Data processing method, device and storage medium
By updating the neural network model parameters and structure, and optimizing the network architecture with jump connection and activation functions, the problem of unstable model performance in neural network architecture search is solved, and the model accuracy and accuracy of classification results are improved.
Patent Information
- Application Number
- CN202010608235.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-06-29
- Publication Date
- 2025-01-03
- Estimated Expiration
- 2040-06-29
AI Technical Summary
During the search process based on neural network architecture, the search results are unstable, and the performance gap between the models searched multiple times is large, and the optimality cannot be achieved, resulting in low accuracy of the target model.
By inputting the marked training data in the training set into the predetermined model to be trained, the model parameters are updated according to the difference between the predicted label and the annotated label, and the target network model is constructed based on the trained network structure, the classification data is classified, and the jump connection parameters are used and activation function are optimized to reduce the difference in the output results of the network structure, and stop training when the training loss value changes less than the threshold.
A target network model with higher accuracy is achieved, and more accurate classification results are obtained, solving the problem of unstable model performance.
Smart Images
Figure CN111753917B_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of neural network technology, and in particular to a data processing method, device and storage medium. Background Art
[0002] Neural network is an algorithmic mathematical model that imitates the behavioral characteristics of animal neural networks and performs distributed parallel information processing. It mainly relies on the complexity of the system and adjusts the interconnected relationships between a large number of internal nodes to achieve the purpose of processing information. It has a wide range of applications in data classification, speech analysis, and image recognition. However, in the process of searching for network models based on the neural network architecture, there are cases where the search results are unstable, the performance gap between the models searched multiple times is large, and the performance cannot reach the optimal level. This will result in low accuracy of the target model. Summary of the invention
[0003] The present disclosure provides a data processing method, device and storage medium.
[0004] According to a first aspect of an embodiment of the present disclosure, there is provided a data processing method, including:
[0005] Input each labeled training data in the training set into a predetermined model to be trained to obtain a first prediction label;
[0006] Obtaining a training loss value according to a difference between the first predicted label and the labeled label of the labeled training data;
[0007] Based on the training loss value, updating the model parameters of the predetermined model;
[0008] The target network model constructed based on the network structure included in the trained predetermined model is used to classify the data to be classified to obtain a classification result.
[0009] Optionally, the predetermined model includes: a first type of network structure and a second type of network structure;
[0010] The network function of the first type of network structure includes a jump connection parameter, and the jump connection parameter is used to reduce the difference between the output result of the first type of network structure and the output result of the second type of network structure by mixing random noise.
[0011] Optionally, the amplitude of the random noise mixed in as indicated by the jump connection parameter is smaller than the output result of the first type of network structure.
[0012] Optionally, the model parameters include: network parameters and structural parameters; and updating the model parameters of the predetermined model based on the training loss value includes:
[0013] Based on the training loss value, updating the network parameters of the predetermined model;
[0014] Input each of the labeled verification data in the verification set into the predetermined model after the network parameters are updated to obtain a second prediction label;
[0015] Obtaining a verification loss value based on a difference between the second predicted label and the labeled verification data;
[0016] Based on the verification loss value, the structural parameters of the predetermined model after the network parameters are updated are updated.
[0017] Optionally, the target network model constructed based on the network structure included in the trained predetermined model classifies the data to be classified to obtain a classification result, including:
[0018] Normalizing the model parameters of each network structure in the trained predetermined model based on a predetermined type of activation function to obtain normalized model parameters;
[0019] From the trained predetermined model, determining a network structure whose normalized model parameter value is greater than a set parameter threshold;
[0020] Based on the network structure whose normalized model parameter value is greater than the set parameter threshold, constructing the target network model;
[0021] The data to be classified is classified based on the target network model to obtain the classification result.
[0022] Optionally, the target network model constructed based on the network structure included in the trained predetermined model classifies the data to be classified to obtain a classification result, including:
[0023] Input the data to be classified into the target network model, and output M category probabilities of the data to be classified in M categories; wherein the category labels corresponding to the first preset number of category probabilities among the M category probabilities are the classification results of the data to be classified;
[0024] The type of the data to be classified includes at least one of the following: an image data type, a text data type, and an audio data type.
[0025] Optionally, the method further includes:
[0026] When the numerical change of the training loss value within a set time period is less than a set change threshold, the training of the predetermined model is stopped.
[0027] According to a second aspect of an embodiment of the present disclosure, there is provided a data processing device, including:
[0028] A first acquisition module is configured to input each labeled training data in the training set into a predetermined model to be trained to obtain a first prediction label;
[0029] A second acquisition module is configured to obtain a training loss value according to a difference between the first predicted label and the labeled label of the labeled training data;
[0030] An updating module, configured to update a model parameter of the predetermined model based on the training loss value;
[0031] The classification module is configured to construct a target network model based on the network structure included in the trained predetermined model, classify the data to be classified, and obtain a classification result.
[0032] Optionally, the predetermined model includes: a first type of network structure and a second type of network structure;
[0033] The network function of the first type of network structure includes a jump connection parameter, and the jump connection parameter is used to reduce the difference between the output result of the first type of network structure and the output result of the second type of network structure by mixing random noise.
[0034] Optionally, the amplitude of the random noise mixed in as indicated by the jump connection parameter is smaller than the output result of the first type of network structure.
[0035] Optionally, the model parameters include: network parameters and structure parameters; the update module is further configured to:
[0036] Based on the training loss value, updating the network parameters of the predetermined model;
[0037] Input each of the labeled verification data in the verification set into the predetermined model after the network parameters are updated to obtain a second prediction label;
[0038] Obtaining a verification loss value based on a difference between the second predicted label and the labeled verification data;
[0039] Based on the verification loss value, the structural parameters of the predetermined model after the network parameters are updated are updated.
[0040] Optionally, the classification module is further configured to:
[0041] Normalizing the model parameters of each network structure in the trained predetermined model based on a predetermined type of activation function to obtain normalized model parameters;
[0042] From the trained predetermined model, determining a network structure whose normalized model parameter value is greater than a set parameter threshold;
[0043] Based on the network structure whose normalized model parameter value is greater than the set parameter threshold, constructing the target network model;
[0044] The data to be classified is classified based on the target network model to obtain the classification result.
[0045] Optionally, the classification module is further configured to:
[0046] Input the data to be classified into the target network model, and output M category probabilities of the data to be classified in M categories; wherein the category labels corresponding to the first preset number of category probabilities among the M category probabilities are the classification results of the data to be classified;
[0047] The type of the data to be classified includes at least one of the following: an image data type, a text data type, and an audio data type.
[0048] Optionally, the device further comprises:
[0049] The stopping module is configured to stop training the predetermined model when the numerical change of the training loss value within a set time period is less than a set change threshold.
[0050] According to a third aspect of an embodiment of the present disclosure, there is provided a data processing device, including:
[0051] processor;
[0052] a memory configured to store processor-executable instructions;
[0053] Wherein, the processor is configured to: implement the steps in any of the above-mentioned data processing methods when executed.
[0054] According to a fourth aspect of an embodiment of the present disclosure, a non-temporary computer-readable storage medium is provided. When instructions in the storage medium are executed by a processor of a data processing device, the device is enabled to perform any of the above-mentioned data processing methods.
[0055] The technical solution provided by the embodiments of the present disclosure may have the following beneficial effects:
[0056] As can be seen from the above embodiments, the present disclosure inputs each labeled training data in the training set into the predetermined model to be trained, updates the model parameters of the predetermined model according to the difference between the obtained first prediction label and the labeled label of the labeled training data, and classifies the data to be classified according to the target network model constructed according to the network structure included in the trained predetermined model to obtain a classification structure. Since the model parameters can include network parameters and structural parameters, in the process of optimizing the predetermined model, the network parameters and structural parameters can be optimized to obtain a target network model with higher accuracy, so that the classification result obtained based on the model is more accurate.
[0057] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0058] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and, together with the description, serve to explain the principles of the present disclosure.
[0059] Figure 1 It is a flowchart of a data processing method according to an exemplary embodiment.
[0060] Figure 2 The diagram is a schematic diagram of a structure of a hypernetwork according to an exemplary embodiment.
[0061] Figure 3 It is a block diagram of a data processing device according to an exemplary embodiment.
[0062] Figure 4 It is a block diagram of a data processing device according to an exemplary embodiment.
[0063] Figure 5 is a block diagram of another data processing device according to an exemplary embodiment. DETAILED DESCRIPTION
[0064] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.
[0065] The present disclosure provides a data processing method. Figure 1 is a flowchart of a data processing method according to an exemplary embodiment. Figure 1As shown, the method mainly includes the following steps:
[0066] In step 101, each labeled training data in the training set is input into a predetermined model to be trained to obtain a first prediction label;
[0067] In step 102, a training loss value is obtained according to the difference between the first predicted label and the labeled label of the labeled training data;
[0068] In step 103, based on the training loss value, the model parameters of the predetermined model are updated;
[0069] In step 104, the target network model constructed based on the network structure included in the trained predetermined model is used to classify the data to be classified to obtain a classification result.
[0070] The data processing method involved in the embodiments of the present disclosure can be applied to electronic devices, where the electronic devices include mobile terminals and fixed terminals, where the mobile terminals include: mobile phones, tablet computers, laptop computers, etc.; the fixed terminals include: personal computers. In other optional embodiments, the data processing method can also be run on network-side devices, where the network-side devices include: servers, processing centers, etc.
[0071] The target network model obtained in the embodiment of the present disclosure can be used to implement a predetermined function, which includes but is not limited to at least one of the following:
[0072] Object segmentation to segment the object and background in the input image;
[0073] Classification of objects in the input image;
[0074] Object tracking based on incoming images;
[0075] Diagnostic assistance based on medical images;
[0076] Functions such as speech recognition and / or speech correction based on input speech.
[0077] The above is only an example of the predetermined function implemented by the predetermined model, and the specific implementation is not limited to the above example.
[0078] In the process of training the predetermined model, it is first necessary to input each labeled training data in the training set into the predetermined model to be trained, and update the model parameters of the predetermined model.
[0079] In some embodiments, the model parameters may include: structural parameters and / or network parameters. Among them, the structural parameters include parameters used to characterize the structural features of each network structure in the predetermined model, and the structural features include: the connection relationship of each network structure and other features. Taking the predetermined network as a super network as an example, since the super network can contain multiple network structures, the connection relationship between each network structure can be characterized based on the structural parameters. The network parameters include weight parameters for training each network structure in the predetermined model. For example, if the predetermined model contains a network structure that can constitute a convolutional neural network, each network structure can be regarded as a filter (convolution kernel) constituting a convolutional neural network. In this case, the network parameter can be the weight of a filter.
[0080] In the embodiment of the present disclosure, each labeled training data in the training set can be input into a predetermined model to be trained, and the model parameters of the predetermined model can be updated based on the output results. The data type of the labeled training data can include: image data type, text data type or audio data type.
[0081] Here, each labeled training data in the training set can be input into the predetermined model after the model parameters are updated to obtain a first prediction label; and then the preset loss function is used to obtain the training loss value of the predetermined model on the training set based on the difference between the first prediction label and the labeled label of the labeled training data. Here, the preset loss function is an algorithm used to measure the quality of the prediction results output by the predetermined model. In the embodiment of the present disclosure, the model parameters of the predetermined model can be reversely adjusted based on the pre-constructed preset loss function, such as using the gradient descent method, so that the model converges, that is, the final model is better. Among them, the selection of the loss function also affects the quality of the model to a certain extent. The preset loss function can include: binary mutual entropy loss and cross entropy loss function.
[0082] In other optional embodiments, the method further includes: when the numerical change of the training loss value within a set time length is less than a set change threshold, stopping the training of the predetermined model. In the disclosed embodiment, it can be based on a pre-constructed preset loss function. For example, after determining the training loss value, the model parameters of the predetermined model can be reversely adjusted based on the training loss value using a gradient descent method to make the model converge, that is, to make the final model better. For example, when the numerical change of the training loss value within a set time length is less than a set change threshold, it indicates that the model has converged, and at this time, the training of the predetermined model can be stopped.
[0083] In the disclosed embodiment, each labeled training data in the training set is input into the predetermined model to be trained, and the model parameters of the predetermined model are updated according to the difference between the obtained first prediction label and the labeled label of the labeled training data, and the target network model constructed according to the network structure included in the trained predetermined model is used to classify the data to be classified to obtain a classification structure. Since the model parameters may include network parameters and structural parameters, in the process of optimizing the predetermined model, the network parameters and structural parameters can be optimized to obtain a target network model with higher accuracy, so that the classification result obtained based on the model is more accurate.
[0084] In other optional embodiments, the predetermined model includes: a first type of network structure and a second type of network structure;
[0085] The network function of the first type of network structure includes a jump connection parameter, and the jump connection parameter is used to reduce the difference between the output result of the first type of network structure and the output result of the second type of network structure by mixing random noise.
[0086] Here, the first type of network structure can include a skip connection network structure, also called a residual connection network structure. When the neural network is forward propagating, taking the lth layer, l+1 layer, and l+2 layer as an example, a[l] represents the output of the lth layer. In the neural network, it is necessary to pass through the l+1 layer to reach the l+2 layer, and the skip connection network structure can directly transmit a[l] to the l+2 layer, so that a[l+2]=g(z[l+2]+a[l]). In this way, a[l+2] is not only related to z in the l+2 layer, but also to a in the lth layer. This is a skip connection, that is, a skip connection can realize the network connection between the network structures in two non-adjacent network layers. Based on the above properties of the skip connection, the skip connection network structure has great advantages over the second type of network structure, and this advantage will be magnified as the model training process progresses, which will lead to the exclusion of other advantageous first type network structures.
[0087] In the disclosed embodiment, the random noise indicated by the jump connection parameter can interfere with the output result of the first type of network structure to a set degree. For example, a noise function can be added to the output function of the first type of network structure, or multiplied by the noise function. In this way, the difference between the final output result of the first type of network structure and the output result of the second type of network structure can be reduced. In some embodiments, the predetermined model can include a first type of network structure, and multiple second type of network structures.
[0088] In the embodiments of the present disclosure, in order to reduce the possibility that the first type of network structure will dominate due to its own characteristics during model training, a jump connection parameter can be set in the network function of the first type of network structure, and random noise can be mixed in through the jump connection parameter to reduce the output result of the first type of network structure and the output result of the second type of network structure, thereby reducing the unfair advantage of the first type of network structure. Here, the mixed noise can be Gaussian noise, uniformly distributed noise, etc., wherein Gaussian noise refers to a type of noise whose probability density function obeys Gaussian distribution (i.e., normal distribution), and uniformly distributed noise refers to a type of noise whose power spectrum density function obeys uniform distribution, for example, white noise. In some embodiments, the mixed noise can also be other types of noise, and the embodiments of the present disclosure do not specifically limit the type of noise mixed.
[0089] In other optional embodiments, the amplitude of the random noise mixed in indicated by the jump connection parameter is smaller than the output result of the first type of network structure. Here, by making the amplitude of the random noise mixed in smaller than the output result of the first type of network structure, it is possible to reduce the unfair advantage of the first type of network structure without causing excessive impact on the output structure. In some embodiments, the amplitude of the random noise can be a fixed value, or it can decay with the training process of the predetermined model, which is not specifically limited here.
[0090] In other optional embodiments, the model parameters include: network parameters and structural parameters; and updating the model parameters of the predetermined model based on the training loss value includes:
[0091] Based on the training loss value, updating the network parameters of the predetermined model;
[0092] Input each of the labeled verification data in the verification set into the predetermined model after the network parameters are updated to obtain a second prediction label;
[0093] Obtaining a verification loss value based on a difference between the second predicted label and the labeled verification data;
[0094] Based on the verification loss value, the structural parameters of the predetermined model after the network parameters are updated are updated.
[0095] Here, after determining the training loss value, the network parameters of the predetermined model can be adjusted based on the training loss value using the gradient descent method. After updating the network parameters of the predetermined model, each labeled verification data in the verification set can be input into the predetermined model after the network parameters are updated to update the structural parameters.
[0096] For example, each labeled verification data in the verification set is input into the predetermined model after the network parameters are updated to obtain a second prediction label; a verification loss value is obtained based on the difference between the second prediction label and the labeled label of the labeled verification data; and then the structural parameters in the predetermined model after the network parameters are updated are updated based on the verification loss value.
[0097] In the disclosed embodiment, the structural parameters in the predetermined model after the network parameters are updated can be updated according to the verification loss value. In the process of optimizing the predetermined model, in addition to optimizing the network parameters, the structural parameters can also be optimized. Through cross-training, a target network model with higher accuracy can be obtained, and the classification results obtained based on the model are more accurate.
[0098] In other optional embodiments, the target network model constructed based on the network structure included in the trained predetermined model classifies the data to be classified to obtain a classification result, including:
[0099] Normalizing the model parameters of each network structure in the trained predetermined model based on a predetermined type of activation function to obtain normalized model parameters;
[0100] From the trained predetermined model, determining a network structure whose normalized model parameter value is greater than a set parameter threshold;
[0101] Based on the network structure whose normalized model parameter value is greater than the set parameter threshold, constructing the target network model;
[0102] The data to be classified is classified based on the target network model to obtain the classification result.
[0103] In the embodiment of the present disclosure, the activation function includes a softmax function. Here, the model parameters can be normalized based on the activation function to obtain the normalized model parameters. Taking the activation function as a softmax function as an example, the formula for normalizing the model parameters includes:
[0104]
[0105] In formula (1), ∑ i softmax(α i )=1; among them, softmax(α i ) represents the normalized model parameters; α i Represents model parameters, where i represents the index of the current model parameters, j represents the index of different model parameters in the same network layer, M represents the number of network structures in the network layer, and i, j and M are all positive integers.
[0106] In other optional embodiments, the method further includes: performing a performance evaluation on the trained predetermined model based on a test data set to obtain an evaluation result; wherein the type of test data in the test data set includes at least one of the following: an image data type, a business data type, and an audio data type.
[0107] Here, after the trained predetermined model is formed, its performance can be evaluated on the test data set, and the network structure can be gradually optimized until the optimal predetermined model is found, such as a predetermined model that minimizes the verification loss or maximizes the reward. Here, the test data in the test data set can be input into the trained predetermined model, and the evaluation result can be output through the predetermined model, and then the output evaluation result can be compared with the preset standard to obtain a comparison result, and the performance of the predetermined model can be evaluated based on the comparison result, wherein the test result can be the speed or accuracy of the predetermined model processing the test data.
[0108] Since the predetermined model after training has multiple network layers, and the network structure in each network layer whose normalized model parameters are greater than the set parameter threshold can be one or more. When there is one network structure in the network layer whose normalized model parameters are greater than the set parameter threshold, the one network structure can be directly connected to the output end of the previous layer of the network layer and the input end of the next layer. When there are multiple network structures in the network layer whose normalized model parameters are greater than the set parameter threshold, the functions characterizing each network structure can be added to obtain a new network structure for the network layer, and the new network structure can be connected to the output end of the previous layer of the network layer and the input end of the next layer. In the disclosed embodiment, after the network structures of each network layer are connected, the target network model is formed. In the disclosed embodiment, after the target network model is formed, the data to be classified can be classified based on the target network model to obtain a classification result.
[0109] In other optional embodiments, the target network model constructed based on the network structure included in the trained predetermined model classifies the data to be classified to obtain a classification result, including:
[0110] Input the data to be classified into the target network model, and output M category probabilities of the data to be classified in M categories; wherein the category labels corresponding to the first preset number of category probabilities among the M category probabilities are the classification results of the data to be classified;
[0111] The type of the data to be classified includes at least one of the following: an image data type, a text data type, and an audio data type.
[0112] The technical solutions in any of the above embodiments of the present disclosure can be applied to Neural Architecture Search (NAS), where NAS is a technology for automatically designing neural networks. Based on NAS, a high-performance neural network structure can be automatically designed based on a sample set, which can effectively reduce the use and implementation costs of neural networks. In the process of searching based on NAS, a super network (predetermined model) containing multiple network structures can be trained to generate a super network containing a search space of all network structures.
[0113] The technical solution in any of the above embodiments can also be applied to differentiable neural network architecture search (Differentiable Architecture Search, DARTS). In the differentiable neural network architecture search, a method of training a super network containing a search space of all network structures is adopted. Among them, the super network is composed of multiple searchable cells stacked in a certain order, and the cells include normal cells and reduction cells. Each unit is a directed acyclic graph (DAG). Each unit has N nodes, and there are multiple optional network structures between any two nodes. Each network structure contains two types of parameters: structural parameters and network parameters. Among them, each structural parameter is constrained by the softmax function and affects each other. Here, the super network can be trained in a cross-training manner, that is, the structural parameters and network parameters are cross-updated. After the super network training converges, the network structure with the largest output value of the softmax function can be selected as the network structure constituting the target network model.
[0114] Figure 2 is a schematic diagram of a structure of a super network according to an exemplary embodiment. Figure 2 As shown, the super network 200 includes a first network layer 201, a second network layer 202, a third network layer 203, ... wherein any two nodes in the first network layer 201, the second network layer 202 and the third network layer 203, ... respectively include a first type of network structure, and M parallel second type of network structures, namely network structure 1, network structure skip, network structure 2, ... network structure M, and the structural parameters of each network structure are constrained by the softmax function.
[0115] In other optional embodiments, the network structure search can also be performed based on a fair differentiable architecture search (Fair Differentiable Architecture Search, Fair DARTS). In the process of network search based on FairDARTS, a network structure whose softmax function output value is greater than a set parameter threshold can be selected as the network structure constituting the target network model.
[0116] The predetermined model in the disclosed embodiment is not limited to the deep learning model in tasks such as image classification, target detection, semantic segmentation, text-to-speech, natural language translation, speech enhancement, speech recognition, etc. In the process of model training, the discretization deviation can be reduced in the fair differentiable neural network search, thereby obtaining a better neural network structure.
[0117] Figure 3 is a block diagram of a data processing device according to an exemplary embodiment. Figure 3 As shown, the data processing device 300 mainly includes:
[0118] A first acquisition module 301 is configured to input each labeled training data in the training set into a predetermined model to be trained to obtain a first prediction label;
[0119] A second acquisition module 302 is configured to obtain a training loss value according to a difference between the first predicted label and the labeled label of the labeled training data;
[0120] An updating module 303 is configured to update the model parameters of the predetermined model based on the training loss value;
[0121] The classification module 304 is configured to classify the data to be classified based on the target network model constructed based on the network structure included in the trained predetermined model to obtain a classification result.
[0122] In other optional embodiments, the predetermined model includes: a first type of network structure and a second type of network structure;
[0123] The network function of the first type of network structure includes a jump connection parameter, and the jump connection parameter is used to reduce the difference between the output result of the first type of network structure and the output result of the second type of network structure by mixing random noise.
[0124] In other optional embodiments, the amplitude of the random noise mixed in as indicated by the jump connection parameter is smaller than the output result of the first type of network structure.
[0125] In other optional embodiments, the model parameters include: network parameters and structure parameters; the updating module 303 is further configured to:
[0126] Based on the training loss value, updating the network parameters of the predetermined model;
[0127] Input each of the labeled verification data in the verification set into the predetermined model after the network parameters are updated to obtain a second prediction label;
[0128] Obtaining a verification loss value based on a difference between the second predicted label and the labeled verification data;
[0129] Based on the verification loss value, the structural parameters of the predetermined model after the network parameters are updated are updated.
[0130] In other optional embodiments, the classification module 304 is further configured to:
[0131] Normalizing the model parameters of each network structure in the trained predetermined model based on a predetermined type of activation function to obtain normalized model parameters;
[0132] From the trained predetermined model, determining a network structure whose normalized model parameter value is greater than a set parameter threshold;
[0133] Based on the network structure whose normalized model parameter value is greater than the set parameter threshold, constructing the target network model;
[0134] The data to be classified is classified based on the target network model to obtain the classification result.
[0135] In other optional embodiments, the classification module 304 is further configured to:
[0136] Input the data to be classified into the target network model, and output M category probabilities of the data to be classified in M categories; wherein the category labels corresponding to the first preset number of category probabilities among the M category probabilities are the classification results of the data to be classified;
[0137] The type of the data to be classified includes at least one of the following: an image data type, a text data type, and an audio data type.
[0138] In other optional embodiments, the device 300 further includes:
[0139] The stopping module is configured to stop training the predetermined model when the numerical change of the training loss value within a set time period is less than a set change threshold.
[0140] Regarding the device in the above embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.
[0141] Figure 4 1 is a block diagram of a data processing device 1200 according to an exemplary embodiment. For example, the device 1200 may be a mobile phone, a computer, a digital broadcast terminal, a messaging device, a game console, a tablet device, a medical device, a fitness device, a personal digital assistant, etc.
[0142] Reference Figure 4 , the device 1200 may include one or more of the following components: a processing component 1202 , a memory 1204 , a power component 1206 , a multimedia component 1208 , an audio component 1210 , an input / output (I / O) interface 1212 , a sensor component 1214 , and a communication component 1216 .
[0143] The processing component 1202 generally controls the overall operation of the device 1200, such as operations associated with display, phone calls, data communications, camera operations, and recording operations. The processing component 1202 may include one or more processors 1220 to execute instructions to perform all or part of the steps of the above-described method. In addition, the processing component 1202 may include one or more modules to facilitate the interaction between the processing component 1202 and other components. For example, the processing component 1202 may include a multimedia module to facilitate the interaction between the multimedia component 1208 and the processing component 1202.
[0144] The memory 1204 is configured to store various types of data to support operations on the device 1200. Examples of such data include instructions for any application or method operating on the device 1200, contact data, phone book data, messages, pictures, videos, etc. The memory 1204 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.
[0145] The power supply component 1206 provides power to the various components of the device 1200. The power supply component 1206 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device 1200.
[0146] The multimedia component 1208 includes a screen that provides an output interface between the device 1200 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touch screen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touch, slide, and gestures on the touch panel. The touch sensor may not only sense the boundaries of the touch or slide action, but also detect the duration and pressure associated with the touch or slide operation. In some embodiments, the multimedia component 1208 includes a front camera and / or a rear camera. When the device 1200 is in an operating mode, such as a shooting mode or a video mode, the front camera and / or the rear camera may receive external multimedia data. Each front camera and rear camera may be a fixed optical lens system or have a focal length and optical zoom capability.
[0147] The audio component 1210 is configured to output and / or input audio signals. For example, the audio component 1210 includes a microphone (MIC), and when the device 1200 is in an operation mode, such as a call mode, a recording mode, and a speech recognition mode, the microphone is configured to receive an external audio signal. The received audio signal can be further stored in the memory 1204 or sent via the communication component 1216. In some embodiments, the audio component 1210 also includes a speaker for outputting audio signals.
[0148] I / O interface 1212 provides an interface between processing component 1202 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include but are not limited to: a home button, a volume button, a start button, and a lock button.
[0149] The sensor assembly 1214 includes one or more sensors for providing various aspects of the status assessment of the device 1200. For example, the sensor assembly 1214 can detect the open / closed state of the device 1200, the relative positioning of components, such as the display and keypad of the device 1200, the sensor assembly 1214 can also detect the position change of the device 1200 or a component of the device 1200, the presence or absence of user contact with the device 1200, the orientation or acceleration / deceleration of the device 1200, and the temperature change of the device 1200. The sensor assembly 1214 may include a proximity sensor configured to detect the presence of a nearby object without any physical contact. The sensor assembly 1214 may also include an optical sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, the sensor assembly 1214 may also include an accelerometer, a gyroscope sensor, a magnetic sensor, a pressure sensor, or a temperature sensor.
[0150] The communication component 1216 is configured to facilitate wired or wireless communication between the device 1200 and other devices. The device 1200 can access a wireless network based on a communication standard, such as WiFi, 2G or 3G, or a combination thereof. In an exemplary embodiment, the communication component 1216 receives a broadcast signal or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 1216 also includes a near field communication (NFC) module to facilitate short-range communication. For example, the NFC module can be implemented based on radio frequency identification (RFID) technology, infrared data association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology and other technologies.
[0151] In an exemplary embodiment, the device 1200 can be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors or other electronic components to perform the above-mentioned methods.
[0152] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 1204 including instructions, and the instructions can be executed by the processor 1220 of the device 1200 to perform the above method. For example, the non-transitory computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, an optical data storage device, etc.
[0153] A non-transitory computer-readable storage medium, when instructions in the storage medium are executed by a processor of a mobile terminal, enables the mobile terminal to perform a data processing method, the method comprising:
[0154] Input each labeled training data in the training set into a predetermined model to be trained to obtain a first prediction label;
[0155] Obtaining a training loss value according to a difference between the first predicted label and the labeled label of the labeled training data;
[0156] Based on the training loss value, updating the model parameters of the predetermined model;
[0157] The target network model constructed based on the network structure included in the trained predetermined model is used to classify the data to be classified to obtain a classification result.
[0158] Figure 5 1 is another block diagram of a data processing device 1300 according to an exemplary embodiment. For example, the device 1300 may be provided as a server. Figure 5 The apparatus 1300 includes a processing component 1322, which further includes one or more processors, and a memory resource represented by a memory 1332 for storing instructions executable by the processing component 1322, such as an application. The application stored in the memory 1332 may include one or more modules, each corresponding to a set of instructions. In addition, the processing component 1322 is configured to execute instructions to perform the above-mentioned data processing method, which includes:
[0159] Input each labeled training data in the training set into a predetermined model to be trained to obtain a first prediction label;
[0160] Obtaining a training loss value according to a difference between the first predicted label and the labeled label of the labeled training data;
[0161] Based on the training loss value, updating the model parameters of the predetermined model;
[0162] The target network model constructed based on the network structure included in the trained predetermined model is used to classify the data to be classified to obtain a classification result.
[0163] The device 1300 may also include a power supply component 1326 configured to perform power management of the device 1300, a wired or wireless network interface 1350 configured to connect the device 1300 to a network, and an input / output (I / O) interface 1358. The device 1300 may operate based on an operating system stored in the memory 1332, such as Windows Server™, MacOS X™, Unix™, Linux™, FreeBSD™, or the like.
[0164] Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed in this disclosure. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.
[0165] It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.
Claims
1. A data processing method, characterized in that: include: Input each labeled training data in the training set into a predetermined model to be trained to obtain a first prediction label; Obtaining a training loss value according to a difference between the first predicted label and the labeled label of the labeled training data; Based on the training loss value, updating the model parameters of the predetermined model; Normalizing the model parameters of each network structure in the trained predetermined model based on a predetermined type of activation function to obtain normalized model parameters; From the trained predetermined model, determining a network structure whose normalized model parameter value is greater than a set parameter threshold; Based on the network structure whose normalized model parameter value is greater than the set parameter threshold, a target network model is formed; Classify the data to be classified based on the target network model to obtain the classification result; Wherein, the predetermined model includes: a first type of network structure and a second type of network structure; The network function of the first type of network structure includes a skip connection parameter, and the skip connection parameter is used to reduce the difference between the output result of the first type of network structure and the output result of the second type of network structure by mixing random noise; The type of the data to be classified includes at least one of the following: an image data type, a text data type, and an audio data type.
2. The method according to claim 1, characterized in that The amplitude of the random noise mixed in as indicated by the jump connection parameter is smaller than the output result of the first type of network structure.
3. The method according to claim 1, characterized in that The model parameters include: network parameters and structural parameters; the updating of the model parameters of the predetermined model based on the training loss value includes: Based on the training loss value, updating the network parameters of the predetermined model; Inputting each labeled verification data in the verification set into the predetermined model after the network parameters are updated to obtain a second prediction label; Obtaining a verification loss value based on a difference between the second predicted label and the labeled verification data; Based on the verification loss value, the structural parameters of the predetermined model after the network parameters are updated are updated.
4. The method according to claim 1, characterized in that: The target network model constructed based on the network structure included in the trained predetermined model classifies the data to be classified to obtain a classification result, including: The data to be classified is input into the target network model, and M category probabilities of the data to be classified in M categories are output; wherein the category labels corresponding to the first preset number of category probabilities among the M category probabilities are the classification results of the data to be classified.
5. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: When the numerical change of the training loss value within a set time period is less than a set change threshold, the training of the predetermined model is stopped.
6. A data processing device, characterized in that: include: A first acquisition module is configured to input each labeled training data in the training set into a predetermined model to be trained to obtain a first prediction label; A second acquisition module is configured to obtain a training loss value according to a difference between the first predicted label and the labeled label of the labeled training data; An updating module, configured to update a model parameter of the predetermined model based on the training loss value; A classification module is configured to normalize the model parameters of each network structure in the trained predetermined model based on a predetermined type of activation function to obtain normalized model parameters; From the trained predetermined model, determining a network structure whose normalized model parameter value is greater than a set parameter threshold; Based on the network structure whose normalized model parameter value is greater than the set parameter threshold, a target network model is formed; Classify the data to be classified based on the target network model to obtain the classification result; Wherein, the predetermined model includes: a first type of network structure and a second type of network structure; The network function of the first type of network structure includes a skip connection parameter, and the skip connection parameter is used to reduce the difference between the output result of the first type of network structure and the output result of the second type of network structure by mixing random noise; The type of the data to be classified includes at least one of the following: an image data type, a text data type, and an audio data type.
7. The device according to claim 6, characterized in that The amplitude of the random noise mixed in as indicated by the jump connection parameter is smaller than the output result of the first type of network structure.
8. The device according to claim 6, characterized in that The model parameters include: network parameters and structural parameters; the update module is further configured to: Based on the training loss value, updating the network parameters of the predetermined model; Inputting each labeled verification data in the verification set into the predetermined model after the network parameters are updated to obtain a second prediction label; Obtaining a verification loss value based on a difference between the second predicted label and the labeled verification data; Based on the verification loss value, the structural parameters of the predetermined model after the network parameters are updated are updated.
9. The device according to claim 6, characterized in that The classification module is further configured as: The data to be classified is input into the target network model, and M category probabilities of the data to be classified in M categories are output; wherein the category labels corresponding to the first preset number of category probabilities among the M category probabilities are the classification results of the data to be classified.
10. The device according to any one of claims 6 to 9, characterized in that The device also includes: The stopping module is configured to stop training the predetermined model when the numerical change of the training loss value within a set time period is less than a set change threshold.
11. A data processing device, characterized in that: include: processor; a memory configured to store processor-executable instructions; Wherein, the processor is configured to: implement the steps in any one of the data processing methods in claims 1 to 5 when executed.
12. A non-transitory computer-readable storage medium, when instructions in the storage medium are executed by a processor of a data processing device, the device is enabled to perform any one of the data processing methods of claims 1 to 5.