first verification data is a data processing result of the first verification data obtained through prediction. In addition, a labeling result of the first verification data may be obtained, where the labeling result of the first verification data is a data processing result of the first verification data obtained by labeling. And then, determining a loss value of the second network model corresponding to the first verification data based on the prediction result of the first verification data and the labeling result of the first verification data. The data processing result may be a processing result of any one of classification processing, translation processing, recognition processing, detection processing, and the like.

In one possible implementation manner, determining a loss value of the second network model corresponding to the first verification data based on the first verification data and the second network model includes: performing data enhancement processing on the first verification data based on the first data enhancement parameter to obtain the first verification data after the data enhancement processing; and determining a loss value of the second network model corresponding to the first verification data based on the first verification data and the second network model after the data enhancement processing.

In this embodiment of the application, a manner of performing data enhancement processing on the first verification data based on the first data enhancement parameter is similar to a manner of performing data enhancement processing on the first training data based on the first data enhancement parameter, and reference may be made to the related description of step 202, which is not described herein again.

In this embodiment of the application, the first verification data after the data enhancement processing may be input to the second network model, and the second network model outputs a prediction result of the first verification data after the data enhancement processing, where the prediction result of the first verification data after the data enhancement processing is a data processing result of the first verification data after the data enhancement processing obtained through prediction. And then, determining a loss value of the second network model corresponding to the first verification data based on the prediction result of the first verification data after the data enhancement processing and the labeling result of the first verification data.

Next, determining a second data enhancement parameter for the next training based on the loss value of the second network model corresponding to the first verification data. The random gradient descent algorithm may be adopted to determine the gradient of the loss function to the first data enhancement parameter based on the loss value of the second network model corresponding to the first verification data according to the following formula (4), where the loss function refers to the loss function of the second network model and is used for determining the loss value of the second network model. And then, adjusting the first data enhancement parameter based on the gradient of the first data enhancement parameter of the loss function to obtain a second data enhancement parameter of the next training.

Wherein,

is the gradient of the loss function to the first data enhancement parameter, gamma is the first data enhancement parameter,

the sign of the gradient descent. E is a mathematically expected symbol, a to p (α, β) represent a first network model obtained by sampling candidate network models based on α and β (the model structure of the first network model is the same as that of the second network model because the model parameters of the first network model are adjusted to obtain the second network model), a represents a candidate network model, α represents an importance parameter of each candidate arithmetic processing corresponding to a target edge, and β represents an importance parameter of each candidate edge.

A pair of loss values w representing a second network model corresponding to the first verification data ^* The gradient of (a) of (b) is,

a loss value, w, representing the second network model corresponding to the first verification data ^* Are the model parameters of the second network model,

gradient of model parameters representing the second network model to the first data enhancement parameters, D _val Representing the first authentication data.

The model parameters of the second network model may be expressed as formula (5) shown below.

Wherein w ^* (γ) is a model parameter of the second network model, which may be abbreviated as w ^* And gamma is a first data enhancement parameter. argmin is used to take the value of the variable at which the minimum value of the objective function is taken, the objective function being E _Γ～p(γ) [L _train (w|Γ(D _train ))]The variable value is w. w is the model parameter of the first network model and E is the mathematically expected symbol. Γ to p (γ) indicate that the first training data is subjected to data enhancement processing based on the first data enhancement parameter γ. L is a radical of an alcohol _train (w|Γ(D _train ) A loss value, Γ (D), representing a first network model to which the first training data corresponds _train ) Representing the first training data after the data enhancement process. Eta is the symbol of the learning rate,

the sign of the gradient descent.

For symbols of definition, i.e. definition

Is the model parameter w' of the second network model, that is, the model parameter of the second network model can be denoted as w ^* And may also be denoted as w'.

The model parameters of the second network model satisfy the following formula (6), and the descriptions of the symbols in the formula (6) can be found in the formulas (4) and (5), which are not described herein again.

According to the strategy gradient, the model parameter w of the second network model ^* The gradient of the first data enhancement parameter γ satisfies the following formula (7), wherein the description of each symbol in the formula (7) can be found in the formulas (4) and (5), which is not described herein again.

In equation (7), log represents a logarithmic function, p (Γ) represents the probability that the candidate data enhancement process Γ is sampled,

representing the gradient of the function log p (Γ) over the first data enhancement parameter γ. The following formula (8) can be obtained from the formulas (4) to (7), wherein the descriptions of the symbols in the formula (8) can be seen in the formulas (4) and (5), which are not described again.

In the formula (8), D _val In order to be the first authentication data,

the gradient of the parameter is enhanced for the first data for the loss function. Thereafter, the ladder of parameters is enhanced for the first data based on the loss functionAnd adjusting the first data enhancement parameter to obtain a second data enhancement parameter of the next training.

According to the embodiment of the application, when the current training does not meet the training end condition, based on the first verification data of the current training, the loss value of the second network model corresponding to the first verification data is determined, and based on the loss value, the second data enhancement parameter of the next training is determined. Since the second data enhancement parameter is information for characterizing data enhancement processing, the embodiment of the present application can determine data enhancement processing corresponding to next training. Because the data enhancement processing corresponding to the next training is determined according to the loss value corresponding to the last training, the data enhancement processing mode has diversity and higher accuracy, and the robustness and the accuracy of the model can be improved.

In this embodiment of the present application, a second network structure parameter for next training may also be determined based on a loss value of a second network model corresponding to the first verification data, and the second network structure parameter is used to represent structure information of the model for next training, so that the embodiment of the present application may determine the model corresponding to the next training. Because the model corresponding to the next training is determined according to the loss value corresponding to the last training, the model structure is continuously adjusted and optimized in the training process, and the accuracy of the model can be improved.

A random gradient descent algorithm may be used to determine the gradient of the loss function to the first network structure parameter based on the loss value of the second network model corresponding to the first verification data according to equation (9) shown below. And then, adjusting the first network structure parameter based on the gradient of the first network structure parameter of the loss function to obtain a second network structure parameter of the next training.

In this embodiment, the first network structure parameter includes α and β, where α is the importance of various candidate operation processes corresponding to the target edgeThe parameter β represents an importance parameter of each candidate edge. In the formula (9), the reaction mixture,

in order to be the sign of the decline in the gradient,

e represents the mathematically expected sign for the gradient of the loss function versus a. A to p (α) represent sampling of candidate arithmetic processing corresponding to candidate edges in the candidate network model based on α, that is, target arithmetic processing corresponding to target edges obtained based on α sampling, and can be abbreviated as a.

And representing the gradient of the loss value of the second network model corresponding to the first verification data to the A. L is _val (w ^* A) represents a loss value of the second network model corresponding to the first verification data, w ^* Is the model parameter of the second network model, and a is the candidate network model.

Represents the gradient of a versus a.

For the gradient of the loss function to β, B to p (β) represent that the candidate edge in the candidate network model is sampled based on β, i.e., the target edge obtained based on β sampling, which can be abbreviated as B.

A gradient of the loss value pair B of the second network model corresponding to the first verification data,

representing the gradient of B versus β.

In the embodiment of the application, the gradient of the loss function to alpha is calculated

After thatBased on

And adjusting alpha to obtain the adjusted alpha. In calculating the gradient of the loss function to beta

Then based on

And adjusting the beta to obtain the adjusted beta. The adjusted alpha and the adjusted beta are the second network structure parameters of the next training.

In the embodiment of the present application, the first network structure parameter and the first data enhancement parameter should satisfy the following formula (10).

Wherein argmin is used for obtaining a variable value when the target function is minimized, the target function is L (α, β, γ), the variable value is β 0, β 1, γ, wherein the first network structure parameter includes β 2, β, α represents an importance parameter of each candidate operation corresponding to the target edge, β represents an importance parameter of each candidate edge, and γ is a first data enhancement parameter. E is the mathematically expected symbol. a to p (α, β) represent a first network model obtained by sampling the candidate network model based on α and β, and α represents the candidate network model. L is _val (w ^* (γ,a),a|D _val ) A loss value of the second network model corresponding to the first validation data is represented. w is a ^* (γ, a) represents a model parameter of the second network model, which may be abbreviated as w ^* 。D _val Representing the first authentication data.

According to the method, during each training, data enhancement processing is performed on first training data corresponding to the current training based on first data enhancement parameters corresponding to the current training, and then model parameters of a first network model corresponding to the current training are adjusted based on the first training data after the data enhancement processing, so that the prediction capability of the model is improved, the robustness and the accuracy of the model are improved, and the accuracy of a data processing result is improved when the data processing is performed based on the data processing model.

Take the first training data as the first training multimedia data and the data processing as the classification processing as an example. The robustness of the classification model obtained by the related art training is poor, so that the accuracy of the classification result is poor when the multimedia data is classified based on the classification model. In the embodiment of the present application, according to the manner from step 201 to step 204, data enhancement processing is performed on the first training multimedia data corresponding to the current training based on the first data enhancement parameter corresponding to the current training, and then the model parameter of the first network model corresponding to the current training is adjusted based on the first training multimedia data after the data enhancement processing, so as to obtain the classification model. The training mode of the classification model can improve the prediction capability of the classification model, thereby improving the robustness and the accuracy of the classification model, and further improving the accuracy of a classification processing result when the classification model is used for classifying multimedia data.

Based on the same principle, any one of the translation model, the recognition model, the detection model and the like can be obtained by training in the way from step 201 to step 204, so that the prediction capability of any one of the models is improved, and the robustness and the accuracy of any one of the models are improved. When any model is a translation model, the accuracy of a translation processing result can be improved when the translation model is used for carrying out translation processing on multimedia data; when any model is the recognition model, the accuracy of the recognition processing result can be improved when the recognition model is used for recognizing the multimedia data; when any model is a detection model, the accuracy of the detection processing result can be improved when the detection model is used for detecting and processing multimedia data.

Based on the foregoing implementation environment, the present embodiment provides a data processing method, which may be executed by the electronic device 11 in fig. 1, taking a flowchart of the data processing method provided in the present embodiment shown in fig. 3 as an example. As shown in fig. 3, the method includes steps 301 to 302.

Step 301, target data is acquired.

The number of the target data is at least one, and the data type and the acquisition mode of the target data are not limited in the embodiment of the application. Illustratively, the data type of the target data is multimedia data (i.e. the target data is target multimedia data), and any multimedia data can be obtained as the target multimedia data, optionally, the multimedia data includes at least one of image data, text data, audio data and video data.

Step 302, inputting the target data into the data processing model to obtain the data processing result of the target data.

The data processing model is obtained by training according to the training method of the data processing model mentioned in the above embodiments.

In the embodiment of the application, the target data is input into the data processing model, and the data processing model performs data processing on the target data to obtain a data processing result of the target data. The data processing result is a processing result of any one of data processing such as classification processing, translation processing, recognition processing, detection processing, and the like.

For example, when the target data is the target multimedia data, the data processing result of the target multimedia data may be a classification result obtained by performing classification processing on the target multimedia data, or may be a detection result obtained by performing detection processing on the target multimedia data.

In the embodiment of the present application, a data processing model is obtained by training in the training manner of the data processing model provided in the above embodiment (for convenience of description, the data processing model is referred to as the data processing model in the embodiment of the present application). In addition, four data processing models are obtained by other four ways of training, and the four data processing models are respectively called a data processing model 1, a data processing model 2, a data processing model 3 and a data processing model 4. The accuracy of the five data processing models was tested using two data sets (denoted as CIFAR-10 and CIFAR-100 data sets, respectively) and the results are shown in Table 1 below.

TABLE 1

Data processing model	CIFAR-10 dataset	CIFAR-100 dataset
			Data processing model 1	97.2％	82.5％
Data processing model 2	97.4％	/
			Data processing model 3	97.1％	80.6％
Data processing model 4	97.3％	79.4％
			Data processing model of the embodiment of the application	97.9％	84.9％

As can be seen from Table 1, the accuracy of the data processing model of the embodiment of the present application on the CIFAR-10 data set and the CIFAR-100 data set is higher than that of the data processing models 1-4. Therefore, the data processing model of the embodiment of the application can improve the accuracy of the data processing result,

in each training of the data processing model, the data enhancement processing is performed on the first training data corresponding to the current training based on the first data enhancement parameter corresponding to the current training, and then the model parameter of the first network model corresponding to the current training is adjusted based on the first training data after the data enhancement processing. The training mode improves the prediction capability of the data processing model, thereby improving the robustness and the accuracy of the data processing model, and further improving the accuracy of a data processing result when data processing is carried out based on the data processing model.

It should be noted that information (including but not limited to user equipment information, user personal information, etc.), data (including but not limited to data for analysis, stored data, presented data, etc.), and signals referred to in this application are authorized by the user or sufficiently authorized by various parties, and the collection, use, and processing of the relevant data is required to comply with relevant laws and regulations and standards in relevant countries and regions. For example, the target data, the first training data, the first verification data, the second verification data, etc. referred to in this application are obtained with sufficient authorization.

Fig. 4 is a schematic structural diagram of a training apparatus for a data processing model according to an embodiment of the present application, and as shown in fig. 4, the apparatus includes:

an obtaining module 401, configured to obtain a first data enhancement parameter, a first network model, and first training data corresponding to current training, where the first data enhancement parameter is used to represent information of data enhancement processing;

a data enhancement module 402, configured to perform data enhancement processing on the first training data based on the first data enhancement parameter, to obtain first training data after the data enhancement processing;

an adjusting module 403, configured to adjust a model parameter of the first network model based on the first training data after the data enhancement processing, to obtain a second network model;

a determining module 404, configured to take the second network model as the data processing model in response to the training end condition being satisfied.

In one possible implementation, the first data enhancement parameter includes a probability of being selected of at least one candidate data enhancement process;

a data enhancement module 402 for determining a target data enhancement process from the at least one candidate data enhancement process based on the probability of being selected of the at least one candidate data enhancement process; and performing target data enhancement processing on the first training data to obtain the first training data after data enhancement processing.

a data enhancement module 402, configured to determine a target sub-data enhancement processing from each sub-data enhancement processing based on an executed probability of each sub-data enhancement processing; and performing target subdata enhancement processing on the first training data to obtain the first training data after data enhancement processing.

In a possible implementation manner, the adjusting module 403 is configured to input the first training data after the data enhancement processing into the first network model, so as to obtain a prediction result of the first training data after the data enhancement processing; acquiring a labeling result of the first training data; and adjusting model parameters of the first network model based on the prediction result of the first training data and the labeling result of the first training data after the data enhancement processing to obtain a second network model.

In a possible implementation manner, the obtaining module 401 is further configured to obtain first verification data of the current training in response to that the training end condition is not satisfied;

a determining module 404, configured to determine, based on the first verification data and the second network model, a loss value of the second network model corresponding to the first verification data; and determining a second data enhancement parameter of the next training based on the loss value of the second network model corresponding to the first verification data, wherein the second data enhancement parameter is used for representing the information of data enhancement processing.

In a possible implementation manner, the determining module 404 is configured to perform data enhancement processing on the first verification data based on the first data enhancement parameter to obtain first verification data after the data enhancement processing; and determining a loss value of the second network model corresponding to the first verification data based on the first verification data and the second network model after the data enhancement processing.

In a possible implementation manner, the obtaining module 401 is configured to obtain a candidate network model and a first network structure parameter corresponding to current training, where the first network structure parameter is used to represent structure information of the first network model; and sampling the candidate network model based on the first network structure parameter to obtain a first network model.

an obtaining module 401, configured to perform sampling processing on multiple candidate edges based on the importance parameter of each candidate edge to obtain multiple target edges; a first network model is determined based on the plurality of target edges.

an obtaining module 401, configured to perform sampling processing on various candidate operation processes corresponding to any one target edge on the basis of the importance parameter of the various candidate operation processes corresponding to any one target edge, so as to obtain a target operation process corresponding to any one target edge; and determining a first network model based on target operation processing corresponding to a plurality of target edges.

In a possible implementation manner, the obtaining module 401 is configured to obtain second verification data corresponding to a last training; determining a loss value of the first network model corresponding to the second verification data based on the second verification data and the first network model; and determining the first network structure parameter based on the loss value of the first network model corresponding to the second verification data.

In a possible implementation manner, the obtaining module 401 is configured to obtain second verification data corresponding to a last training; determining a loss value of the first network model corresponding to the second verification data based on the second verification data and the first network model; and determining a first data enhancement parameter based on the loss value of the first network model corresponding to the second verification data.

When the device is used for each training, data enhancement processing is carried out on first training data corresponding to the current training based on first data enhancement parameters corresponding to the current training, and then model parameters of a first network model corresponding to the current training are adjusted based on the first training data after the data enhancement processing, so that the prediction capability of the model is improved, and the robustness and the accuracy of the model are improved.

It should be understood that, when the apparatus provided in fig. 4 implements its functions, it is only illustrated by the division of the functional modules, and in practical applications, the above functions may be distributed by different functional modules according to needs, that is, the internal structure of the apparatus is divided into different functional modules to implement all or part of the functions described above. In addition, the apparatus and method embodiments provided by the above embodiments belong to the same concept, and specific implementation processes thereof are detailed in the method embodiments, which are not described herein again, so that when data processing is performed based on a data processing model, accuracy of a data processing result is improved.

Fig. 5 is a schematic structural diagram of a data processing apparatus according to an embodiment of the present application, and as shown in fig. 5, the apparatus includes:

an obtaining module 501, configured to obtain target data;

an obtaining module 502, configured to input the target data into a data processing model, so as to obtain a data processing result of the target data, where the data processing model is obtained by training according to any one of the above-mentioned training methods of the data processing model.

When the data processing model of the device is trained each time, the data enhancement processing is firstly carried out on the first training data corresponding to the current training based on the first data enhancement parameters corresponding to the current training, and then the model parameters of the first network model corresponding to the current training are adjusted based on the first training data after the data enhancement processing. The training mode improves the prediction capability of the data processing model, thereby improving the robustness and the accuracy of the data processing model, and further improving the accuracy of a data processing result when data processing is carried out based on the data processing model.

It should be understood that, when the apparatus provided in fig. 5 implements its functions, it is only illustrated by the division of the functional modules, and in practical applications, the above functions may be distributed by different functional modules according to needs, that is, the internal structure of the apparatus is divided into different functional modules to implement all or part of the functions described above. In addition, the apparatus and method embodiments provided by the above embodiments belong to the same concept, and specific implementation processes thereof are described in the method embodiments for details, which are not described herein again.

Fig. 6 shows a block diagram of a terminal device 600 according to an exemplary embodiment of the present application. The terminal device 600 may be a portable mobile terminal such as: a smartphone, a tablet, a laptop, or a desktop computer. The terminal device 600 may also be referred to by other names such as user equipment, portable terminal, laptop terminal, desktop terminal, etc.

In general, the terminal device 600 includes: a processor 601 and a memory 602.

The processor 601 may include one or more processing cores, such as a 4-core processor, an 8-core processor, and so on. The processor 601 may be implemented in at least one hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), and a PLA (Programmable Logic Array). The processor 601 may also include a main processor and a coprocessor, where the main processor is a processor for Processing data in an awake state, and is also called a Central Processing Unit (CPU); a coprocessor is a low power processor for processing data in a standby state. In some embodiments, the processor 601 may be integrated with a GPU (Graphics Processing Unit), which is responsible for rendering and drawing the content that the display screen needs to display. In some embodiments, processor 601 may also include an AI (Artificial Intelligence) processor for processing computational operations related to machine learning.

The memory 602 may include one or more computer-readable storage media, which may be non-transitory. The memory 602 may also include high-speed random access memory, as well as non-volatile memory, such as one or more magnetic disk storage devices, flash memory storage devices. In some embodiments, a non-transitory computer readable storage medium in the memory 602 is used to store at least one instruction for execution by the processor 601 to implement the data processing model training method or the data processing method provided by the method embodiments in the present application.

In some embodiments, the terminal device 600 may further include: a peripheral interface 603 and at least one peripheral. The processor 601, memory 602, and peripheral interface 603 may be connected by buses or signal lines. Various peripheral devices may be connected to the peripheral interface 603 via a bus, signal line, or circuit board. Specifically, the peripheral device includes: at least one of a radio frequency circuit 604, a display screen 605, a camera assembly 606, an audio circuit 607, and a power supply 608.

The peripheral interface 603 may be used to connect at least one peripheral related to I/O (Input/Output) to the processor 601 and the memory 602. In some embodiments, the processor 601, memory 602, and peripheral interface 603 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 601, the memory 602, and the peripheral interface 603 may be implemented on a separate chip or circuit board, which is not limited in this embodiment.

The Radio Frequency circuit 604 is used for receiving and transmitting RF (Radio Frequency) signals, also called electromagnetic signals. The radio frequency circuitry 604 communicates with communication networks and other communication devices via electromagnetic signals. The rf circuit 604 converts an electrical signal into an electromagnetic signal to transmit, or converts a received electromagnetic signal into an electrical signal. Optionally, the radio frequency circuit 604 comprises: an antenna system, an RF transceiver, one or more amplifiers, a tuner, an oscillator, a digital signal processor, a codec chipset, a subscriber identity module card, and so forth. The radio frequency circuit 604 may communicate with other terminals via at least one wireless communication protocol. The wireless communication protocols include, but are not limited to: the world wide web, metropolitan area networks, intranets, generations of mobile communication networks (2G, 3G, 4G, and 5G), Wireless local area networks, and/or WiFi (Wireless Fidelity) networks. In some embodiments, the rf circuit 604 may further include NFC (Near Field Communication) related circuits, which are not limited in this application.

The display 605 is used to display a UI (User Interface). The UI may include graphics, text, icons, video, and any combination thereof. When the display screen 605 is a touch display screen, the display screen 605 also has the ability to capture touch signals on or over the surface of the display screen 605. The touch signal may be input to the processor 601 as a control signal for processing. At this point, the display 605 may also be used to provide virtual buttons and/or a virtual keyboard, also referred to as soft buttons and/or a soft keyboard. In some embodiments, the display 605 may be one, and is disposed on the front panel of the terminal device 600; in other embodiments, the display 605 may be at least two, which are respectively disposed on different surfaces of the terminal device 600 or in a folding design; in other embodiments, the display 605 may be a flexible display, disposed on a curved surface or a folded surface of the terminal device 600. Even more, the display 605 may be arranged in a non-rectangular irregular pattern, i.e., a shaped screen. The Display 605 may be made of LCD (Liquid Crystal Display), OLED (Organic Light-Emitting Diode), and the like.

The camera assembly 606 is used to capture images or video. Optionally, camera assembly 606 includes a front camera and a rear camera. Generally, a front camera is disposed at a front panel of the terminal, and a rear camera is disposed at a rear surface of the terminal. In some embodiments, the number of the rear cameras is at least two, and each rear camera is any one of a main camera, a depth-of-field camera, a wide-angle camera and a telephoto camera, so that the main camera and the depth-of-field camera are fused to realize a background blurring function, and the main camera and the wide-angle camera are fused to realize panoramic shooting and VR (Virtual Reality) shooting functions or other fusion shooting functions. In some embodiments, camera assembly 606 may also include a flash. The flash lamp can be a monochrome temperature flash lamp or a bicolor temperature flash lamp. The double-color-temperature flash lamp is a combination of a warm-light flash lamp and a cold-light flash lamp, and can be used for light compensation at different color temperatures.

Audio circuitry 607 may include a microphone and a speaker. The microphone is used for collecting sound waves of a user and the environment, converting the sound waves into electric signals, and inputting the electric signals to the processor 601 for processing or inputting the electric signals to the radio frequency circuit 604 to realize voice communication. For the purpose of stereo sound collection or noise reduction, a plurality of microphones may be provided at different positions of the terminal device 600. The microphone may also be an array microphone or an omni-directional pick-up microphone. The speaker is used to convert electrical signals from the processor 601 or the radio frequency circuit 604 into sound waves. The loudspeaker can be a traditional film loudspeaker or a piezoelectric ceramic loudspeaker. When the speaker is a piezoelectric ceramic speaker, the speaker can be used for purposes such as converting an electric signal into a sound wave audible to a human being, or converting an electric signal into a sound wave inaudible to a human being to measure a distance. In some embodiments, audio circuitry 607 may also include a headphone jack.

The power supply 608 is used to supply power to various components in the terminal device 600. The power supply 608 may be alternating current, direct current, disposable or rechargeable. When the power supply 608 includes a rechargeable battery, the rechargeable battery may be a wired rechargeable battery or a wireless rechargeable battery. The wired rechargeable battery is a battery charged through a wired line, and the wireless rechargeable battery is a battery charged through a wireless coil. The rechargeable battery may also be used to support fast charge technology.

In some embodiments, the terminal device 600 further includes one or more sensors 609. The one or more sensors 609 include, but are not limited to: acceleration sensor 611, gyro sensor 612, pressure sensor 613, optical sensor 614, and proximity sensor 615.

The acceleration sensor 611 can detect the magnitude of acceleration in three coordinate axes of the coordinate system established with the terminal apparatus 600. For example, the acceleration sensor 611 may be used to detect components of the gravitational acceleration in three coordinate axes. The processor 601 may control the display screen 605 to display the user interface in a landscape view or a portrait view according to the gravitational acceleration signal collected by the acceleration sensor 611. The acceleration sensor 611 may also be used for acquisition of motion data of a game or a user.

The gyro sensor 612 may detect a body direction and a rotation angle of the terminal device 600, and the gyro sensor 612 may acquire a 3D motion of the user on the terminal device 600 in cooperation with the acceleration sensor 611. The processor 601 may implement the following functions according to the data collected by the gyro sensor 612: motion sensing (such as changing the UI according to a user's tilting operation), image stabilization at the time of photographing, game control, and inertial navigation.

The pressure sensor 613 may be disposed on a side frame of the terminal device 600 and/or on a lower layer of the display 605. When the pressure sensor 613 is disposed on the side frame of the terminal device 600, the holding signal of the user to the terminal device 600 can be detected, and the processor 601 performs left-right hand recognition or shortcut operation according to the holding signal collected by the pressure sensor 613. When the pressure sensor 613 is disposed at the lower layer of the display screen 605, the processor 601 controls the operability control on the UI interface according to the pressure operation of the user on the display screen 605. The operability control comprises at least one of a button control, a scroll bar control, an icon control and a menu control.

The optical sensor 614 is used to collect the ambient light intensity. In one embodiment, processor 601 may control the display brightness of display screen 605 based on the ambient light intensity collected by optical sensor 614. Specifically, when the ambient light intensity is high, the display brightness of the display screen 605 is increased; when the ambient light intensity is low, the display brightness of the display screen 605 is adjusted down. In another embodiment, processor 601 may also dynamically adjust the shooting parameters of camera assembly 606 based on the ambient light intensity collected by optical sensor 614.

The proximity sensor 615, also called a distance sensor, is generally provided on the front panel of the terminal device 600. The proximity sensor 615 is used to collect a distance between the user and the front surface of the terminal device 600. In one embodiment, when the proximity sensor 615 detects that the distance between the user and the front surface of the terminal device 600 gradually decreases, the processor 601 controls the display 605 to switch from the bright screen state to the dark screen state; when the proximity sensor 615 detects that the distance between the user and the front surface of the terminal device 600 is gradually increased, the processor 601 controls the display 605 to switch from the screen-on state to the screen-on state.

Those skilled in the art will appreciate that the configuration shown in fig. 6 is not limiting of terminal device 600 and may include more or fewer components than shown, or some components may be combined, or a different arrangement of components may be used.

Fig. 7 is a schematic structural diagram of a server 700 according to an embodiment of the present application, where the server 700 may generate a relatively large difference due to a difference in configuration or performance, and may include one or more processors 701 and one or more memories 702, where at least one program code is stored in the one or more memories 702, and is loaded and executed by the one or more processors 701 to implement the data processing model training method or the data processing method provided in the foregoing method embodiments, and the processor 701 is, for example, a CPU. Of course, the server 700 may also have components such as a wired or wireless network interface, a keyboard, and an input/output interface, so as to perform input and output, and the server 700 may also include other components for implementing the functions of the device, which are not described herein again.

In an exemplary embodiment, there is also provided a computer readable storage medium having at least one program code stored therein, the at least one program code being loaded and executed by a processor to cause an electronic device to implement any one of the above-mentioned training methods for a data processing model or data processing methods.

Alternatively, the computer-readable storage medium may be a Read-Only Memory (ROM), a Random Access Memory (RAM), a Compact Disc Read-Only Memory (CD-ROM), a magnetic tape, a floppy disk, an optical data storage device, and the like.

In an exemplary embodiment, a computer program or a computer program product is further provided, in which at least one computer instruction is stored, the at least one computer instruction being loaded and executed by a processor, so as to make a computer implement the training method or the data processing method of any one of the data processing models described above.

It should be understood that reference to "a plurality" herein means two or more. "and/or" describes the association relationship of the associated objects, meaning that there may be three relationships, e.g., a and/or B, which may mean: a exists alone, A and B exist simultaneously, and B exists alone. The character "/" generally indicates that the former and latter associated objects are in an "or" relationship.

The above-mentioned serial numbers of the embodiments of the present application are merely for description and do not represent the merits of the embodiments.

The above description is only exemplary of the present application and should not be taken as limiting the present application, and any modifications, equivalents, improvements and the like that are made within the principles of the present application should be included in the protection scope of the present application.

Claims

1. A method of training a data processing model, the method comprising:

performing data enhancement processing on the first training data based on the first data enhancement parameter to obtain first training data after the data enhancement processing;

2. The method of claim 1, wherein the first data enhancement parameter comprises a probability of being selected for at least one candidate data enhancement process;

3. The method of claim 2, wherein the first data enhancement parameter further comprises a probability of execution of each sub-data enhancement process of the target data enhancement process;

4. The method according to claim 1, wherein the adjusting model parameters of the first network model based on the first training data after the data enhancement processing to obtain a second network model comprises:

acquiring a labeling result of the first training data;

5. The method of claim 1, wherein after obtaining the second network model, further comprising:

and determining a second data enhancement parameter for next training based on the loss value of the second network model corresponding to the first verification data, wherein the second data enhancement parameter is used for representing information of data enhancement processing.

6. The method of claim 5, wherein determining the loss value of the second network model corresponding to the first validation data based on the first validation data and the second network model comprises:

performing data enhancement processing on the first verification data based on the first data enhancement parameter to obtain first verification data after data enhancement processing;

7. The method of any of claims 1 to 6, wherein obtaining the first network model comprises:

8. The method according to claim 7, wherein the candidate network model includes a plurality of candidate edges, the candidate edges characterize performing operation processing on feature vectors, and the first network structure parameter includes an importance parameter of each candidate edge;

a first network model is determined based on the plurality of target edges.

9. The method of claim 8, wherein any candidate edge corresponds to at least one candidate operation, and the first network configuration parameters further comprise importance parameters of various candidate operation corresponding to any target edge;

10. The method of claim 7, wherein obtaining the first network configuration parameter comprises:

acquiring second verification data corresponding to the last training;

11. The method of any of claims 1 to 6, wherein obtaining the first data enhancement parameter comprises:

acquiring second verification data corresponding to the last training;

12. A method of data processing, the method comprising:

acquiring target data;

inputting the target data into a data processing model to obtain a data processing result of the target data, wherein the data processing model is obtained by training according to the method of any one of claims 1 to 11.

13. An apparatus for training a data processing model, the apparatus comprising:

14. A data processing apparatus, characterized in that the apparatus comprises:

the acquisition module is used for acquiring target data;

an obtaining module, configured to input the target data into a data processing model, so as to obtain a data processing result of the target data, where the data processing model is obtained by training according to the method of any one of claims 1 to 11.

15. An electronic device, characterized in that the electronic device comprises a processor and a memory, wherein at least one program code is stored in the memory, and the at least one program code is loaded and executed by the processor to cause the electronic device to implement the training method of the data processing model according to any one of claims 1 to 11 or the data processing method according to claim 12.