CN114927210A - Diagnostic correlation classification model training method, apparatus, device, medium, and program - Google Patents

Abstract

A diagnostic correlation classification model training method, apparatus, device, medium, and program, the method comprising: the control equipment sends the first parameters of the classification models to the training equipment, the training equipment trains the classification models respectively based on the first parameters and local training samples to obtain second parameters corresponding to the training equipment, and the classification models are arranged in the training equipment. The control equipment combines the second parameters corresponding to the training equipment to obtain third parameters, and informs the training equipment to update the parameters of the classification model to the third parameters. The classification model is trained in parallel by combining a plurality of training sample providers, and the finally trained classification model learns the characteristic information of the training samples of each training sample provider, so that the performance is improved, and the time cost is low.

Description

Diagnostic-related classification model training method, apparatus, equipment, medium and program

technical field

The invention relates to the technical field of artificial intelligence, and in particular, to a method, device, equipment, medium and program for training a diagnosis-related classification model.

Background technique

Classification tasks are encountered in many application scenarios. For example, classify images according to whether they contain a specific object, and classify articles according to their corresponding fields. Another example is DiagnosisRelated Groups (DRGs). DRGs refer to classifying patients into 500-2000 diagnosis-related groups based on factors such as the patient's hospital stay, clinical diagnosis, surgery, symptoms, disease severity, comorbidities and complications, and then deciding how much compensation should be given to the hospital.

With the continuous development of machine learning, neural network and other technologies, it has become a mainstream trend to use neural network models to complete classification and recognition tasks.

Taking DRGs prediction as an example, because the medical data of each medical institution is limited and the data is unevenly distributed, medical data is private and should not be shared among medical institutions. The way each medical institution conducts model training based on its own medical data may make the trained model perform poorly and be difficult to generalize.

SUMMARY OF THE INVENTION

Embodiments of the present invention provide a method, apparatus, device, medium and program for training a diagnosis-related classification model, which can obtain a model with good performance and versatility.

In a first aspect, an embodiment of the present invention provides a model training method, which is applied to a control device, and the method includes:

Sending the first parameter of the classification model to a plurality of training devices, so that the plurality of training devices respectively train the classification model based on the first parameter and the local training samples, so as to obtain the corresponding correspondence of the plurality of training devices The second parameter of the plurality of training devices is provided with the classification model;

receiving second parameters corresponding to each of the plurality of training devices;

combining the respective second parameters of the plurality of training devices to obtain the third parameter;

Notifying the plurality of training devices to update the parameters of the classification model to the third parameters.

In a second aspect, an embodiment of the present invention provides a model training device, which is applied to a control device, and the device includes:

The sending module is configured to send the first parameter of the classification model to multiple training devices, so that the multiple training devices respectively train the classification model based on the first parameter and local training samples to obtain the multiple training devices. second parameters corresponding to each of the plurality of training devices, the plurality of training devices correspond to a plurality of training sample providers, and each of the plurality of training devices is provided with the classification model;

a receiving module, configured to receive the respective second parameters corresponding to the multiple training devices;

a processing module, configured to combine the respective second parameters of the plurality of training devices to obtain the third parameters;

The sending module is further configured to: notify the plurality of training devices to update the parameters of the classification model to the third parameters.

In a third aspect, an embodiment of the present invention provides an electronic device, including: a memory and a processor; wherein, executable code is stored on the memory, and when the executable code is executed by the processor, the The processor may at least implement the model training method described in the first aspect.

In a fourth aspect, an embodiment of the present invention provides a non-transitory machine-readable storage medium, where executable code is stored on the non-transitory machine-readable storage medium, and when the executable code is executed by a processor of an electronic device When executed, the processor can at least implement the model training method described in the first aspect.

In a fifth aspect, an embodiment of the present invention provides an application program, which, when the application program is executed by a processor of an electronic device, causes the processor to execute the model training method according to the first aspect.

In a sixth aspect, an embodiment of the present invention provides a model training method, which is applied to a target training device, and the method includes:

receiving a first parameter of the classification model sent by a control device, the control device is connected to multiple training devices, and the target training device is any one of the multiple training devices;

training the classification model based on the first parameter and local training samples to obtain the second parameter corresponding to the target training device;

sending the second parameter corresponding to the target training device to the control device, so that the control device combines the second parameters corresponding to the multiple training devices to obtain the third parameter, wherein the multiple training devices The device separately trains the classification model based on the first parameter and local training samples to obtain second parameters corresponding to each of the plurality of training devices;

In response to the notification from the control device, the parameter of the classification model is updated to the third parameter.

In a seventh aspect, an embodiment of the present invention provides a model training device, which is applied to target training equipment, and the device includes:

a receiving module, configured to receive a first parameter of the classification model sent by a control device, the control device is connected to a plurality of training devices, and the target training device is any one of the plurality of training devices;

a training module, configured to train the classification model based on the first parameter and local training samples to obtain a second parameter corresponding to the target training device;

a sending module, configured to send the second parameter corresponding to the target training device to the control device, so that the control device combines the second parameters corresponding to the multiple training devices to obtain the third parameter, wherein, The plurality of training devices respectively train the classification model based on the first parameters and local training samples to obtain second parameters corresponding to the plurality of training devices;

An update module, configured to update the parameter of the classification model to the third parameter in response to the notification from the control device.

In an eighth aspect, an embodiment of the present invention provides an electronic device, including: a memory and a processor; wherein, executable code is stored on the memory, and when the executable code is executed by the processor, the The processor may at least implement the model training method described in the sixth aspect.

In a ninth aspect, an embodiment of the present invention provides a non-transitory machine-readable storage medium, where executable code is stored on the non-transitory machine-readable storage medium, and when the executable code is executed by a processor of an electronic device When executed, the processor can at least implement the model training method described in the sixth aspect.

In a tenth aspect, an embodiment of the present invention provides an application program, which, when the application program is executed by a processor of an electronic device, causes the processor to execute the model training method according to the sixth aspect.

In an eleventh aspect, an embodiment of the present invention provides a method for training a diagnosis-related classification model, which is applied to a control device. The method includes:

Sending the first parameter of the classification model to a plurality of training devices, so that the plurality of training devices respectively train the classification model based on the first parameter and the local training samples, so as to obtain the corresponding correspondence of the plurality of training devices The second parameter of , the classification model is provided in the multiple training devices, and the multiple training devices correspond to multiple medical institutions;

receiving second parameters corresponding to each of the plurality of training devices;

combining the respective second parameters of the plurality of training devices to obtain the third parameter;

Notifying the plurality of training devices to update the parameters of the classification model to the third parameters.

In a twelfth aspect, an embodiment of the present invention provides an apparatus for training a diagnosis-related classification model, which is applied to a control device, and the apparatus includes:

The sending module is configured to send the first parameter of the classification model to multiple training devices, so that the multiple training devices respectively train the classification model based on the first parameter and local training samples to obtain the multiple training devices. second parameters corresponding to each of the training devices, the classification models are all provided in the plurality of training devices, and the plurality of training devices correspond to a plurality of medical institutions;

a receiving module, configured to receive the respective second parameters corresponding to the multiple training devices;

a merging module for merging the respective second parameters of the plurality of training devices to obtain the third parameter;

The sending module is further configured to: notify the plurality of training devices to update the parameters of the classification model to the third parameters.

In a thirteenth aspect, an embodiment of the present invention provides an electronic device, including: a memory and a processor; wherein, executable code is stored on the memory, and when the executable code is executed by the processor, all The processor can at least implement the method for training a diagnosis-related classification model according to the eleventh aspect.

In a fourteenth aspect, an embodiment of the present invention provides a non-transitory machine-readable storage medium, where executable code is stored on the non-transitory machine-readable storage medium, and when the executable code is processed by an electronic device When the processor is executed, the processor can at least implement the method for training a diagnosis-related classification model according to the eleventh aspect.

In a fifteenth aspect, an embodiment of the present invention provides an application program, which, when the application program is executed by a processor of an electronic device, causes the processor to execute the method for training a diagnosis-related classification model as described in the eleventh aspect.

In a sixteenth aspect, an embodiment of the present invention provides a method for training a diagnosis-related classification model, which is applied to a target training device. The method includes:

Receive a first parameter of the classification model sent by a control device, the control device is connected to a plurality of training devices, the target training device is any one of the plurality of training devices, and the plurality of training devices are connected to a plurality of medical devices. institutional correspondence;

training the classification model based on the first parameter and local training samples to obtain the second parameter corresponding to the target training device;

sending the second parameter corresponding to the target training device to the control device, so that the control device combines the second parameters corresponding to the multiple training devices to obtain the third parameter, wherein the multiple training devices The device separately trains the classification model based on the first parameter and local training samples to obtain second parameters corresponding to each of the plurality of training devices;

In response to the notification from the control device, the parameter of the classification model is updated to the third parameter.

In a seventeenth aspect, an embodiment of the present invention provides an apparatus for training a diagnosis-related classification model, which is applied to target training equipment. The apparatus includes:

a receiving module, configured to receive the first parameter of the classification model sent by a control device, the control device is connected to multiple training devices, the target training device is any one of the multiple training devices, the multiple training devices The equipment corresponds to multiple medical institutions;

a training module, configured to train the classification model based on the first parameter and local training samples to obtain a second parameter corresponding to the target training device;

a sending module, configured to send the second parameter corresponding to the target training device to the control device, so that the control device combines the second parameters corresponding to the multiple training devices to obtain the third parameter, wherein, The plurality of training devices respectively train the classification model based on the first parameters and local training samples to obtain second parameters corresponding to the plurality of training devices;

An update module, configured to update the parameter of the classification model to the third parameter in response to the notification from the control device.

In an eighteenth aspect, an embodiment of the present invention provides an electronic device, including: a memory and a processor; wherein, executable code is stored on the memory, and when the executable code is executed by the processor, all The processor can at least implement the method for training a diagnosis-related classification model according to the sixteenth aspect.

In a nineteenth aspect, an embodiment of the present invention provides a non-transitory machine-readable storage medium, where executable code is stored on the non-transitory machine-readable storage medium, and when the executable code is processed by an electronic device When the processor is executed, the processor can at least implement the diagnosis-related classification model training method according to the sixteenth aspect.

In a twentieth aspect, an embodiment of the present invention provides an application program, which, when the application program is executed by a processor of an electronic device, causes the processor to execute the method for training a diagnosis-related classification model as described in the sixteenth aspect.

In the model training scheme provided by the above embodiments of the present invention, the training samples of multiple training sample providers need to be used to complete the training of the classification model, so that the finally trained classification model can be adapted to the classification task requirements of multiple training sample providers , which is universal. In order to ensure the information security of each training sample provider, each training sample provider locally trains the classification model based on its corresponding training device.

Specifically, the training process is as follows: first, the control device for managing multiple training devices will set the first parameter of the classification model, and then send the first parameter to each training device, so that each training device is based on the first parameter of the training device. The parameters and the local training samples train the locally stored classification model, that is, multiple training devices train the classification model based on the first parameter and the local training samples in parallel, and after training for a period of time, the corresponding classification model is obtained. A plurality of second parameters, one second parameter corresponds to one training device. Each training device uploads the second parameter obtained by its own training to the control device, the control device combines the respective second parameters of the multiple training devices to obtain the third parameter, and the control device informs the multiple training devices to use the local classification model. The parameter is updated to the third parameter. If it is finally determined that the classification model with the third parameter converges, the training of the classification model is ended, and the classification model with the third parameter is subsequently used to perform the respective classification tasks of the multiple training sample providers.

Since the third parameter incorporates the feature information learned by each training device based on its own training samples, the classification model with the third parameter can be generally applied to the classification tasks of multiple training sample providers, and can obtain more accurate Classification results.

Description of drawings

In order to illustrate the technical solutions in the embodiments of the present invention more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort.

1 is a schematic diagram of the composition of a model training system according to an embodiment of the present invention;

2 is a flowchart of a model training method provided by an embodiment of the present invention;

3 is a flowchart of a model testing method provided by an embodiment of the present invention;

4 is a flowchart of a model training method provided by an embodiment of the present invention;

5 is a flowchart of a model training method provided by an embodiment of the present invention;

6 is a flowchart of a method for training a diagnosis-related classification model according to an embodiment of the present invention;

7 is a flowchart of a method for training a diagnosis-related classification model according to an embodiment of the present invention;

8 is a schematic diagram of a model training scenario provided by an embodiment of the present invention;

9 is a schematic structural diagram of a model training apparatus provided by an embodiment of the present invention;

10 is a schematic structural diagram of an electronic device corresponding to the model training apparatus provided in the embodiment shown in FIG. 9;

FIG. 11 is a schematic structural diagram of a model training apparatus according to an embodiment of the present invention;

12 is a schematic structural diagram of an electronic device corresponding to the model training apparatus provided in the embodiment shown in FIG. 11;

13 is a schematic structural diagram of an apparatus for training a diagnosis-related classification model according to an embodiment of the present invention;

FIG. 14 is a schematic structural diagram of an electronic device corresponding to the apparatus provided by the embodiment shown in FIG. 13;

15 is a schematic structural diagram of an apparatus for training a diagnosis-related classification model according to an embodiment of the present invention;

FIG. 16 is a schematic structural diagram of an electronic device corresponding to the apparatus provided in the embodiment shown in FIG. 15 .

Detailed ways

In order to make the purposes, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments These are some embodiments of the present invention, but not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

The terms used in the embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit the present invention. The singular forms "a," "the," and "the" as used in the embodiments of the present invention and the appended claims are intended to include the plural forms as well, unless the context clearly dictates otherwise, "a plurality" Generally at least two are included.

Depending on the context, the words "if", "if" as used herein may be interpreted as "at" or "when" or "in response to determining" or "in response to detecting". Similarly, the phrases "if determined" or "if detected (the stated condition or event)" can be interpreted as "when determined" or "in response to determining" or "when detected (the stated condition or event)," depending on the context )" or "in response to detection (a stated condition or event)".

In addition, the sequence of steps in the following method embodiments is only an example, and is not strictly limited.

In practical applications, there will be such a situation: a classification model needs to be trained to complete multi-classification tasks. However, due to the limited training samples provided by a single training sample provider and the uneven data distribution (such as incomplete types of training samples) Or some types of training samples have a small number), if only the training samples provided by one training sample provider are used to complete the training of the classification model, the performance of the trained classification model may be poor and not universal.

A method of combining training samples provided by multiple training sample providers to train a classification model may be a transfer learning method. In some application scenarios, the training samples provided by the training sample provider are private and cannot be directly provided to others. Based on this, in the transfer learning mode, the training process of the classification model is simply: first, each Each training sample provider labels its own training samples with supervision information. For example, training sample provider 1 marks its first type of training samples as 01, and the second type of training samples as 02. The sample provider B is selected to mark its third type of training samples as 01 and the fourth type of training samples as 02. Among them, 01 and 02 represent category labels. After that, the training sample provider 1 trains the classification model 1 based on its own training samples, and then the training sample provider 1 tests the classification model 1 based on its own test samples. Assuming that the test results meet the requirements, the classification model 1 is provided for training. Sample provider 2. Training sample provider 2 uses its own training samples to train classification model 2 on the basis of classification model 1, and uses its own test samples to test classification model 2, assuming that the test results meet the requirements (note that classification model 2 does not must show good test results for the training sample provider 1), then continue to provide the classification model 3 to the next training sample provider 3, and the training sample provider 3 will, on the basis of the classification model 2, The training samples continue to be trained, and so on, assuming that the classification model n is finally trained by the training sample provider N. Finally, the classification model n is provided to the above n training sample providers, and each training sample provider will subsequently use the classification model n to complete their respective classification tasks. For example, any training sample provider i, when receiving a classification task triggered by a user, uses a classification model n to complete the classification task, so as to output a classification prediction result for the input data.

In the above transfer learning method, each training sample provider is used to train the classification model sequentially in a serial manner. When there are many training sample providers, the time cost of training the classification model is relatively large.

In addition, in the above transfer learning method, on the basis of the classification models trained by other training sample providers, the local training samples are used to continue training the classification models, and the performance (or quality) of the obtained classification models is obtained at this time. It depends on the number of local training samples and whether the data distribution of the training samples used by the previously trained classification model is similar to the data distribution of the local training samples. The resulting classification model does not perform well for local or other training sample providers. Therefore, the classification model finally obtained under the transfer learning method is not universal among multiple training sample providers, and cannot obtain a globally optimal classification performance.

In addition, in the above-mentioned transfer learning method, sharing the classification model between different training sample providers will also cause data security issues. For example, after the training sample provider 2 obtains the classification model 1, based on the parameters of the classification model 1, the data distribution characteristics of the training sample provider 1 for training the classification model 1 can be reversed, which brings data security risks to the training sample provider 1.

In order to avoid one or more problems existing in the above transfer learning manner, an embodiment of the present invention provides a new solution for training a classification model. Figuratively speaking, compared with the transfer learning method, the training of the classification model is completed in the federated learning method in the embodiment of the present invention. The so-called federated learning method is simply to combine multiple training sample providers to complete the training of the classification model in parallel, so as to save the training time cost, and make the classification model obtained by the final training have good performance among multiple training sample providers. Universality, to achieve good global performance (that is, the classification model obtained by the final training obtains better accuracy on the classification tasks of multiple training sample providers).

The following describes the training process of the classification model in the above federated learning manner with reference to the following embodiments.

FIG. 1 is a schematic diagram of the composition of a model training system provided by an embodiment of the present invention. As shown in FIG. 1 , the model training system includes: a control device and a plurality of training devices (such as the training device A shown in FIG. 1 , the training device B), a plurality of training devices are respectively connected in communication with the control device.

The multiple training devices may correspond to multiple training sample providers. Specifically, one training sample provider corresponds to one training device.

It is understandable that the training sample provider can provide the training samples required for training the classification model. However, after the training of the classification model is completed, the training sample provider can also be the user of the classification model and use the classification model to complete local classification tasks. .

Taking the training of a classification model for implementing DRGs (ie, a diagnosis-related classification model) as an example, the providers of the above-mentioned multiple training samples may be multiple medical institutions. Multiple medical institutions can use their own stored historical medical data as training samples to complete the training of the classification model. After the training of the classification model is completed, the medical institution can use the trained classification model to complete the classification task of DRGs prediction. A medical data to a classification model to obtain the predicted results of DRGs output by the classification model.

In the embodiment of the present invention, the control device is a device for controlling multiple training devices to perform model training. The control device can be a physical/virtual server or server cluster in the cloud, and the training device can be a terminal device or server located locally at the training sample provider. Each training device stores training samples and test samples of the corresponding training sample provider, so that the training samples and test samples of the training sample provider do not need to leave the training sample provider to ensure data security.

It can be understood that the training samples are used to train the classification model, and the test samples are used to verify the accuracy of the classification model.

In addition, in the embodiment of the present invention, a classification model is provided in multiple training devices, that is, the training process of the classification model is performed locally in multiple training devices (each training device is based on its own stored training samples locally training the classification model), and the control device only controls the training process.

The classification model in this embodiment of the present invention may be any neural network model that can implement a classification task, for example, it may be implemented as a recurrent neural network (Recurrent Neural Network, RNN for short) model, a long short-term memory network (Long Short-Term Memory) , referred to as LSTM) model, bidirectional long short-term memory network (Bi-directional Long Short-Term Memory, referred to as Bi-LSTM) model, and so on.

Based on the system architecture shown in FIG. 1 , from the perspective of the control device, in the process of training the classification model, the control device can perform the various steps shown in FIG. 2 .

FIG. 2 is a flowchart of a model training method provided by an embodiment of the present invention. As shown in FIG. 2 , the method includes the following steps:

201. The control device sends the first parameter of the classification model to multiple training devices, so that the multiple training devices respectively train the classification model based on the first parameter and the local training samples to obtain second parameters corresponding to each of the multiple training devices .

202. The control device receives second parameters corresponding to each of the multiple training devices.

203. The control device merges the second parameters corresponding to the multiple training devices to obtain the third parameter, and notifies the multiple training devices to update the parameters of the classification model to the third parameters.

In this embodiment, since multiple training sample providers are to be combined to train the classification model, it is necessary to label the training samples provided by the multiple training sample providers with unified supervision information, that is, multiple training samples The sample provider needs to complete the annotation of the supervision information of the training samples based on the same annotation standard. In addition, in practical applications, different training sample providers may have different habits of recording information, and there may be errors in recording. deal with.

Based on this, the control device may send data processing information to multiple training devices in advance, so that the multiple training devices perform data processing on local training samples according to the data processing information.

The data processing information includes labeling information corresponding to the training samples. In short, the labeling information describes what kind of class label should be labeled with what class of training samples. For example, tell all medical institutions to label a certain eye disease as 01, a certain heart disease as 02, and so on.

Optionally, the above-mentioned data processing information also includes at least one of the following information:

Data structuring processing information, which is used to perform set structuring processing on the data contained in each training sample;

Data filtering rule information, which is used to filter training samples that do not meet the set requirements.

For any training sample, the data information contained in the training sample can be organized into a data structure in the form of key-value pairs, expressed as key:value. Among them, key represents the key, and value represents the value. For example, taking medical data as an example, medical data will include data content such as age, various inspection indicators, symptoms, diagnosis results, etc., and each data content can be structured according to the set structural requirements. For example, a piece of medical data records: Zhang San, 25. Another piece of medical data records: Name: Li Si, 30. According to the record requirements of name and age specified in the data structure processing information, it can be converted into: Name: Zhang San; Age: 25. And, Name: Li Si; Age: 30.

The above data filtering rule information may describe filtering rules for various data contents included in the training samples. For example, if the age of a patient recorded in a piece of medical data is 150, this piece of medical data will be filtered because this age is unreasonable.

For ease of understanding, the execution process of the above model training method is exemplarily described with reference to the system shown in FIG. 1 .

In practical applications, when the classification model needs to be trained, the control device may first initialize the parameters of the classification model, assuming that the initialization is the first parameter, which is represented as W0. It can be understood that the process of training the classification model is the process of iteratively updating the parameters of the classification model.

After that, the control device sends the first parameter W0 to a plurality of training devices, such as the training device A and the training device B shown in FIG. 1 . The multiple training devices locally store the same classification model with the same structure, and after receiving the first parameter W0, the multiple training devices set the parameters of the locally stored classification model to the first parameter W0.

After that, each training device uses the local training samples to train the classification model whose parameter is set as the first parameter W0, and after training for a period of time, each training device obtains the second parameter corresponding to the locally maintained classification model. For example, in FIG. 1, training device A trains the classification model with the first parameter W0 for a period of time based on local training samples to obtain the second parameter WA, and training device B classifies the classification model with the first parameter W0 based on the local training samples After the model is trained for a period of time, the second parameter WB is obtained. Afterwards, the training device A sends the second parameter WA to the control device, and similarly, the training device B sends the second parameter WB to the control device.

Wherein, optionally, the above-mentioned "a period of time" may be a preset period of time. In this case, the preset period of time is applicable to multiple training devices, for example, all training devices train the classification model with the first parameter W0 30 minutes. Optionally, "a period of time" can also determine the length of the "period of time" corresponding to each training device according to the total number of training samples stored in each training device, such as training each training device to have a classification with the first parameter W0. The period of time for the models is set to be 60% of the total number of training samples corresponding to each.

After the control device receives the respective second parameters corresponding to the multiple training devices, the second parameters are combined to obtain the third parameter W1. In FIG. 1, it is assumed that the control device obtains the third parameter W1 after combining the second parameter WA and the second parameter WB, and notifies multiple training devices to update the parameters of the local classification model to the third parameter W1.

Optionally, the control device may perform weighted average processing on the second parameters corresponding to each of the plurality of training devices to obtain the third parameter W1.

The weight of the second parameter corresponding to each training device may be determined according to the total number of training samples stored in the training device and/or the level category of the training sample provider corresponding to the training device.

Taking the training sample provider as a medical institution as an example, the weight of the second parameter corresponding to each medical institution may be determined according to the preset grade category corresponding to the medical institution. For example, the weights corresponding to the same level category are the same, and the weights corresponding to different level categories are different.

When the weight of the second parameter corresponding to each training device is determined according to the total number of training samples, the weight value corresponding to the total number of training samples in different value ranges can be preset, and the weight of the second parameter corresponding to each training device can be determined accordingly. .

After obtaining the third parameter W1, the control device may also determine whether the classification model with the third parameter W1 meets a set condition, and the set condition may be a convergence condition. In short, it is to determine whether the classification model with the third parameter W1 can generally meet the classification task requirements of multiple training sample providers and obtain good performance.

In general, the control device can test the classification model with the third parameter W1 by controlling a plurality of training devices to determine whether the classification model with the third parameter W1 meets the set condition. The testing process can be performed with reference to the embodiment shown in FIG. 3 .

FIG. 3 is a flowchart of a model testing method provided by an embodiment of the present invention. As shown in FIG. 3 , the control device can determine whether the classification model with the third parameter W1 meets the set conditions through the following steps:

301. The control device sends the third parameter W1 to multiple training devices, so that the multiple training devices respectively test the classification model with the third parameter W1 according to local test samples to obtain respective test indicators corresponding to the multiple training devices.

302. The control device receives respective test indicators corresponding to the multiple training devices.

303. The control device determines whether the classification model with the third parameter W1 meets the set condition according to the respective test indicators corresponding to the multiple training devices.

Specifically, the control device sends the third parameter W1 to a plurality of training devices, and at this time, each training device (such as training device A and training device B in FIG. 1 ) sets the parameters of various stored classification models as the third parameter W1. After that, each training device uses the local test samples to test the classification model whose parameter is set as the third parameter W1 to obtain the test index. Test metrics may include classification accuracy.

After that, each training device sends its own test index to the control device, and the control device determines whether the classification model with the third parameter W1 meets the set condition according to the respective test indexes corresponding to the plurality of training devices.

Optionally, the control device may perform weighted average processing on the test indicators corresponding to each of the multiple training devices, and if the weighted average processing result is greater than or equal to the set threshold, it is determined that the classification model with the third parameter W1 meets the set condition.

Wherein, optionally, the control device may determine the weights of the respective test indicators corresponding to the multiple training devices according to at least one of the following information:

The number of training samples corresponding to each of the multiple training sample providers, the set level category corresponding to each of the multiple training sample providers, and the test indicators corresponding to each of the multiple training equipment.

For example, for the training device A and the training device B shown in FIG. 1 , it is assumed that the classification accuracy obtained by the training device A testing the classification model with the third parameter W1 is Pa, and the training device B has the third parameter W1. The classification accuracy of the classification model of parameter W1 is Pb, and it is assumed that Pa is greater than Pb. Assuming that the weights of the respective test indicators corresponding to the multiple training devices are determined only based on the respective test indicators corresponding to the multiple training devices, it can be determined that the weight corresponding to Pa is e1, and the weight corresponding to Pb is e2, where e1 is greater than e2. Wherein, a plurality of value ranges of classification accuracy rates and a weight corresponding to each value range may be preset, and a value range with a higher classification accuracy rate corresponds to a higher weight.

Similarly, for the level categories that the training sample provider may correspond to, weights corresponding to different level categories may be preset, and the weights of the respective test indicators corresponding to the multiple training devices are determined accordingly.

Similarly, the corresponding relationship between the number of training samples in different number intervals and the weights may be preset, and the weights of the test indicators corresponding to each of the plurality of training devices are determined accordingly.

After determining the weights of the respective test indicators corresponding to the multiple training devices, the control device performs weighted average processing on the respective test indicators corresponding to the multiple training devices (that is, after adding up the weights, and then dividing by the total number of training devices), if If the weighted average processing result is greater than or equal to the set threshold, it is determined that the classification model with the third parameter W1 meets the set condition; otherwise, if it is less than the set threshold, it is determined that the classification model with the third parameter W1 does not meet the set condition.

When the control device determines that the classification model with the third parameter W1 meets the set conditions, it means that the classification model with the third parameter W1 shows a better effect on multiple training sample providers. At this time, the control device can notify Multiple training devices can stop model training so that they use the classification model with the third parameter W1 to process subsequent classification tasks, that is, the classification model with the third parameter W1 is the classification model obtained by final training.

When the control device determines that the classification model with the third parameter W1 does not meet the set condition, it means that the classification model with the third parameter W1 needs to be further trained to optimize the classification model. At this time, the control device sends a notification message to the multiple training devices, so that the multiple training devices respectively continue to train the classification model with the third parameter based on the local training samples. That is, each training device continues to train the classification model with the third parameter W1 by using the local training samples based on the classification model with the third parameter W1.

After a plurality of training devices use their respective training samples to train the classification model with the third parameter W1 for a period of time, the plurality of training devices will obtain their corresponding fourth parameters respectively, and then send the fourth parameters to the control device, and the control device The fourth parameters corresponding to each of the multiple training devices are combined to obtain the fifth parameter. After that, the control device controls a plurality of training devices to test the classification model with the fifth parameter to determine whether the classification model with the fifth parameter meets the set condition. For the above execution process, reference may be made to the relevant descriptions above, which will not be repeated here. It is understandable that the above training and testing processes are performed iteratively until a classification model with certain parameters that meets the set conditions is trained.

To sum up, in the federated learning mode, multiple training devices are controlled by the control device to train the classification model in parallel, so that the training time cost is not restricted by the number of training sample providers. In addition, by scheduling multiple training devices to train the classification model locally at the same time, the classification model is not transferred between different training devices, which helps to protect the data security of different training sample providers. In addition, during the training process of the classification model, the data distribution characteristics of multiple training sample providers are combined to determine the parameters of the classification model, so that the final classification model has good generality among multiple training sample providers, and can obtain better performance.

FIG. 4 is a flowchart of a model training method provided by an embodiment of the present invention. The method may be executed by a control device. As shown in FIG. 4 , the method may include the following steps:

401. The control device sends the first parameter of the classification model to multiple training devices, so that the multiple training devices respectively train the classification model based on the first parameter and the local training samples to obtain second parameters corresponding to each of the multiple training devices. .

402. The control device receives the encrypted second parameters corresponding to each of the multiple training devices.

403. The control device sends the encrypted second parameters corresponding to the multiple training devices to the trusted execution space, so as to decrypt the second parameters corresponding to the multiple training devices in the trusted execution space, and perform the operation of the multiple training devices. The corresponding second parameters are transformed.

404. The control device merges the transformed second parameters corresponding to each of the multiple training devices to obtain the third parameter, and notifies the multiple training devices to update the parameters of the classification model to the third parameters.

In order to further improve the data security of multiple training sample providers, the solution of this embodiment is provided.

In this embodiment, the core point of improving the data security of multiple training sample providers is that after the training device corresponding to each training sample provider trains the classification model based on the local training samples to obtain the parameters of the classification model, the subsequent During the transmission of the parameter, the security of the parameter should be ensured to avoid inferring the data distribution characteristics of the training sample provider based on the parameter.

Continuing from the previous embodiment, specifically, when each training device receives the first parameter sent by the control device, and uses the local training samples to train the classification model with the first parameter for a period of time to obtain the corresponding second parameter, The second parameter is encrypted and transmitted to the control device.

Still taking the example in the foregoing embodiment, after obtaining the second parameter WA, the training device A encrypts the second parameter WA, obtains the encrypted second parameter WA', and sends the encrypted second parameter WA' to the control device. . Similarly, after obtaining the second parameter WB, the training device B encrypts the second parameter WB to obtain the encrypted second parameter WB', and sends the encrypted second parameter WB' to the control device. The encryption method can be any existing encryption method, such as asymmetric encryption.

The control device does not have the right to directly decrypt the encrypted second parameter WA' and the encrypted second parameter WB', because if the control device directly decrypts the second parameter WA and the second parameter WB in plaintext, the two The training sample provider corresponding to the training device poses a data security threat.

Specifically, the control device sends the encrypted second parameter WA' and the encrypted second parameter WB' to a trusted execution space, and decrypts the second parameter WA and the second parameter WB in the trusted execution space, because The second parameter WA and the second parameter WB are located in the trusted execution space and cannot be accessed by the outside world, so there is no problem of leaking the second parameter WA and the second parameter WB. In addition, the second parameter WA and the second parameter WB are further subjected to a certain set of transformation processing in the trusted execution space, and the transformation processing result is fed back to the control device. It is assumed that the second parameter WA is transformed after transformation processing. After transforming the second parameter WAA, the second parameter WB is transformed to obtain the transformed second parameter WBB. In this way, the transformed second parameter WAA and the transformed second parameter WBB obtained by the control device do not directly obtain the plaintext second parameter WA and the second parameter WB.

After that, the control device combines the transformed second parameter WAA and the transformed second parameter WBB to obtain a third parameter, and sends the third parameter to each training device.

Wherein, optionally, the above transformation processing may include: differential privacy processing. Of course, it is not limited to this, and other transformation processing methods may also be used. However, it is worth noting that no matter which transformation method is adopted, there are two purposes: first, the second parameter of the plaintext is not transmitted; second, the data distribution of the training samples reflected by the transformed second parameter The features are the same as or similar to the data distribution features of the training samples reflected by the untransformed second parameter. That is to say, the transformation processing methods that satisfy the above two conditions can be adopted.

In practical applications, the above-mentioned trusted execution space can be implemented by any existing trusted computing technology, such as SGX (software guard extensions).

FIG. 5 is a flowchart of a model training method provided by an embodiment of the present invention. The method may be executed by a target training device. As shown in FIG. 5 , the method may include the following steps:

501. The target training device receives the first parameter of the classification model sent by the control device, the control device is connected to multiple training devices, and the target training device is any one of the multiple training devices.

As described above, optionally, the target training device may also receive data processing information sent by the control device before, and perform data processing on the local training samples according to the received data processing information. The data processing information includes labeling information corresponding to the training samples.

Optionally, the data processing information also includes at least one of the following information:

Data structuring processing information, which is used to perform set structuring processing on the data contained in each training sample;

Data filtering rule information, which is used to filter training samples that do not meet the set requirements.

502. The target training device trains the classification model based on the first parameter and the local training sample to obtain the second parameter corresponding to the target training device.

503. The target training device sends the second parameter corresponding to itself to the control device, so that the control device merges the second parameters corresponding to each of the multiple training devices to obtain the third parameter, wherein the multiple training devices are based on the first parameter and the local The training samples of the training samples respectively train the classification model to obtain the second parameters corresponding to each of the multiple training devices.

As described above, optionally, the target training device may encrypt the second parameter corresponding to itself, and then send the encrypted second parameter corresponding to the target training device to the control device. The same goes for other training equipment.

504. The target training device updates the parameter of the classification model to the third parameter in response to the notification from the control device.

As mentioned above, the control device also needs to determine whether the classification model with the third parameter meets the set condition, and the determination process is realized by controlling a plurality of training devices to test the classification model with the third parameter. To complete this testing process, the target training device performs the following steps:

Test the classification model with the third parameter according to the local test sample to obtain the test index corresponding to the target training device;

The test index corresponding to the target training device is sent to the control device, so that the control device determines whether the classification model with the third parameter meets the set condition according to the test index corresponding to each of the multiple training devices, wherein the multiple training devices are based on the local The test samples respectively test the classification model with the third parameter to obtain respective test indicators corresponding to the plurality of training devices.

As mentioned above, if the control device determines that the classification model with the third parameter meets the set condition, it notifies each training device including the target training device to use the classification model with the third parameter to perform various classification tasks. Conversely, if the control device determines that the classification model with the third parameter does not meet the set condition, the third parameter is sent to each training device. At this time, the target training device receives the third parameter sent by the control device when it is determined that the classification model with the third parameter does not meet the set condition, and then continues to train the classification model based on the third parameter and the local training samples.

In this embodiment, for the detailed execution process of the target training device in the classification model training process, reference may be made to the relevant descriptions in the foregoing embodiments, and details are not described herein.

The training solution for the classification model provided by the embodiment of the present invention may be applicable to any application scenario that needs to complete the training of a shared classification model by combining the training samples provided by multiple training sample providers. For example, in the DGRs application scenario mentioned above.

In the DGRs application scenario, a diagnosis-related classification model needs to be trained. To complete the training of the classification model, as shown in Figure 6, the control device needs to perform the following steps:

601. The control device sends the first parameter of the classification model to multiple training devices, so that the multiple training devices respectively describe the classification model for training based on the first parameter and the local training samples to obtain the respective second parameters of the multiple training devices. , the classification model is provided in multiple training devices, and multiple training devices correspond to multiple medical institutions.

602. The control device receives second parameters corresponding to each of the multiple training devices.

603. The control device merges the second parameters corresponding to the multiple training devices to obtain the third parameter, and notifies the multiple training devices to update the parameters of the classification model to the third parameters.

Correspondingly, in order to complete the training of the classification model, as shown in FIG. 7 , the target training device among the multiple training devices needs to perform the following steps:

701. The target training device receives the first parameter of the classification model sent by the control device, the control device is connected to multiple training devices, the target training device is any one of the multiple training devices, and the multiple training devices correspond to multiple medical institutions.

702. The target training device trains the classification model based on the first parameters and local training samples to obtain second parameters corresponding to the target training device.

703. The target training device sends the second parameter corresponding to the target training device to the control device, so that the control device merges the second parameters corresponding to each of the multiple training devices to obtain the third parameter, wherein the multiple training devices are based on the first parameter. The classification model is trained separately with the local training samples to obtain second parameters corresponding to each of the multiple training devices.

704. In response to the notification from the control device, the target training device updates the parameter of the classification model to the third parameter.

The following exemplarily describes the execution process of the model training method provided by the embodiment of the present invention in the DGRs application scenario with reference to FIG. 8 .

In Figure 8, it is assumed that the respective medical data of medical institution X and medical institution Y need to be combined to complete the classification model for realizing DGRs prediction.

In the initial stage, it is assumed that the classification model has been created, and the parameter of the control device to initialize the classification model is the first parameter C1, and the control device sends the first parameter C1 to medical institution X and medical institution Y, and medical institution X and medical institution Y send the local The parameters of the maintained classification model are set as the first parameter C1, and then the medical institution X and the medical institution Y use their respective local medical data (training samples) to train the classification model with the first parameter C1. After a period of training, medical institution X obtains the second parameter CA2, medical institution Y obtains the second parameter CB2, medical institution X encrypts and transmits the second parameter CA2 to the control device, and medical institution Y encrypts and transmits the second parameter CB2 to the control device .

The control device sends the obtained encryption parameters to the trusted execution space, decrypts the second parameter CA2 and the second parameter CB2 in the trusted execution space, and processes the two second parameters respectively through the privacy differential method to obtain the first parameter. The second parameter CA2' and the second parameter CB2'.

The control device obtains the second parameter CA2' and the second parameter CB2' calculated in the trusted execution space, and combines these two parameters to obtain the third parameter C2. The third parameter C2 is sent to the medical institution X and the medical institution Y, and the medical institution X and the medical institution Y are ordered to test the classification model with the third parameter C2.

Medical institution X and medical institution Y use their respective local test data to test the classification model with the third parameter C2, obtain the classification accuracy rates PA and PB, and send the classification accuracy rates to the control device.

The control device determines that the classification model with the third parameter C2 meets the conditions based on the classification accuracy rates PA and PB, and notifies the medical institution X and the medical institution Y to use the classification model with the third parameter C2 to complete the subsequent DGRs prediction task.

In this way, assuming that medical institution X generates medical data subsequently, when the medical data needs to be predicted by DGRs, the medical data is input into the classification model with the third parameter C2, and the classification output result of the classification model is obtained.

The model training scheme provided by the embodiments of the present invention can be applied not only to the application scenarios of DRGs, but also to other application scenarios. For example, in financial scenarios, for data security reasons, the data of different financial institutions are also stored independently and in isolation. When it is necessary to combine the data of multiple financial institutions to train a data processing system (such as data classification) that can be applied to each financial institution When the model is used, the solution provided by the embodiment of the present invention can also be applied. For another example, in the government affairs scenario, the data between different government affairs systems are also independent of each other, and in the e-commerce scenario, the data of different businesses are also independent. When it is necessary to combine the data of multiple government affairs systems or multiple merchants to train a certain model, the solution provided by the embodiment of the present invention can be adopted.

The model training apparatus of one or more embodiments of the present invention will be described in detail below. Those skilled in the art can understand that these devices can be configured by using commercially available hardware components through the steps taught in this solution.

FIG. 9 is a schematic structural diagram of a model training apparatus according to an embodiment of the present invention, where the apparatus is applied to a control device. As shown in FIG. 9 , the apparatus includes: a sending module 11 , a receiving module 12 , and a processing module 13 .

The sending module 11 is configured to send the first parameter of the classification model to multiple training devices, so that the multiple training devices respectively train the classification model based on the first parameter and local training samples to obtain the The second parameter corresponding to each of the plurality of training devices, and the classification model is provided in each of the plurality of training devices.

The receiving module 12 is configured to receive the respective second parameters corresponding to the multiple training devices.

The processing module 13 is configured to combine the respective second parameters corresponding to the multiple training devices to obtain the third parameter.

The sending module 11 is further configured to: notify the plurality of training devices to update the parameters of the classification model to the third parameters.

Optionally, the processing module 13 is specifically configured to: perform weighted average processing on the second parameters corresponding to the plurality of training devices to obtain the third parameter.

Optionally, the receiving module 12 is specifically configured to: receive the encrypted second parameters corresponding to each of the multiple training devices. At this time, the processing module 13 is specifically configured to: send the encrypted second parameters corresponding to the multiple training devices to the trusted execution space, so as to decrypt the multiple training devices in the trusted execution space Transform the second parameters corresponding to each of the multiple training devices, and combine the transformed second parameters corresponding to the multiple training devices to obtain the third parameter.

Wherein, optionally, the transformation processing includes: differential privacy processing.

Optionally, the sending module 11 is further configured to: if it is determined that the classification model with the third parameter does not meet the set condition, notify the plurality of training devices to continue the training of all training samples based on the local training samples. The classification model is trained.

Optionally, the sending module 11 is further configured to: if it is determined that the classification model with the third parameter meets the set condition, notify the plurality of training devices to stop the training of the classification model.

Optionally, the receiving module 12 is further configured to: receive respective test indicators corresponding to the plurality of training devices, wherein the plurality of training devices respectively measure the parameters having the third parameter according to local test samples. The classification model is tested to obtain respective test indicators corresponding to the plurality of training devices. The processing module 13 is further configured to: determine whether the classification model with the third parameter meets the set condition according to the respective test indicators corresponding to the multiple training devices.

Wherein, optionally, the processing module 13 is specifically configured to: perform weighted average processing on the respective corresponding test indicators of the multiple training devices; if the weighted average processing result is greater than or equal to a set threshold, determine that the The classification model of the third parameter reaches the set condition.

Optionally, the processing module 13 is specifically configured to: determine the weights of the respective test indicators corresponding to the multiple training devices according to at least one of the following information:

The number of training samples corresponding to each of the plurality of training sample providers, the set level category corresponding to each of the plurality of training sample providers, and the test indicators corresponding to each of the plurality of training devices.

Optionally, the sending module 11 is further configured to: send data processing information to the multiple training devices, so that the multiple training devices perform data processing on local training samples according to the data processing information, the The data processing information includes labeling information corresponding to the training samples.

Optionally, the data processing information also includes at least one of the following information:

Data structuring processing information, which is used to perform set structuring processing on the data contained in each training sample;

Data filtering rule information, which is used to filter training samples that do not meet the set requirements.

Optionally, the multiple training devices correspond to multiple training sample providers. Wherein, optionally, the multiple training sample providers include multiple medical institutions, and the classification model is used to complete diagnosis-related classification.

The apparatus shown in FIG. 9 may execute the model training method provided in the embodiments shown in FIG. 2 to FIG. 4 . For the detailed execution process and technical effects, refer to the descriptions in the foregoing embodiments, which will not be repeated here.

In a possible design, the structure of the model training apparatus shown in FIG. 9 may be implemented as an electronic device. As shown in FIG. 10 , the electronic device may include: a first processor 21 and a first memory 22 . The first memory 22 stores executable codes, and when the executable codes are executed by the first processor 21, the first processor 21 can at least implement the above-mentioned provision in the embodiments shown in FIG. 2 to FIG. 4 . model training method.

Optionally, the electronic device may further include a first communication interface 23 for communicating with other devices.

In addition, an embodiment of the present invention provides a non-transitory machine-readable storage medium, where executable codes are stored on the non-transitory machine-readable storage medium, and when the executable codes are executed by a processor of an electronic device , so that the processor can at least implement the model training method provided in the foregoing embodiments shown in FIG. 2 to FIG. 4 .

In addition, an embodiment of the present invention provides an application program. When the application program is executed by a processor of an electronic device, the processor is caused to execute the model training method provided in the embodiments shown in the foregoing FIG. 2 to FIG. 4 . .

FIG. 11 is a schematic structural diagram of a model training apparatus according to an embodiment of the present invention, where the apparatus is applied to a target training device. As shown in FIG. 11 , the apparatus includes: a receiving module 31 , a training module 32 , a sending module 33 , and an updating module 34 .

The receiving module 31 is configured to receive the first parameter of the classification model sent by the control device, the control device is connected to multiple training devices, and the target training device is any one of the multiple training devices.

A training module 32, configured to train the classification model based on the first parameters and local training samples to obtain second parameters corresponding to the target training device.

A sending module 33, configured to send the second parameter corresponding to the target training device to the control device, so that the control device combines the second parameters corresponding to the multiple training devices to obtain the third parameter, wherein , the multiple training devices respectively train the classification model based on the first parameters and local training samples to obtain second parameters corresponding to the multiple training devices.

The updating module 34 is configured to update the parameter of the classification model to the third parameter in response to the notification from the control device.

Optionally, the apparatus further includes: an encryption module configured to encrypt the second parameter corresponding to the target training device. At this time, the sending module 33 is specifically configured to: send the encrypted second parameter corresponding to the target training device to the control device.

Optionally, the training module 32 is further configured to: test the classification model with the third parameter according to a local test sample to obtain a test index corresponding to the target training device. The sending module 33 is further configured to: send the test index corresponding to the target training device to the control device, so that the control device determines, according to the test index corresponding to each of the plurality of training devices, that the third device has the third parameter. Whether the classification model of the parameter meets the set condition, wherein the plurality of training devices respectively test the classification model with the third parameter according to local test samples to obtain the corresponding corresponding test indicators.

Optionally, the receiving module 31 is further configured to: receive the data processing information sent by the control device. The device further includes: a data processing module configured to perform data processing on the local training samples according to the data processing information, where the data processing information includes label information corresponding to the training samples.

Optionally, the data processing information also includes at least one of the following information:

Data structuring processing information, which is used to perform set structuring processing on the data contained in each training sample;

Data filtering rule information, which is used to filter training samples that do not meet the set requirements.

The apparatus shown in FIG. 11 can execute the model training method provided in the embodiment shown in FIG. 5. For the detailed execution process and technical effect, refer to the description in the foregoing embodiment, and details are not repeated here.

In a possible design, the structure of the model training apparatus shown in FIG. 11 may be implemented as an electronic device. As shown in FIG. 12 , the electronic device may include: a second processor 41 and a second memory 42 . The second memory 42 stores executable codes, and when the executable codes are executed by the second processor 41 , the second processor 41 can at least implement the model training method provided in the foregoing embodiment in FIG. 5 .

Optionally, the electronic device may further include a second communication interface 43 for communicating with other devices.

In addition, an embodiment of the present invention provides a non-transitory machine-readable storage medium, where executable codes are stored on the non-transitory machine-readable storage medium, and when the executable codes are executed by a processor of an electronic device , so that the processor can at least implement the model training method provided in the foregoing embodiment in FIG. 5 .

In addition, an embodiment of the present invention provides an application program, and when the application program is executed by a processor of an electronic device, the processor is caused to execute the model training method provided in the embodiment shown in the foregoing FIG. 5 .

FIG. 13 is a schematic structural diagram of an apparatus for training a diagnosis-related classification model according to an embodiment of the present invention, where the apparatus is applied to a control device. As shown in FIG. 13 , the apparatus includes: a sending module 51 , a receiving module 52 , and a combining module 53 .

The sending module 51 is configured to send the first parameter of the classification model to multiple training devices, so that the multiple training devices respectively train the classification model based on the first parameter and local training samples to obtain the The second parameter corresponding to each of the plurality of training devices, where the classification model is set in each of the plurality of training devices, and the plurality of training devices correspond to a plurality of medical institutions.

The receiving module 52 is configured to receive the respective second parameters corresponding to the multiple training devices.

The combining module 53 is configured to combine the respective second parameters corresponding to the multiple training devices to obtain the third parameters.

The sending module 51 is further configured to: notify the plurality of training devices to update the parameters of the classification model to the third parameters.

The apparatus shown in FIG. 13 can execute the diagnosis-related classification model training method provided in the embodiment shown in FIG. 6 . For the detailed execution process and technical effects, refer to the description in the foregoing embodiment, which will not be repeated here.

In a possible design, the structure of the diagnostic-related classification model training apparatus shown in FIG. 13 may be implemented as an electronic device. As shown in FIG. 14 , the electronic device may include: a third processor 61 and a third memory 62 . The third memory 62 stores executable codes, and when the executable codes are executed by the third processor 61, the third processor 61 can at least implement the diagnosis-related classification model provided in the foregoing embodiment of FIG. 6 . training method.

Optionally, the electronic device may further include a third communication interface 63 for communicating with other devices.

In addition, an embodiment of the present invention provides a non-transitory machine-readable storage medium, where executable codes are stored on the non-transitory machine-readable storage medium, and when the executable codes are executed by a processor of an electronic device , so that the processor can at least implement the diagnosis-related classification model training method provided in the embodiment of FIG. 6 above.

In addition, an embodiment of the present invention provides an application program. When the application program is executed by a processor of an electronic device, the processor is caused to execute the method for training a diagnosis-related classification model provided in the embodiment shown in FIG. 6 above. .

FIG. 15 is a schematic structural diagram of an apparatus for training a diagnosis-related classification model according to an embodiment of the present invention, where the apparatus is applied to a target training device. As shown in FIG. 15 , the apparatus includes: a receiving module 71 , a training module 72 , a sending module 73 , and an updating module 74 .

A receiving module 71, configured to receive the first parameter of the classification model sent by the control device, the control device is connected to multiple training devices, the target training device is any one of the multiple training devices, the multiple training devices The training equipment corresponds to a plurality of medical institutions.

A training module 72, configured to train the classification model based on the first parameters and local training samples to obtain second parameters corresponding to the target training device.

A sending module 73, configured to send the second parameter corresponding to the target training device to the control device, so that the control device combines the second parameters corresponding to the multiple training devices to obtain the third parameter, wherein , the multiple training devices respectively train the classification model based on the first parameters and local training samples to obtain second parameters corresponding to the multiple training devices.

The updating module 74 is configured to update the parameter of the classification model to the third parameter in response to the notification from the control device.

The apparatus shown in FIG. 15 can execute the diagnosis-related classification model training method provided in the embodiment shown in FIG. 7 . For the detailed execution process and technical effects, refer to the descriptions in the foregoing embodiments, which will not be repeated here.

In a possible design, the structure of the diagnostic-related classification model training apparatus shown in FIG. 15 may be implemented as an electronic device. As shown in FIG. 16 , the electronic device may include: a fourth processor 81 and a fourth memory 82 . The fourth memory 82 stores executable codes, and when the executable codes are executed by the fourth processor 81, the fourth processor 81 can at least implement the diagnosis-related classification model provided in the foregoing embodiment of FIG. 7 . training method.

Optionally, the electronic device may further include a fourth communication interface 83 for communicating with other devices.

In addition, an embodiment of the present invention provides a non-transitory machine-readable storage medium, where executable codes are stored on the non-transitory machine-readable storage medium, and when the executable codes are executed by a processor of an electronic device , so that the processor can at least implement the diagnosis-related classification model training method provided in the embodiment of FIG. 7 .

In addition, an embodiment of the present invention provides an application program. When the application program is executed by a processor of an electronic device, the processor is caused to execute the method for training a diagnosis-related classification model provided in the embodiment shown in FIG. 7 above. .

The apparatus embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separate. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution in this embodiment. Those of ordinary skill in the art can understand and implement it without creative effort.

From the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by adding a necessary general hardware platform, and certainly can also be implemented by combining hardware and software. Based on this understanding, the above-mentioned technical solutions can be embodied in the form of computer products in essence or that contribute to the prior art. In the form of a computer program product embodied on a medium (including but not limited to disk storage, CD-ROM, optical storage, etc.).

Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, but not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that it can still be The technical solutions described in the foregoing embodiments are modified, or some technical features thereof are equivalently replaced; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (36)

1. a model training method, is characterized in that, comprises: Sending the first parameter of the classification model to a plurality of training devices, so that the plurality of training devices respectively train the classification model based on the first parameter and the local training samples, so as to obtain the corresponding correspondence of the plurality of training devices The second parameter of the plurality of training devices is provided with the classification model; receiving second parameters corresponding to each of the plurality of training devices; combining the respective second parameters of the plurality of training devices to obtain the third parameter; Notifying the plurality of training devices to update the parameters of the classification model to the third parameters. 2. The method according to claim 1, wherein the combining the respective second parameters of the plurality of training devices to obtain the third parameter comprises: Perform a weighted average process on the second parameters corresponding to each of the plurality of training devices to obtain the third parameter. 3. The method according to claim 1, wherein the receiving second parameters corresponding to the multiple training devices comprises: receiving encrypted second parameters corresponding to each of the plurality of training devices; The combining the respective second parameters of the multiple training devices to obtain the third parameters, including: Send the encrypted second parameters corresponding to the plurality of training devices to the trusted execution space, so as to decrypt the second parameters corresponding to the plurality of training devices in the trusted execution space and perform the operation on the trusted execution space. The second parameters corresponding to each of the plurality of training devices are transformed; The transformed second parameters corresponding to each of the plurality of training devices are combined to obtain the third parameter. 4. The method according to claim 3, wherein the transformation processing comprises: differential privacy processing. 5. The method according to claim 1, wherein the method further comprises: If it is determined that the classification model with the third parameter meets the set condition, the plurality of training devices are notified to stop the training of the classification model. 6. The method of claim 1, wherein the method further comprises: If it is determined that the classification model with the third parameter does not meet the set condition, the plurality of training devices are notified to continue training the classification model based on local training samples. 7. The method according to claim 5 or 6, wherein the method further comprises: Receive test indicators corresponding to each of the plurality of training devices, wherein the plurality of training devices respectively test the classification model with the third parameter according to local test samples to obtain the respective test indicators of the plurality of training devices. Corresponding test indicators; Whether the classification model with the third parameter meets the set condition is determined according to the respective test indicators corresponding to the plurality of training devices. 8 . The method according to claim 7 , wherein the determining whether the classification model with the third parameter meets the set condition according to the test indicators corresponding to each of the plurality of training devices comprises: 9 . : Perform weighted average processing on the respective corresponding test indicators of the multiple training devices; If the weighted average processing result is greater than or equal to a set threshold, it is determined that the classification model with the third parameter meets the set condition. 9. The method according to claim 8, wherein the method further comprises: The weights of the respective test indicators corresponding to the multiple training devices are determined according to at least one of the following information: The number of training samples corresponding to each of the plurality of training sample providers, the set level category corresponding to each of the plurality of training sample providers, and the test indicators corresponding to each of the plurality of training devices. 10. The method of claim 1, wherein the method further comprises: Sending data processing information to the multiple training devices, so that the multiple training devices perform data processing on local training samples according to the data processing information, where the data processing information includes label information corresponding to the training samples. 11. The method according to claim 10, wherein the data processing information further comprises at least one of the following information: Data structuring processing information, which is used to perform set structuring processing on the data contained in each training sample; Data filtering rule information, which is used to filter training samples that do not meet the set requirements. 12. The method of claim 1, wherein the plurality of training devices correspond to a plurality of training sample providers. 13. The method according to claim 11, wherein the plurality of training sample providers comprises a plurality of medical institutions, and the classification model is used to complete diagnosis-related classification. 14. A model training method, characterized in that, applied to a target training device, the method comprising: receiving a first parameter of the classification model sent by a control device, the control device is connected to multiple training devices, and the target training device is any one of the multiple training devices; training the classification model based on the first parameter and local training samples to obtain the second parameter corresponding to the target training device; sending the second parameter corresponding to the target training device to the control device, so that the control device combines the second parameters corresponding to the multiple training devices to obtain the third parameter, wherein the multiple training devices The device separately trains the classification model based on the first parameter and local training samples to obtain second parameters corresponding to each of the plurality of training devices; In response to the notification from the control device, the parameter of the classification model is updated to the third parameter. 15. The method according to claim 14, wherein the sending the second parameter corresponding to the target training device to the control device comprises: encrypting the second parameter corresponding to the target training device; Send the encrypted second parameter corresponding to the target training device to the control device. 16. The method of claim 14, wherein the method further comprises: Test the classification model with the third parameter according to the local test sample to obtain the test index corresponding to the target training device; Send the test index corresponding to the target training device to the control device, so that the control device determines whether the classification model with the third parameter conforms to the design according to the test index corresponding to each of the multiple training devices. set conditions, wherein the plurality of training devices respectively test the classification model with the third parameter according to local test samples to obtain respective test indicators corresponding to the plurality of training devices. 17. The method of claim 14, wherein the method further comprises: Receive data processing information sent by the control device; Data processing is performed on the local training samples according to the data processing information, where the data processing information includes label information corresponding to the training samples. 18. The method according to claim 17, wherein the data processing information further comprises at least one of the following information: Data structuring processing information, which is used to perform set structuring processing on the data contained in each training sample; Data filtering rule information, which is used to filter training samples that do not meet the set requirements. 19. A method for training a diagnosis-related classification model, comprising: Sending the first parameter of the classification model to a plurality of training devices, so that the plurality of training devices respectively train the classification model based on the first parameter and the local training samples, so as to obtain the corresponding correspondence of the plurality of training devices The second parameter of , the classification model is provided in the multiple training devices, and the multiple training devices correspond to multiple medical institutions; receiving second parameters corresponding to each of the plurality of training devices; combining the respective second parameters of the plurality of training devices to obtain the third parameter; Notifying the plurality of training devices to update the parameters of the classification model to the third parameters. 20. A method for training a diagnosis-related classification model, characterized in that, applied to target training equipment, the method comprising: Receive a first parameter of the classification model sent by a control device, the control device is connected to a plurality of training devices, the target training device is any one of the plurality of training devices, and the plurality of training devices are connected to a plurality of medical devices. institutional correspondence; training the classification model based on the first parameter and local training samples to obtain the second parameter corresponding to the target training device; sending the second parameter corresponding to the target training device to the control device, so that the control device combines the second parameters corresponding to the multiple training devices to obtain the third parameter, wherein the multiple training devices The device separately trains the classification model based on the first parameter and local training samples to obtain second parameters corresponding to each of the plurality of training devices; In response to the notification from the control device, the parameter of the classification model is updated to the third parameter. 21. A model training device, wherein the device comprises: The sending module is configured to send the first parameter of the classification model to multiple training devices, so that the multiple training devices respectively train the classification model based on the first parameter and local training samples to obtain the multiple training devices. second parameters corresponding to each of the training devices, the classification models are all provided in the plurality of training devices; a receiving module, configured to receive the respective second parameters corresponding to the multiple training devices; a processing module, configured to combine the respective second parameters of the plurality of training devices to obtain the third parameters; The sending module is further configured to: notify the plurality of training devices to update the parameters of the classification model to the third parameters. 22. An electronic device, comprising: a memory and a processor; wherein, executable code is stored on the memory, and when the executable code is executed by the processor, the processor is executed The model training method according to any one of claims 1 to 13. 23. A non-transitory machine-readable storage medium, wherein executable codes are stored on the non-transitory machine-readable storage medium, and when the executable codes are executed by a processor of an electronic device, The processor performs the model training method of any one of claims 1 to 13. 24. An application program, characterized in that, when the application program is executed by a processor of an electronic device, the processor is caused to execute the model training method according to any one of claims 1 to 13. 25. A model training device, characterized in that, applied to target training equipment, the device comprising: a receiving module, configured to receive a first parameter of the classification model sent by a control device, the control device is connected to a plurality of training devices, and the target training device is any one of the plurality of training devices; a training module, configured to train the classification model based on the first parameter and local training samples to obtain a second parameter corresponding to the target training device; a sending module, configured to send the second parameter corresponding to the target training device to the control device, so that the control device combines the second parameters corresponding to the multiple training devices to obtain the third parameter, wherein, The plurality of training devices respectively train the classification model based on the first parameters and local training samples to obtain second parameters corresponding to the plurality of training devices; An update module, configured to update the parameter of the classification model to the third parameter in response to the notification from the control device. 26. An electronic device, comprising: a memory and a processor; wherein, executable code is stored on the memory, and when the executable code is executed by the processor, the processor is caused to execute The model training method according to any one of claims 14 to 18. 27. A non-transitory machine-readable storage medium, wherein executable codes are stored on the non-transitory machine-readable storage medium, and when the executable codes are executed by a processor of an electronic device, The processor performs the model training method of any one of claims 14 to 187. 28. An application program, characterized in that, when the application program is executed by a processor of an electronic device, the processor is caused to execute the model training method according to any one of claims 14 to 18. 29. An apparatus for training a diagnosis-related classification model, wherein the apparatus comprises: The sending module is configured to send the first parameter of the classification model to multiple training devices, so that the multiple training devices respectively train the classification model based on the first parameter and local training samples to obtain the multiple training devices. second parameters corresponding to each of the training devices, the classification models are all provided in the plurality of training devices, and the plurality of training devices correspond to a plurality of medical institutions; a receiving module, configured to receive the respective second parameters corresponding to the multiple training devices; a merging module for merging the respective second parameters of the plurality of training devices to obtain the third parameter; The sending module is further configured to: notify the plurality of training devices to update the parameters of the classification model to the third parameters. 30. An electronic device, comprising: a memory and a processor; wherein executable code is stored on the memory, and when the executable code is executed by the processor, the processor is made to execute The diagnostic-related classification model training method according to claim 19 . 31. A non-transitory machine-readable storage medium, wherein executable codes are stored on the non-transitory machine-readable storage medium, and when the executable codes are executed by a processor of an electronic device, The processor executes the diagnostic-related classification model training method of claim 19 . 32. An application program, wherein when the application program is executed by a processor of an electronic device, the processor is caused to execute the method for training a diagnosis-related classification model according to claim 19. 33. A diagnostic-related classification model training device, characterized in that, when applied to target training equipment, the device comprises: a receiving module, configured to receive the first parameter of the classification model sent by a control device, the control device is connected to multiple training devices, the target training device is any one of the multiple training devices, the multiple training devices The equipment corresponds to multiple medical institutions; a training module, configured to train the classification model based on the first parameter and local training samples to obtain a second parameter corresponding to the target training device; a sending module, configured to send the second parameter corresponding to the target training device to the control device, so that the control device combines the second parameters corresponding to the multiple training devices to obtain the third parameter, wherein, The plurality of training devices respectively train the classification model based on the first parameters and local training samples to obtain second parameters corresponding to the plurality of training devices; An update module, configured to update the parameter of the classification model to the third parameter in response to the notification from the control device. 34. An electronic device, comprising: a memory and a processor; wherein, executable code is stored on the memory, and when the executable code is executed by the processor, the processor is executed The diagnosis-related classification model training method according to claim 20. 35. A non-transitory machine-readable storage medium, wherein executable codes are stored on the non-transitory machine-readable storage medium, and when the executable codes are executed by a processor of an electronic device, The processor executes the diagnostic-related classification model training method of claim 20 . 36. An application program, wherein when the application program is executed by a processor of an electronic device, the processor is caused to execute the method for training a diagnosis-related classification model according to claim 20.

Images