Artificial intelligence-based deep learning model self-adaptive testing method and related device

Abstract

The application relates to the fields of artificial intelligence and digital medicine, and proposes a deep learning model self-adaptive testing method based on artificial intelligence and related equipment.The method comprises the following steps: obtaining a deep learning model and test data, the deep learning model can obtain feature representation and classification results of any test data; a first data set for storing the feature representation and the classification results is created; real-time test batches are input into the deep learning model to obtain real-time feature representation and real-time classification results of the test data; the first data set is queried based on the real-time feature representation to construct a feature consistency loss; proxy feature representation of each category in the first data set is calculated to construct a feature uniformity loss; the deep learning model and the first data set are updated based on the feature consistency loss and the feature uniformity loss; and the self-adaptive testing of the deep learning model is completed until a preset condition is met. The application can improve the precision of the deep learning model applied to the field of digital medicine on medical test data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence and digital medical technology, and particularly relates to a deep learning model self-adaptive testing method based on artificial intelligence and related equipment. BACKGROUND

[0002] Deep learning models have been widely used in various industries and fields, such as the digital medical field and the financial technology field, but the biggest problem is that the generalization of deep learning models is poor, that is, a deep learning model trained on one data set is difficult to obtain good results on another data set, especially when the distribution gap between the two data sets is large. The accuracy of the deep learning model is poor, for example, a deep learning model trained using one medical data set is difficult to obtain good results on another medical data set. How to ensure the accuracy of the deep learning model on other data sets is a problem to be solved.

[0003] Self-adaptive testing is a solution to the above problem. Self-adaptive testing uses test data to update the deep learning model when testing the deep learning model, so that the deep learning model can better conform to the distribution of the test data set, thereby improving the accuracy of the deep learning model on the test data set.

[0004] Currently, test data is usually input into a deep learning model in batches, a loss function is constructed according to the global information of the test data in the historical batch and the global information of the current batch, and then the deep learning model is updated. However, this approach does not take into account the feature distribution relationship between different test data in the historical batch, and does not fully utilize the test data in the historical batch, resulting in poor accuracy of the deep learning model on the test data set. SUMMARY

[0005] In view of the above, it is necessary to propose a deep learning model self-adaptive testing method based on artificial intelligence and related equipment to solve the technical problem of how to improve the accuracy of the deep learning model on the test data set in self-adaptive testing. The related equipment includes a deep learning model self-adaptive testing device based on artificial intelligence, an electronic device and a storage medium.

[0006] The present application provides a deep learning model self-adaptive testing method based on artificial intelligence, which comprises:

[0007] S10, obtaining a deep learning model and a plurality of test data, the deep learning model comprising at least a feature extraction layer and a classification layer, the feature extraction layer being used for feature extraction of any test data to output a feature representation, and the classification layer being used for classification of the feature representation to output a classification result;

[0008] S11, creating a first dataset for storing feature representations and classification results of each test data in an adaptive testing process;

[0009] S12, selecting a preset number of test data from the plurality of test data as a real-time test batch, and inputting the real-time test batch into the deep learning model to obtain real-time feature representations and real-time classification results of each test data;

[0010] S13, querying the first dataset based on the real-time feature representations to obtain similar feature representations of each test data, and constructing a feature consistency loss of the real-time test batch based on the similar feature representations, the real-time feature representations and the real-time classification results of each test data;

[0011] S14, calculating a proxy feature representation of each category in the first dataset, and calculating a distance between the real-time feature representation of each test data and the proxy feature representation of each category to construct a feature uniformity loss of the real-time test batch;

[0012] S15, constructing a target loss based on the feature consistency loss and the feature uniformity loss, and storing the real-time feature representations and the real-time classification results of each test data in the real-time test batch in the first dataset after updating the deep learning model based on the target loss;

[0013] S16, returning to step S12, and continuously updating the deep learning model until the value of the target loss is less than a preset value, and completing the adaptive testing of the deep learning model.

[0014] In some embodiments, the querying the first dataset based on the real-time feature representations to obtain similar feature representations of each test data comprises:

[0015] For any test data in the real-time test batch, calculating a similarity between the real-time feature representation of the test data and all feature representations in the first dataset;

[0016] Arranging all feature representations in the first dataset in descending order of the similarity;

[0017] Selecting K top-ranked feature representations as the similar feature representations of the test data;

[0018] After traversing all test data in the real-time test batch, obtaining the similar feature representations of each test data.

[0019] In some embodiments, the constructing a feature consistency loss of the real-time test batch based on the similar feature representations, the real-time feature representations and the real-time classification results of each test data comprises:

[0020] For any test data in the real-time test batch, a feature consistency sub-loss of the test data is calculated based on the real-time feature representation of the test data, the real-time classification result and the similar feature representations, and the feature consistency sub-loss satisfies a relationship:

[0021]

[0022] wherein z i is the real-time feature representation of test data i, K represents the total number of similar feature representations of the test data, z j is the jth similar feature representation corresponding to test data i, sim(z i , z j ) represents the similarity between z i and z j , p i is the real-time classification result of test data i, p j is the classification result corresponding to the jth similar feature representation in the first data set, L i is the feature consistency sub-loss of test data i;

[0023] The sum of the feature consistency sub-losses of all test data in the real-time test batch is taken as the feature consistency loss of the real-time test batch, and the feature consistency loss satisfies a relationship:

[0024]

[0025] wherein N is the number of all test data in the real-time test batch, L i is the feature consistency sub-loss of test data i, L mslc is the feature consistency loss of the real-time test batch.

[0026] In some embodiments, the classification result includes a probability value of the test data belonging to each category, and the calculation of the proxy feature representation of each category in the first data set comprises:

[0027] For any classification result in the first data set, the information entropy of the classification result is calculated;

[0028] If the information entropy is greater than a preset information entropy, the classification result is discarded, and if the information entropy is not greater than the preset information entropy, the category corresponding to the maximum probability value in the classification result is taken as a target category, and the feature representation corresponding to the classification result is written into the feature representation set of the target category;

[0029] After traversing all the classification results in the first data set, a feature representation set of each category is obtained;

[0030] The mean of all feature representations in the feature representation set of each category is calculated as the proxy feature representation of the category.

[0031] In some embodiments, the calculating the distance between the real-time feature representation of each test data and the proxy feature representation of each category to construct the feature uniformity loss of the real-time test batch comprises:

[0032] For any test data in the real-time test batch, the category distance between the real-time feature representation of the test data and the proxy feature representation of each category is calculated, which satisfies the relationship:

[0033]

[0034] wherein, and respectively represent the proxy feature representation of category c and category u, zi represents the real-time feature representation of test data i, represents the similarity between z i , Num represents the number of all categories, represents the category distance between the real-time feature representation of the test data i and the proxy feature representation of category c;

[0035] Based on the category distance and the real-time classification result of the test data, the feature uniformity sub-loss of the test data is constructed, which satisfies the relationship:

[0036]

[0037] wherein, is the probability value corresponding to category c in the real-time classification result of test data i, is the category distance between the real-time feature representation of test data i and the proxy feature representation of category c, Num represents the number of all categories, T i is the feature uniformity sub-loss of test data i;

[0038] Based on the category distance and the real-time classification result of the test data, the training factor of each test data is calculated, and the feature uniformity loss of the real-time test batch is constructed based on the training factor and the feature uniformity sub-loss of each test data.

[0039] In some embodiments, the training factor of each test data is calculated based on the category distance and the real-time classification result of the test data, and the feature uniformity loss of the real-time test batch is constructed based on the training factor of each test data and the feature uniformity sub-loss, comprising:

[0040] For any test data in the real-time test batch, the category corresponding to the minimum category distance is taken as the first attribution category of the test data, and the category corresponding to the maximum probability value in the real-time classification result is taken as the second attribution category of the test data;

[0041] If the first attribution category and the second attribution category are the same, the training factor of the test data is set to 1, and if the first attribution category and the second attribution category are different, the training factor of the test data is set to 0;

[0042] The feature uniformity loss of the real-time test batch is constructed based on the training factor of each test data and the feature uniformity sub-loss, and the feature uniformity loss satisfies the relationship:

[0043]

[0044] Wherein, is the training factor of test data i, T i is the feature uniformity sub-loss of test data i, N is the number of all test data in the real-time test batch, L tsd is the feature uniformity loss of the real-time test batch, and when L tsd = 0.

[0045] In some embodiments, the target loss satisfies the relationship:

[0046] LM= L tsd + λL mslc

[0047] Wherein, λ is a hyperparameter, L tsd is the feature uniformity loss of the real-time test batch, L mslc is the feature consistency loss of the real-time test batch, and LM is the target loss.

[0048] The embodiments of the present application also provide an artificial intelligence-based deep learning model adaptive testing device, comprising:

[0049] An acquisition unit is configured to acquire a deep learning model and a plurality of test data, wherein the deep learning model at least includes a feature extraction layer and a classification layer, the feature extraction layer is configured to perform feature extraction on any test data to output a feature representation, and the classification layer is configured to classify the feature representation to output a classification result.

[0050] a creating unit configured to create a first data set for storing feature representations and classification results of each test data in an adaptive testing process;

[0051] a selecting unit configured to select a preset number of test data from the plurality of test data as a real-time test batch, and input the real-time test batch into the deep learning model to obtain real-time feature representations and real-time classification results of each test data;

[0052] a feature consistency unit configured to query the first data set based on the real-time feature representations to obtain similar feature representations of each test data, and construct a feature consistency loss of the real-time test batch based on the similar feature representations, the real-time feature representations and the real-time classification results of each test data;

[0053] a feature uniformity unit configured to calculate proxy feature representations of each category in the first data set, and calculate distances between the real-time feature representations of each test data and the proxy feature representations of each category to construct a feature uniformity loss of the real-time test batch;

[0054] an updating unit configured to construct a target loss based on the feature consistency loss and the feature uniformity loss, and store the real-time feature representations and the real-time classification results of each test data in the real-time test batch in the first data set after updating the deep learning model based on the target loss;

[0055] an iterating unit configured to return to the selecting unit, and constantly update the deep learning model until a value of the target loss is less than a preset value, and complete the adaptive testing of the deep learning model.

[0056] Embodiments of the present application also provide an electronic device, which comprises:

[0057] a memory configured to store at least one instruction;

[0058] a processor configured to execute the instruction stored in the memory to implement the adaptive testing method of the deep learning model based on artificial intelligence.

[0059] Embodiments of the present application also provide a computer readable storage medium, which stores at least one instruction, and the at least one instruction is executed by a processor in an electronic device to implement the adaptive testing method of the deep learning model based on artificial intelligence.

[0060] In summary, the application stores all test data feature representations and classification results in the first data set. In the adaptive testing process, the feature uniformity loss and the feature consistency loss are constructed according to the test data in the real-time test batch and the first data set, which are used to constrain the deep learning model to learn the feature consistency between the same categories and the feature uniformity between different categories in the test data, so that the boundaries between different categories are more obvious. Through adaptive testing, the deep learning model can conform to and adapt to the feature distribution of the test data, thereby ensuring the accuracy of the deep learning model on the test data. BRIEF DESCRIPTION OF DRAWINGS

[0061] Figure 1 is a flowchart of a preferred embodiment of the adaptive testing method of the deep learning model based on artificial intelligence involved in the application.

[0062] Figure 2 is a function module diagram of a preferred embodiment of the adaptive testing device of the deep learning model based on artificial intelligence involved in the application.

[0063] Figure 3 is a structural schematic diagram of an electronic device of a preferred embodiment of the adaptive testing method of the deep learning model based on artificial intelligence involved in the application. DETAILED DESCRIPTION

[0064] In order to more clearly understand the purpose, features and advantages of the application, the application will be described in detail below in combination with the drawings and specific embodiments. It should be noted that the embodiments of the application and the features in the embodiments can be combined with each other without conflict. In the following description, a large number of specific details are set forth in order to facilitate a full understanding of the application, and the described embodiments are only part of the embodiments of the application, not all embodiments.

[0065] In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0066] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the application belongs. The terms used in the specification of the application are only for the purpose of describing the specific embodiments of the application, and are not intended to limit the application. The term "and / or" used herein includes any and all combinations of one or more related listed items.

[0067] This application provides an adaptive testing method for deep learning models based on artificial intelligence, which can be applied to one or more electronic devices. An electronic device is a device that can automatically perform numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0068] Electronic devices can be any electronic product that allows human-computer interaction with a customer, such as personal computers, tablets, smartphones, personal digital assistants (PDAs), game consoles, interactive network television (IPTV), smart wearable devices, etc.

[0069] Electronic devices may also include network devices and / or client devices. The network devices include, but are not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of hosts or network servers.

[0070] The networks in which electronic devices are located include, but are not limited to, the Internet, wide area networks, metropolitan area networks, local area networks, and virtual private networks (VPNs).

[0071] like Figure 1 The diagram shown is a flowchart of a preferred embodiment of the AI-based adaptive testing method for deep learning models according to this application. The order of steps in this flowchart can be changed, and some steps can be omitted, depending on different needs. The AI-based adaptive testing method for deep learning models provided in this application can be applied to any scenario requiring adaptive testing of deep learning models. Therefore, this method can be applied to products in these scenarios, such as classifying medical images based on deep learning models in the digital healthcare field.

[0072] S10, acquire a deep learning model and multiple test data. The deep learning model includes at least a feature extraction layer and a classification layer. The feature extraction layer is used to extract features from any test data to output a feature representation. The classification layer is used to classify the feature representation to output a classification result.

[0073] In an optional embodiment, the deep learning model is a model that has been trained on a training dataset, and the deep learning model comprises at least a feature extraction layer and a classification layer. The feature extraction layer can be a fully connected network or a convolutional neural network or a graph neural network, without limitation. The classification layer is a fully connected layer comprising a softmax function, which is a common classification function and will not be described here.

[0074] The training dataset can be a medical dataset for storing medical data such as medical images or medical texts. The medical images refer to internal organs such as the stomach, abdomen, heart, knee, and brain that are obtained in a non-invasive manner for medical or medical research, such as CT (Computed Tomography), MRI (Magnetic Resonance Imaging), US (ultrasonic), X-ray images, electroencephalogram, and optical photography images generated by medical instruments. The medical texts can be electronic healthcare records, electronic personal health records, including medical records, electrocardiograms, medical images, and a series of electronic records with archival value.

[0075] In an optional embodiment, the test data is data other than the training dataset. The types of test data include but are not limited to text information, image information, and voice information, and the types of test data are related to the specific application scenarios of the deep learning model. For example, when the application scenario is to classify medical images, the test data can be medical images or medical texts, or voice information in the process of intelligent diagnosis and treatment, etc.

[0076] In an optional embodiment, the input of the feature extraction layer is any test data, and the output is the feature representation corresponding to the test data. The feature representation is input into the classification layer to output the classification result of the test data, which includes the probability value of the test data belonging to each category. The feature representation can be used to represent the data features of the test data related to classification, and the number of categories is at least two, which is related to the specific application scenarios of the deep learning model.

[0077] In this way, the deep learning model trained on the training dataset and the plurality of test data are obtained, which provides a data basis and a model basis for deep learning model self-adaption test.

[0078] S11, a first dataset is created, which is used to store the feature representation and classification result of each test data in the self-adaption test process.

[0079] In an optional embodiment, the first data set stores the feature representation and the classification result of each test data in the adaptive testing process, and the feature representation and the classification result are one-to-one corresponding. It should be noted that before the adaptive testing starts, there is no data in the first data set, and the number of feature representations and classification results in the first data set gradually increases as the adaptive testing proceeds.

[0080] In this way, the first data set for storing the feature representation and the classification result of each test data in the adaptive testing process is created, and the target loss in the adaptive testing process is constructed.

[0081] S12, selecting a preset number of test data from the plurality of test data as a real-time test batch, and inputting the real-time test batch into the deep learning model to obtain the real-time feature representation and the real-time classification result of each test data.

[0082] In an optional embodiment, the preset number is 32, that is, 32 test data are included in one real-time test batch. Each test data in the real-time test batch is sequentially input into the deep learning model to obtain the real-time feature representation and the real-time classification result of each test data.

[0083] In this way, the real-time feature representation and the real-time classification result of each test data in one real-time test batch are obtained, which provides a data basis for constructing the target loss.

[0084] S13, querying the first data set based on the real-time feature representation to obtain the similar feature representation of each test data, and constructing the feature consistency loss of the real-time test batch based on the similar feature representation, the real-time feature representation and the real-time classification result of each test data.

[0085] In an optional embodiment, the querying the first data set based on the real-time feature representation to obtain the similar feature representation of each test data comprises:

[0086] For any test data in the real-time test batch, the similarity between the real-time feature representation of the test data and all feature representations in the first data set is calculated;

[0087] The feature representations in the first data set are arranged in descending order of the similarity;

[0088] The top K feature representations are selected as the similar feature representation of the test data;

[0089] After traversing all test data in the real-time test batch, the similar feature representation of each test data is obtained.

[0090] wherein K is a preset integer, the similarity can be selected as a cosine similarity, and in other embodiments, other similarity measurement methods such as a Euclidean distance, a Pearson correlation coefficient, a modified cosine similarity, a Hamming distance, or a Manhattan distance can be selected.

[0091] In an optional embodiment, the constructing the feature consistency loss of the real-time test batch based on the similar feature representation, the real-time feature representation, and the real-time classification result of each test data comprises:

[0092] For any test data in the real-time test batch, a feature consistency sub-loss of the test data is calculated based on the real-time feature representation of the test data, the real-time classification result, and the similar feature representation, and the feature consistency sub-loss satisfies a relationship formula:

[0093]

[0094] wherein z i is the real-time feature representation of test data i, K represents the total number of similar feature representations of the test data, z j is the jth similar feature representation corresponding to test data i, sim(z i , z j ) represents the similarity between z i , z j , p i is the real-time classification result of test data i, p j is the classification result corresponding to the jth similar feature representation in the first data set, L j is the feature consistency sub-loss of test data i.

[0095] The sum of the feature consistency sub-losses of all test data in the real-time test batch is taken as the feature consistency loss of the real-time test batch, and the feature consistency loss satisfies a relationship formula:

[0096]

[0097] wherein N is the number of all test data in the real-time test batch, L i is the feature consistency sub-loss of test data i, and L mslc is the feature consistency loss of the real-time test batch.

[0098] wherein K and N are both preset positive integers.

[0099] It should be noted that in the process of calculating the feature consistency loss, for each test data in the real-time test batch, K similar feature representations most similar to the test data are selected from the first data set, and the similar feature representation corresponding classification result is aligned with the real-time classification result of the test data to ensure that the deep learning model can learn the consistent features between test data of the same category, so that the distribution between the corresponding feature representations of test data of the same category is more compact.

[0100] In this way, the feature consistency loss of the real-time test batch is constructed, which is used to constrain the deep learning model to learn consistent features between test data of the same category, so that the distribution between the corresponding feature representations of test data of the same category is more compact.

[0101] S14, calculating the proxy feature representation of each category in the first data set, and calculating the distance between the real-time feature representation of each test data and the proxy feature representation of each category to construct the feature uniformity loss of the real-time test batch.

[0102] In an optional embodiment, the classification result includes a probability value of the test data belonging to each category, and the calculation of the proxy feature representation of each category in the first data set includes:

[0103] For any one classification result in the first data set, the information entropy of the classification result is calculated;

[0104] If the information entropy is greater than a preset information entropy, the classification result is discarded, and if the information entropy is not greater than the preset information entropy, the category corresponding to the maximum probability value in the classification result is taken as a target category, and the feature representation corresponding to the classification result is written into the feature representation set of the target category;

[0105] After traversing all classification results in the first data set, the feature representation set of each category is obtained;

[0106] The mean of all feature representations in the feature representation set of each category is calculated as the proxy feature representation of the category.

[0107] The information entropy can reflect the degree of information confusion in the classification result, and the greater the value of the information entropy, the less accurate the classification result, and the smaller the value of the information entropy, the more accurate the classification result; the value of the preset information entropy is 0.1; if the information entropy is greater than the preset information entropy, it means that the corresponding classification result has a large error, and discarding it can ensure the accuracy of the proxy feature representation of each category.

[0108] In an optional embodiment, the calculating the distance between the real-time feature representation of each test data and the proxy feature representation of each category to construct the feature uniformity loss of the real-time test batch comprises:

[0109] For any test data in the real-time test batch, the category distance between the real-time feature representation of the test data and the proxy feature representation of each category is calculated, and the category distance satisfies the relationship:

[0110]

[0111] wherein, and respectively represent the proxy feature representation of category c and category u, z i represents the real-time feature representation of test data i, represents the similarity between z i , Num represents the number of all categories, represents the category distance between the real-time feature representation of the test data i and the proxy feature representation of category c;

[0112] Based on the category distance and the real-time classification result of the test data, a feature uniformity sub-loss of the test data is constructed, and the feature uniformity sub-loss satisfies the relationship:

[0113]

[0114] wherein, is the probability value corresponding to category c in the real-time classification result of test data i, is the category distance between the real-time feature representation of test data i and the proxy feature representation of category c, Num represents the number of all categories, T i is the feature uniformity sub-loss of test data i;

[0115] Based on the category distance and the real-time classification result of the test data, a training factor of each test data is calculated, and based on the training factor and the feature uniformity sub-loss of each test data, a feature uniformity loss of the real-time test batch is constructed.

[0116] The category distance between the real-time feature representation of the test data and the agent feature representation of each category can represent the belonging relationship between the test data and different categories, that is, the smaller the category distance, the greater the probability that the test data belongs to the category corresponding to the category distance; and the belonging relationship should be consistent with the real-time classification result of the test data, so the feature uniformity sub-loss of the test data is constructed based on the category distance and the real-time classification result, which can constrain the deep learning model to learn the feature difference between different categories, so that the distribution of the feature representations corresponding to the test data of different categories is more easily distinguished.

[0117] In an optional embodiment, the training factor of each test data is calculated based on the category distance and the real-time classification result of the test data, and the feature uniformity loss of the real-time test batch is constructed based on the training factor and the feature uniformity sub-loss of each test data, including:

[0118] For any test data in the real-time test batch, the category corresponding to the minimum category distance is taken as the first belonging category of the test data, and the category corresponding to the maximum probability value in the real-time classification result is taken as the second belonging category of the test data;

[0119] If the first belonging category and the second belonging category are the same, the training factor of the test data is set to 1, and if the first belonging category and the second belonging category are different, the training factor of the test data is set to 0;

[0120] The feature uniformity loss of the real-time test batch is constructed based on the training factor and the feature uniformity sub-loss of each test data, and the feature uniformity loss satisfies the relationship:

[0121]

[0122] wherein, is the training factor of test data i, T i is the feature uniformity sub-loss of test data i, N is the number of all test data in the real-time test batch, L tsd is the feature uniformity loss of the real-time test batch, and when L tsd = 0.

[0123] It should be noted that when the first belonging category and the second belonging category are different, at least one of the first belonging category and the second belonging category is a wrong result, in order to ensure that the deep learning model can accurately learn the feature distribution of the test data, the training factor of the corresponding test data is set to 0, so that the test data does not participate in the subsequent update of the deep learning model.

[0124] Thus, the feature uniformity loss of the real-time test batch is constructed, which is used to constrain the deep learning model to learn the feature difference between different categories, so that the distribution of the feature representations corresponding to the test data of different categories is more easily distinguished.

[0125] S15, based on the feature consistency loss and the feature uniformity loss, a target loss is constructed, and after updating the deep learning model based on the target loss, the real-time feature representation and the real-time classification result of each test data in the real-time test batch are stored in the first data set.

[0126] In an optional embodiment, the target loss satisfies the relationship:

[0127] LM=L tsd +λL mslc

[0128] Wherein, λ is a hyperparameter, L tsd is the feature uniformity loss of the real-time test batch, L mslc is the feature consistency loss of the real-time test batch, and LM is the target loss.

[0129] In an optional embodiment, the deep learning model is updated according to the gradient descent method in the direction of reducing the value of the target loss, and after updating the deep learning model, the real-time feature representation and the real-time classification result of each test data in the real-time test batch are stored in the first data set, and the feature representation and the classification result in the first data set are updated.

[0130] Thus, the deep learning model is updated by the target loss function, so that the deep learning model can learn the feature distribution of the test data, so that the deep learning model learns the feature consistency between the same categories in the test data and the feature uniformity between different categories.

[0131] S16, return to step S12, and continuously update the deep learning model until the value of the target loss is less than a preset value, and the adaptive test of the deep learning model is completed.

[0132] In an optional embodiment, return to step S12 to select a new real-time test batch, and continuously update the deep learning model, and when the value of the target loss is less than a preset value, it indicates that the deep learning model has learned the feature consistency between the same categories in the test data and the feature uniformity between different categories, and the adaptive test of the deep learning model is completed. The preset value is 0.001.

[0133] Thus, the adaptive testing of the deep learning model is completed, and the deep learning model after the adaptive testing can conform to and adapt to the feature distribution of the test data, thereby ensuring the accuracy of the deep learning model on the test data.

[0134] From the above technical solutions, it can be seen that the application stores the feature representation and the classification result of all test data in the first data set by constructing the first data set. In the adaptive testing process, the feature uniformity loss and the feature consistency loss are constructed according to the test data in the real-time test batch and the first data set, which are used to constrain the deep learning model to learn the feature consistency between the same categories and the feature uniformity between different categories in the test data, so that the boundary between different categories is more obvious. Through adaptive testing, the deep learning model can conform to and adapt to the feature distribution of the test data, thereby ensuring the accuracy of the deep learning model on the test data.

[0135] Please refer to Figure 2 , Figure 2 is a functional module diagram of a preferred embodiment of the adaptive testing device for the deep learning model based on artificial intelligence. The adaptive testing device for the deep learning model based on artificial intelligence 11 includes an acquisition unit 110, a creation unit 111, a selection unit 112, a feature consistency unit 113, a feature uniformity unit 114, an update unit 115, and an iteration unit 116. The module / unit referred to in the application refers to a series of computer readable instructions that can be executed by the processor 13 and can complete a fixed function, which is stored in the memory 12. In this embodiment, the functions of each module / unit will be described in detail in the subsequent embodiments.

[0136] In an optional embodiment, the acquisition unit 110 is configured to acquire a deep learning model and a plurality of test data. The deep learning model includes at least a feature extraction layer and a classification layer. The feature extraction layer is configured to perform feature extraction on any test data to output a feature representation. The classification layer is configured to classify the feature representation to output a classification result.

[0137] In an optional embodiment, the creation unit 111 is configured to create a first data set. The first data set is configured to store the feature representation and the classification result of each test data in the adaptive testing process.

[0138] In an optional embodiment, the selection unit 112 is configured to select a preset number of test data from the plurality of test data as a real-time test batch, and input the real-time test batch into the deep learning model to obtain a real-time feature representation and a real-time classification result of each test data.

[0139] In an optional embodiment, the feature consistency unit 113 is configured to query the first dataset based on the real-time feature representation to obtain a similar feature representation for each test data, and construct a feature consistency loss for the real-time test batch based on the similar feature representation for each test data, the real-time feature representation, and the real-time classification result.

[0140] In some embodiments, the querying the first dataset based on the real-time feature representation to obtain a similar feature representation for each test data comprises:

[0141] For any test data in the real-time test batch, calculating a similarity between the real-time feature representation of the test data and all feature representations in the first dataset;

[0142] Arranging all feature representations in the first dataset in a descending order of the similarity;

[0143] Selecting top K feature representations as the similar feature representation for the test data;

[0144] After traversing all test data in the real-time test batch, obtaining the similar feature representation for each test data.

[0145] In some embodiments, the constructing a feature consistency loss for the real-time test batch based on the similar feature representation for each test data, the real-time feature representation, and the real-time classification result comprises:

[0146] For any test data in the real-time test batch, calculating a feature consistency sub-loss for the test data based on the real-time feature representation of the test data, the real-time classification result, and the similar feature representation, the feature consistency sub-loss satisfying a relationship:

[0147]

[0148] wherein z i is the real-time feature representation of test data i, K represents a total number of the similar feature representations of the test data, z j is the jth similar feature representation corresponding to the test data i, sim(z i , z j ) represents a similarity between z i and z j , p i is the real-time classification result of test data i, p j is a classification result corresponding to the jth similar feature representation in the first dataset, L i is the feature consistency sub-loss for test data i.

[0149] The sum of the feature consistency sub-losses of all the test data in the real-time test batch is taken as the feature consistency loss of the real-time test batch, and the feature consistency loss satisfies the relationship:

[0150]

[0151] wherein N is the number of all the test data in the real-time test batch, L j is the feature consistency sub-loss of test data i, L mslc is the feature consistency loss of the real-time test batch.

[0152] In an optional embodiment, the feature uniformity unit 114 is configured to calculate the proxy feature representation of each category in the first data set, and calculate the distance between the real-time feature representation of each test data and the proxy feature representation of each category to construct the feature uniformity loss of the real-time test batch.

[0153] In some embodiments, the classification result includes a probability value of the test data belonging to each category, and the calculation of the proxy feature representation of each category in the first data set includes:

[0154] For any classification result in the first data set, the information entropy of the classification result is calculated;

[0155] If the information entropy is greater than a preset information entropy, the classification result is discarded, and if the information entropy is not greater than the preset information entropy, the category corresponding to the maximum probability value in the classification result is taken as a target category, and the feature representation corresponding to the classification result is written into the feature representation set of the target category;

[0156] After traversing all the classification results in the first data set, the feature representation set of each category is obtained;

[0157] The mean of all the feature representations in the feature representation set of each category is calculated as the proxy feature representation of the category.

[0158] In some embodiments, the calculation of the distance between the real-time feature representation of each test data and the proxy feature representation of each category to construct the feature uniformity loss of the real-time test batch includes:

[0159] For any test data in the real-time test batch, the category distance between the real-time feature representation of the test data and the proxy feature representation of each category is calculated, and the category distance satisfies the relationship:

[0160]

[0161] wherein, and respectively represent the proxy feature representation of the category c and the category u, zi represents the real-time feature representation of the test data i, represents zi, the similarity between them, Num represents the number of all categories, represents the category distance between the real-time feature representation of the test data i and the proxy feature representation of the category c;

[0162] construct a feature uniformity sub-loss of the test data based on the category distance and the real-time classification result of the test data, and the feature uniformity sub-loss satisfies a relationship:

[0163]

[0164] wherein, is a probability value corresponding to the category c in the real-time classification result of the test data i, is the category distance between the real-time feature representation of the test data i and the proxy feature representation of the category c, Num represents the number of all categories, and T i is the feature uniformity sub-loss of the test data i;

[0165] calculate a training factor of each test data based on the category distance and the real-time classification result of the test data, and construct a feature uniformity loss of the real-time test batch based on the training factor and the feature uniformity sub-loss of each test data.

[0166] In some embodiments, the calculation of the training factor of each test data based on the category distance and the real-time classification result of the test data, and the construction of the feature uniformity loss of the real-time test batch based on the training factor and the feature uniformity sub-loss of each test data, include:

[0167] For any test data in the real-time test batch, the category corresponding to the minimum category distance is taken as the first belonging category of the test data, and the category corresponding to the maximum probability value in the real-time classification result is taken as the second belonging category of the test data;

[0168] If the first belonging category and the second belonging category are the same, the training factor of the test data is set to 1, and if the first belonging category and the second belonging category are different, the training factor of the test data is set to 0;

[0169] construct a feature uniformity loss of the real-time test batch based on the training factor and the feature uniformity sub-loss of each test data, and the feature uniformity loss satisfies a relationship:

[0170]

[0171] wherein, is a training factor of the test data i, T i is a feature uniformity sub-loss of the test data i, N is a number of all test data in the real-time test batch, L tsd is a feature uniformity loss of the real-time test batch, and when , L tsd = 0.

[0172] In an optional embodiment, the updating unit 115 is configured to construct a target loss based on the feature consistency loss and the feature uniformity loss, and store the real-time feature representation and the real-time classification result of each test data in the real-time test batch in the first data set after updating the deep learning model based on the target loss.

[0173] In some embodiments, the target loss satisfies the relationship:

[0174] L M = L tsd + λL mslc

[0175] wherein λ is a hyperparameter, L tsd is the feature uniformity loss of the real-time test batch, L mslc is the feature consistency loss of the real-time test batch, and L M is the target loss.

[0176] In an optional embodiment, the iteration unit 116 is configured to return the selecting unit to continuously update the deep learning model until the value of the target loss is less than a preset value, and complete the adaptive testing of the deep learning model.

[0177] As can be seen from the above technical solutions, the present application stores the feature representation and the classification result of all test data in the first data set, and in the adaptive testing process, the feature uniformity loss and the feature consistency loss are constructed according to the test data in the real-time test batch and the first data set, which are used to constrain the deep learning model to learn the feature consistency between the same categories and the feature uniformity between different categories in the test data, so that the boundary between different categories is more obvious; through adaptive testing, the deep learning model can conform to and adapt to the feature distribution of the test data, thereby ensuring the accuracy of the deep learning model on the test data.

[0178] Please refer to Figure 3 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. The electronic device 1 comprises a memory 12 and a processor 13. The memory 12 is configured to store computer readable instructions, and the processor 13 is configured to execute the computer readable instructions stored in the memory to implement the adaptive testing method of the deep learning model based on artificial intelligence according to any of the above embodiments.

[0179] In an optional embodiment, the electronic device 1 further comprises a bus, a computer program stored in the memory 12 and executable on the processor 13, for example, an artificial intelligence-based deep learning model adaptive testing program.

[0180] Figure 3 Only the electronic device 1 with the memory 12 and the processor 13 is shown, and those skilled in the art can understand that, Figure 3 The structure shown does not constitute a limitation on the electronic device 1, and can include fewer or more components than shown, or combine certain components, or different component arrangements.

[0181] In combination Figure 1 The memory 12 in the electronic device 1 stores a plurality of computer readable instructions to implement an artificial intelligence-based deep learning model adaptive testing method, and the processor 13 can execute the plurality of instructions to implement:

[0182] S10, obtaining a deep learning model and a plurality of test data, the deep learning model at least including a feature extraction layer and a classification layer, the feature extraction layer being used for feature extraction on any test data to output a feature representation, and the classification layer being used for classification on the feature representation to output a classification result;

[0183] S11, creating a first data set, the first data set being used to store the feature representation and the classification result of each test data in the adaptive testing process;

[0184] S12, selecting a preset number of test data from the plurality of test data as a real-time test batch, and inputting the real-time test batch into the deep learning model to obtain a real-time feature representation and a real-time classification result of each test data;

[0185] S13, querying the first data set based on the real-time feature representation to obtain a similar feature representation of each test data, and constructing a feature consistency loss of the real-time test batch based on the similar feature representation, the real-time feature representation and the real-time classification result of each test data;

[0186] S14, calculating a proxy feature representation of each category in the first data set, and calculating a distance between the real-time feature representation of each test data and the proxy feature representation of each category to construct a feature uniformity loss of the real-time test batch;

[0187] S15, constructing a target loss based on the feature consistency loss and the feature uniformity loss, and storing the real-time feature representation and the real-time classification result of each test data in the real-time test batch in the first data set after updating the deep learning model based on the target loss.

[0188] S16, returning to step S12, constantly updating the deep learning model until the value of the target loss is less than a preset value, completing the adaptive test of the deep learning model.

[0189] Specifically, the specific implementation method of the processor 13 to the above instructions can refer to Figure 1 The description of related steps in corresponding embodiments will not be repeated here.

[0190] Those skilled in the art can understand that the schematic diagram is only an example of the electronic device 1 and does not constitute a limitation on the electronic device 1. The electronic device 1 can be a bus type structure or a star type structure. The electronic device 1 can also include more or less other hardware or software or different component arrangements, for example, the electronic device 1 can also include an input / output device, a network access device, etc.

[0191] It should be noted that the electronic device 1 is only an example. Other existing or future electronic products, such as those adaptable to the present application, should also be included in the protection scope of the present application and are hereby incorporated by reference.

[0192] The memory 12 includes at least one type of readable storage medium, which can be non-volatile or volatile. The readable storage medium includes a flash memory, a mobile hard disk, a multimedia card, a card-type memory (e.g., an SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, etc. The memory 12 can be an internal storage unit of the electronic device 1 in some embodiments, such as a mobile hard disk of the electronic device 1. The memory 12 can also be an external storage device of the electronic device 1 in other embodiments, such as a plug-in mobile hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. The memory 12 can be used to store application software and various data installed in the electronic device 1, such as the code of the deep learning model adaptive test program based on artificial intelligence, and can also be used to temporarily store data that has been output or will be output.

[0193] The processor 13 can be composed of integrated circuits in some embodiments, for example, can be composed of a single packaged integrated circuit, or can be composed of multiple packaged integrated circuits with the same function or different functions, including one or more central processing units (CPU), microprocessors, digital processing chips, graphics processors, combinations of various control chips, etc. The processor 13 is the control core of the electronic device 1, which connects various components of the entire electronic device 1 through various interfaces and lines, executes programs or modules stored in the memory 12 (such as an artificial intelligence-based deep learning model adaptive test program), and calls data stored in the memory 12 to execute various functions and process data of the electronic device 1.

[0194] The processor 13 executes the operating system of the electronic device 1 and various installed application programs. The processor 13 executes the application programs to implement the steps in each of the above artificial intelligence-based deep learning model adaptive test method embodiments, such as Figure 1 The steps shown.

[0195] For example, the computer program can be divided into one or more modules / units, which are stored in the memory 12 and executed by the processor 13 to complete the present application. The one or more modules / units can be a series of computer-readable instruction segments that can complete a specific function, which are used to describe the execution process of the computer program in the electronic device 1. For example, the computer program can be divided into an acquisition unit 110, a creation unit 111, a selection unit 112, a feature consistency unit 113, a feature uniformity unit 114, an update unit 115, and an iteration unit 116.

[0196] The integrated units implemented in the form of software function modules described above can be stored in a computer-readable storage medium. The software function modules described above are stored in a storage medium, including a number of instructions for causing a computer device (which can be a personal computer, a computer device, or a network device, etc.) or a processor (Processor) to execute part of the artificial intelligence-based deep learning model adaptive test method described in each embodiment of the present application.

[0197] The modules / units integrated in the electronic device 1, if implemented in the form of software functional units and sold or used as independent products, can be stored in a computer-readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiments can also be implemented by a computer program instructing related hardware devices to complete, and the computer program can be stored in a computer-readable storage medium and can implement the steps of each method embodiment when executed by a processor.

[0198] The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms. The computer-readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory and other memories.

[0199] Further, the computer-readable storage medium can mainly include a storage program area and a storage data area, wherein the storage program area can store an operating system, at least one application required by a function, etc.; and the storage data area can store data created according to the use of the blockchain node, etc.

[0200] The blockchain referred to in the present application is a new application mode of distributed data storage, peer-to-peer transmission, consensus mechanism, encryption algorithm and other computer technologies. Blockchain, in essence, is a decentralized database, which is a series of data blocks associated using cryptographic methods, each data block containing information about a batch of network transactions, used to verify the validity (anti-fake) of the information and generate the next block. The blockchain can include a blockchain underlying platform, a platform product service layer, and an application service layer, etc.

[0201] The bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one arrow is used in Figure 3 , but it does not mean that there is only one bus or one type of bus. The bus is arranged to realize the connection and communication between the memory 12, the at least one processor 13 and the like.

[0202] The embodiment of the present application further provides a computer readable storage medium (not shown in the figure), which stores computer readable instructions, and the computer readable instructions are executed by a processor in an electronic device to implement the method for self-adapting test of the deep learning model based on artificial intelligence.

[0203] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other manners. For example, the above-described device embodiments are merely illustrative, and the division of the modules is merely a logical function division, and there can be another division manner in actual implementation.

[0204] The modules described as separated components can or can not be physically separated, and the components displayed as modules can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. According to actual needs, some or all of the modules can be selected to achieve the purpose of the embodiments.

[0205] In addition, each functional module in each embodiment of the present application can be integrated in one processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware, or in the form of hardware plus software functional modules.

[0206] In addition, it is obvious that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. The plurality of units or devices stated in the specification can also be implemented by one unit or device through software or hardware. The words "first", "second" and the like are used to indicate names, and do not indicate any specific order.

[0207] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit, although the present application has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present application.

Claims

1. An artificial intelligence-based deep learning model self-adaptive testing method, characterized in that, The method comprises: S10, obtaining a deep learning model and a plurality of test data, the deep learning model comprising at least a feature extraction layer and a classification layer, the feature extraction layer being used for feature extraction on any test data to output a feature representation, and the classification layer being used for classification on the feature representation to output a classification result; the test data comprising medical images, medical texts or voice information in a diagnosis and treatment process; S11, creating a first data set, the first data set being used for storing the feature representation and the classification result of each test data in an adaptive test process; S12, selecting a preset number of test data from the plurality of test data as a real-time test batch, and inputting the real-time test batch into the deep learning model to obtain a real-time feature representation and a real-time classification result of each test data; S13, querying the first data set based on the real-time feature representation to obtain a similar feature representation of each test data, and constructing a feature consistency loss of the real-time test batch based on the similar feature representation, the real-time feature representation and the real-time classification result of each test data; S14, calculating a proxy feature representation of each category in the first data set, and calculating a distance between the real-time feature representation of each test data and the proxy feature representation of each category to construct a feature uniformity loss of the real-time test batch; S15, constructing a target loss based on the feature consistency loss and the feature uniformity loss, and storing the real-time feature representation and the real-time classification result of each test data in the real-time test batch in the first data set after updating the deep learning model based on the target loss; S16, returning to step S12, and continuously updating the deep learning model until the value of the target loss is less than a preset value, and the adaptive test of the deep learning model is completed. 2.The AI-based deep learning model self-adaption test method of claim 1, wherein, The querying of the first data set based on the real-time feature representation to obtain the similar feature representation of each test data comprises: For any test data in the real-time test batch, calculating a similarity between the real-time feature representation of the test data and all feature representations in the first data set; Arranging all feature representations in the first data set in descending order of the similarity; Selecting the top K feature representations as the similar feature representation of the test data; After traversing all test data in the real-time test batch, obtaining the similar feature representation of each test data. 3.The AI-based deep learning model self-adaption test method of claim 1, wherein, The constructing of the feature consistency loss of the real-time test batch based on the similar feature representation, the real-time feature representation and the real-time classification result of each test data comprises: For any test data in the real-time test batch, calculating a feature consistency sub-loss of the test data based on the real-time feature representation, the real-time classification result and the similar feature representation of the test data, the feature consistency sub-loss satisfying a relationship formula: wherein, K represents the total number of similar feature representations of the test data i, represents the jth similar feature representation corresponding to the test data i, represents the similarity between the test data i and the jth similar feature representation, represents the real-time classification result of the test data i, represents the classification result corresponding to the jth similar feature representation in the first data set, represents the feature consistency sub-loss of the test data i; Taking a sum of the feature consistency sub-losses of all test data in the real-time test batch as the feature consistency loss of the real-time test batch, the feature consistency loss satisfying a relationship formula: wherein N is the number of all testing data in the real-time testing batch, is a feature consistency sub-loss of the testing data i, is a feature consistency loss of the real-time testing batch. 4.The AI-based deep learning model self-adaption test method of claim 1, wherein, The classification result includes a probability value of the test data belonging to each category, and the calculation of the proxy feature representation of each category in the first data set includes: For any classification result in the first data set, calculate the information entropy of the classification result; If the information entropy is greater than a preset information entropy, discard the classification result, if the information entropy is not greater than the preset information entropy, obtain the category corresponding to the maximum probability value in the classification result as a target category, and write the feature representation corresponding to the classification result into the feature representation set of the target category; After traversing all classification results in the first data set, the feature representation set of each category is obtained; Calculate the mean of all feature representations in the feature representation set of each category as the proxy feature representation of the category. 5.The AI-based deep learning model self-adaption test method of claim 1, wherein, The calculation of the distance between the real-time feature representation of each test data and the proxy feature representation of each category to construct the feature uniformity loss of the real-time test batch includes: For any test data in the real-time test batch, calculate the category distance between the real-time feature representation of the test data and the proxy feature representation of each category, which satisfies the relationship: wherein, and respectively represent the proxy feature representation of the class c and the proxy feature representation of the class u, represents the real-time feature representation of the test data i, represents the similarity between the real-time feature representation of the test data i and the proxy feature representation of the class c, Num represents the number of all classes, represents the class distance between the real-time feature representation of the test data i and the proxy feature representation of the class c; Based on the category distance and the real-time classification result of the test data, construct the feature uniformity sub-loss of the test data, which satisfies the relationship: wherein, is the probability value of the class c corresponding to the real-time classification result of the test data i, is the class distance between the real-time feature representation of the test data i and the proxy feature representation of the class c, and Num represents the number of all classes, is the feature uniformity sub-loss of the test data i; Based on the category distance and the real-time classification result of the test data, calculate the training factor of each test data, and based on the training factor of each test data and the feature uniformity sub-loss, construct the feature uniformity loss of the real-time test batch. 6.The AI-based deep learning model self-adaption test method of claim 5, wherein, The calculation of the distance between the real-time feature representation of each test data and the proxy feature representation of each category to construct the feature uniformity loss of the real-time test batch includes: For any test data in the real-time test batch, the category corresponding to the minimum category distance is taken as the first belonging category of the test data, and the category corresponding to the maximum probability value in the real-time classification result is taken as the second belonging category of the test data; If the first belonging category and the second belonging category are the same, the training factor of the test data is set to 1, if the first belonging category and the second belonging category are different, the training factor of the test data is set to 0; Based on the training factor of each test data and the feature uniformity sub-loss, construct the feature uniformity loss of the real-time test batch, which satisfies the relationship: wherein, is a training factor for test data i, is a feature uniformity sub-loss for test data i, and N is the number of all test data in the real-time test batch, is a feature uniformity loss for the real-time test batch, and when , . 7.The AI-based deep learning model self-adaption test method of claim 1, wherein, The target loss satisfies the relationship: wherein, is a hyperparameter, is a feature uniformity loss for the real-time test batch, is a feature consistency loss for the real-time test batch, is a target loss.

8. An artificial intelligence-based deep learning model self-adaptive testing device, characterized in that, The device includes: An acquisition unit is configured to acquire a deep learning model and a plurality of test data, the deep learning model at least includes a feature extraction layer and a classification layer, the feature extraction layer is configured to perform feature extraction on any test data to output a feature representation, and the classification layer is configured to classify the feature representation to output a classification result; the test data includes medical images, medical texts or voice information in the diagnosis and treatment process; The creating unit is configured to create a first data set for storing feature representations and classification results of each test data in an adaptive testing process; The selecting unit is configured to select a preset number of test data from the plurality of test data as a real-time test batch, and input the real-time test batch into the deep learning model to obtain real-time feature representations and real-time classification results of each test data; The feature consistency unit is configured to query the first data set based on the real-time feature representations to obtain similar feature representations of each test data, and construct a feature consistency loss of the real-time test batch based on the similar feature representations, the real-time feature representations and the real-time classification results of each test data; The feature uniformity unit is configured to calculate proxy feature representations of each category in the first data set, and calculate distances between the real-time feature representations of each test data and the proxy feature representations of each category to construct a feature uniformity loss of the real-time test batch; The updating unit is configured to construct a target loss based on the feature consistency loss and the feature uniformity loss, and store the real-time feature representations and the real-time classification results of each test data in the real-time test batch in the first data set after updating the deep learning model based on the target loss; The iteration unit is configured to return to the selecting unit, and continuously update the deep learning model until the value of the target loss is less than a preset value, and then complete the adaptive testing of the deep learning model.

9. An electronic device, comprising: The electronic device comprises: a memory storing computer readable instructions; and a processor executing the computer readable instructions stored in the memory to implement the adaptive testing method of the deep learning model based on artificial intelligence according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer readable instructions, and the computer readable instructions are executed by the processor to implement the adaptive testing method of the deep learning model based on artificial intelligence according to any one of claims 1 to 7.

Images