CN107766940A - Method and apparatus for generation model - Google Patents

Abstract

The present application discloses methods and apparatus for generating models. A specific implementation of the method includes: in response to receiving a model generation request including a user ID sent by a user terminal, searching for a model information set corresponding to the user ID from a preset model table, and sending the model information set to the terminal , wherein the model information in the model information set includes model categories and model parameters, and the model table is used to represent the correspondence between user identification and model information; in response to receiving the model category and model parameters selected by the user from the model information set sent by the terminal , determining a neural network matching the model category and model parameters selected by the user; in response to receiving the sample data set sent by the terminal, using a machine learning method to obtain a model based on the sample data set and neural network training. This embodiment can generate a user-defined neural network model.

Description

Method and apparatus for generating a model

technical field

The embodiments of the present application relate to the field of computer technology, specifically to the field of Internet technology, and in particular to a method and device for generating a model.

Background technique

Artificial Intelligence (Artificial Intelligence), the English abbreviation is AI. It is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. Artificial intelligence is a branch of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a manner similar to human intelligence. Research in this field includes robotics, language recognition, image recognition, natural language processing and expert systems, etc.

With the development of deep learning, the existing AI technology can often achieve better results in general fields, such as: general text recognition, limited set image classification, general word segmentation, and general speech recognition. However, scenarios that require AI capabilities in practical applications are often customized, such as text recognition for specific documents, word segmentation for medical terms, and speech recognition for special scenarios.

Contents of the invention

The embodiments of the present application propose a method and an apparatus for generating a model.

In the first aspect, the embodiment of the present application provides a method for generating a model, including: in response to receiving a model generation request including a user ID sent by a user terminal, searching for a model corresponding to the user ID from a preset model table. model information set, and send the model information set to the terminal, wherein the model information in the model information set includes model categories and model parameters, and the model table is used to represent the corresponding relationship between user identification and model information; in response to receiving the The model category and model parameters selected by the user from the model information set determine the neural network that matches the model category and model parameters selected by the user; in response to receiving the sample data set sent by the terminal, using machine learning methods, based on the sample data set and neural network The network is trained to obtain a model.

In some embodiments, the sample data in the sample data set includes input sample data and output sample data; and the method further includes: obtaining a predetermined number of input sample data and corresponding to the predetermined number of input sample data from the sample data set A predetermined number of output sample data; for each piece of input sample data obtained, input the input sample data into the model to obtain an output result corresponding to the input sample data, if the output result corresponds to the output sample data corresponding to the input sample data If the similarity of the data is greater than the predetermined similarity threshold, the number of correct verification times is accumulated; the ratio of the correct number of verification times to the predetermined number is determined as the accuracy rate.

In some embodiments, the method further includes: if the accuracy rate is greater than a predetermined accuracy rate threshold, converting the model into an application, and publishing the application to a target server for at least one terminal to download and use.

In some embodiments, the method further includes: receiving feedback data sent by the terminal downloading and using the application and adding the feedback data to the sample data set, wherein the feedback data includes actual input data input into the application by the terminal downloading and using the application, The actual output result corresponding to the actual input data, the expected output result input by the user who downloads and uses the terminal of the application, and the similarity between the actual output result and the expected output result is less than a predetermined similarity threshold; based on the actual input data, the actual output result, and the expected output As a result, the model is retrained and the updated model is republished to the target server.

In some embodiments, the method further includes: if the accuracy rate is less than or equal to the predetermined accuracy rate threshold, then determine the recommended model information from the model information set according to the accuracy rate and the model category and model parameters selected by the user, and send the recommended model information to the terminal Model information for users to reselect model category and model parameters.

In some embodiments, the model is obtained based on the sample data set and neural network training, including: selecting sample data from the sample data set to perform the following training steps: using the sample data to train the neural network; based on the training results, adjusting the network parameters of the neural network; Determine whether the training completion condition has been satisfied; in response to determining that it is not satisfied and the stop training message sent by the user's terminal is not received, select other sample data from the sample data set to continue the training step; wherein the training completion condition includes at least one of the following : The number of training times of the neural network reaches the preset threshold of training times; in two adjacent training sessions, the loss value between the outputs of the neural network is less than the preset threshold.

In a second aspect, the embodiment of the present application provides an apparatus for generating a model, including: a receiving unit configured to, in response to receiving a model generation request including a user identification sent by a user terminal, select a model from a preset model table Find the model information set corresponding to the user ID, and send the model information set to the terminal, wherein the model information in the model information set includes model categories and model parameters, and the model table is used to represent the corresponding relationship between the user ID and the model information; determine The unit is configured to determine a neural network that matches the model category and model parameters selected by the user in response to receiving the model category and model parameters selected by the user from the model information set sent by the terminal; the training unit is configured to respond to the received The sample data set sent by the terminal uses a machine learning device to obtain a model based on the sample data set and neural network training.

In some embodiments, the sample data in the sample data set includes input sample data and output sample data; and the device further includes a verification unit configured to: acquire a predetermined number of input sample data and a predetermined number of input sample data from the sample data set A predetermined number of output sample data corresponding to the input sample data; for each piece of input sample data obtained, input the input sample data into the model to obtain the output result corresponding to the input sample data, if the output result is consistent with the input If the similarity of the output sample data corresponding to the sample data is greater than the predetermined similarity threshold, the number of correct verification times is accumulated; the ratio of the correct number of verification times to the predetermined number is determined as the accuracy rate.

In some embodiments, the device further includes a publishing unit configured to: if the accuracy rate is greater than a predetermined accuracy rate threshold, convert the model into an application, and publish the application to a target server for at least one terminal to download and use.

In some embodiments, the device further includes a feedback unit configured to: receive feedback data sent by the terminal that downloads and uses the application and add the feedback data to the sample data set, wherein the feedback data includes input to the application by the terminal that downloads and uses the application The actual input data, the actual output result corresponding to the actual input data, the expected output result input by the user who downloads and uses the terminal of the application, and the similarity between the actual output result and the expected output result is less than a predetermined similarity threshold; based on the actual input data, Actual output results, expected output results retrain the model, and republish the updated model to the target server.

In some embodiments, the device further includes a recommendation unit configured to: if the accuracy rate is less than or equal to a predetermined accuracy rate threshold, determine the recommended model information from the model information set according to the accuracy rate, the model category and model parameters selected by the user , and send the recommended model information to the terminal for the user to reselect the model category and model parameters.

In some embodiments, the training unit is further used to: select sample data from the sample data set to perform the following training steps: use the sample data to train the neural network; adjust the network parameters of the neural network based on the training results; determine whether the training completion condition has been met ; In response to determining that it is not satisfied and does not receive the stop training message sent by the user's terminal, select other sample data from the sample data set to continue the training step; wherein, the training completion condition includes at least one of the following: the number of training times of the neural network reaches The preset threshold of training times; in two adjacent training sessions, the loss value between the outputs of the neural network is less than the preset threshold.

In a third aspect, the embodiment of the present application provides a server, including: one or more processors; a storage device for storing one or more programs, when one or more programs are executed by one or more processors, One or more processors are caused to implement the method according to any one of the first aspect.

In a fourth aspect, the embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored, wherein, when the program is executed by a processor, the method according to any one of the first aspect is implemented.

The method and device for generating a model provided in the embodiments of the present application trains a user-customized model based on the model category and model parameters of the neural network selected by the user and the training sample data uploaded by the user. In this way, the customized data is effectively utilized to generate a customized model.

Description of drawings

Other characteristics, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

FIG. 1 is an exemplary system architecture diagram to which the present application can be applied;

FIG. 2 is a flowchart of one embodiment of a method for generating a model according to the present application;

FIG. 3 is a schematic diagram of an application scenario of a method for generating a model according to the present application;

Fig. 4 is a flowchart of another embodiment of the method for generating a model according to the present application;

Fig. 5 is a schematic structural diagram of an embodiment of a device for generating a model according to the present application;

Fig. 6 is a schematic structural diagram of a computer system suitable for implementing a terminal device or a server according to an embodiment of the present application.

Detailed ways

The application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described here are only used to explain related inventions, rather than to limit the invention. It should also be noted that, for the convenience of description, only the parts related to the related invention are shown in the drawings.

It should be noted that, in the case of no conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and embodiments.

FIG. 1 shows an exemplary system architecture 100 to which an embodiment of the method for generating a model or the apparatus for generating a model of the present application can be applied.

As shown in FIG. 1 , a system architecture 100 may include terminal devices 101 , 102 , 103 , a network 104 and a server 105 . The network 104 is used as a medium for providing communication links between the terminal devices 101 , 102 , 103 and the server 105 . Network 104 may include various connection types, such as wires, wireless communication links, or fiber optic cables, among others.

Users can use terminal devices 101 , 102 , 103 to interact with server 105 via network 104 to receive or send messages and the like. Various communication client applications can be installed on the terminal devices 101, 102, 103, such as model generation applications, web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, etc.

The terminal devices 101, 102, and 103 can be various electronic devices that have display screens and support the selection of model training parameters and the uploading of sample data used in model training, including but not limited to smartphones, tablet computers, e-book readers, MP3 Players (Moving Picture Experts Group Audio Layer III, moving picture experts compressed standard audio layer 3), MP4 (Moving Picture Experts Group Audio Layer IV, moving picture experts compressed standard audio layer 4) players, laptops and desktops computer and so on.

The server 105 may be a server that provides various services, for example, a background model generation server that provides support for model information displayed on the terminal devices 101 , 102 , 103 . The background model generation server can analyze and process the received model generation request and sample data to generate a model.

It should be noted that the method for generating a model provided in the embodiment of the present application is generally executed by the server 105 , and correspondingly, the device for generating a model is generally disposed in the server 105 .

It should be understood that the numbers of terminal devices, networks and servers in Fig. 1 are only illustrative. According to the implementation needs, there can be any number of terminal devices, networks and servers. In an optional implementation, the method for generating the model can be performed on a server or a server cluster composed of multiple servers, and the trained neural network model can be run on a server, PC, mobile terminal, vehicle-mounted terminal, etc. on various types of electronic equipment.

Continuing to refer to FIG. 2 , a flow 200 of an embodiment of the method for generating a model according to the present application is shown. The method for generating a model includes the following steps:

Step 201, in response to receiving a model generation request including a user ID from a user terminal, look up a model information set corresponding to the user ID from a preset model table, and send the model information set to the terminal.

In this embodiment, the electronic device on which the method for generating a model runs (such as the server shown in FIG. 1 ) can receive a model generation request from a user terminal through a wired connection or a wireless connection, and then from a preset Look up the model information set corresponding to the user ID in the model table, and send the model information set to the terminal for the user to select the model category and model parameters, wherein, the model information in the model information set includes the model category and model parameters, and the model table uses It is used to represent the corresponding relationship between user identification and model information. A model class can be, for example, a convolutional neural network, a recurrent neural network, etc. Model parameters may include, but are not limited to, the number of network layers, kernel function type, error precision, learning rate, etc. The above-mentioned electronic device can set model information for each user, and can also divide users into different rights, and then associate the model table with user rights. The server can provide users with different permissions with different model categories and model parameters for users to choose. Different levels of permissions can be set for users in advance according to the user ID. For example, there are two levels of permissions. Users with high-level permissions can choose neural networks with layers above 10, while users with low-level permissions can only choose neural networks with layers below 10. .

Step 202, in response to receiving the model category and model parameters selected by the user from the model information set sent by the terminal, determine the neural network matching the model category and model parameters selected by the user.

In this embodiment, after receiving the model information set, the terminal displays the model information set for the user to select a model category and model parameters. Users can select model parameters from the model information list, or manually input model parameters. The user can also input the name of the model to be generated through the terminal, which is user-defined, and the server can verify the name to avoid duplication with other user's model names. In addition, if the user inputs model parameters manually, the server also needs to verify the parameters input by the user. The model category in the model information set is associated with the model parameters. After the user selects the model category, the optional model parameters are related to the selected model category. For example, if the user selects a convolutional neural network, the number of network layers can be selected from 3-10 layers. And if the feedforward neural network is selected, the number of network layers can be selected from 3-5 layers. The server determines from the candidate neural network set the neural network that matches the model category and model parameters selected by the user sent by the terminal. The neural network in the candidate neural network set is a mathematical model that uses a structure similar to that of the brain's synaptic connections for information processing. In engineering and academia, it is often referred to directly as "neural network" or neural network.

Optionally, the user can also input the name of the model to be generated through the terminal, which is user-defined, and the server can verify the name to avoid duplication with other user's model names.

Step 203, in response to receiving the sample data set sent by the terminal, use a machine learning method to obtain a model based on the sample data set and neural network training.

In this embodiment, the terminal may send the sample data set while sending the model category and model parameters. It is also possible to wait until the server verifies the model category and model parameters, and then sends a message to the terminal to prompt the user to send the sample data set after the neural network is determined. In this way, it is avoided to still upload the sample data set when the specified model category and model parameters cannot be generated for the user, resulting in a waste of network traffic. Taking the neural network used to identify roads in remote sensing images and the sample data set of remote sensing images as an example, the training process of the model is illustrated. The remote sensing image sample data set includes the original remote sensing image and the road information in the original remote sensing image. The electronic device can input the original remote sensing image into the neural network to obtain the output result. If the similarity between the output result and the road information in the original remote sensing image is less than the predetermined similarity threshold, adjust the network parameters and re-input the original remote sensing image into the neural network. And compare the output result with the road information in the original remote sensing image. Constantly adjust the network parameters until the similarity between the output result and the road information in the original remote sensing image is greater than the predetermined similarity threshold. Then select other original remote sensing images for training until the similarity between the output results of a predetermined number of original remote sensing images and the road information in the original remote sensing images is greater than a predetermined similarity threshold, and the model is trained. The electronic device training can be an initial neural network, and the initial neural network can be an untrained neural network or an untrained neural network. Each layer of the initial neural network can be set with initial parameters, and the parameters can be set during the training process of the neural network. are constantly adjusted. The initial neural network may be various types of untrained or untrained artificial neural networks or a model obtained by combining multiple untrained or untrained artificial neural networks. In this way, the electronic device can input the original remote sensing image from the input side of the initial neural network, process the parameters of each layer in the initial neural network in turn, and output it from the output side of the initial neural network, and the information output from the output side is the original Road information in remote sensing images.

In some optional implementations of this embodiment, the training of the model based on the sample data set and the neural network includes: performing the following training steps for each sample data in the sample data set: using the sample data to train the neural network, based on the training As a result, the network parameters of the neural network are adjusted; it is determined whether the training completion condition has been satisfied; in response to the determination that it is not satisfied and the stop training message sent by the user's terminal is not received, other sample data is selected from the sample data set to continue the training step. Setting a training completion condition avoids training the neural network in an infinite loop. The training completion condition may include but not limited to at least one of the following: the number of training times for training the neural network reaches a preset threshold for training times; in two adjacent training sessions, the loss value between the outputs of the neural network is less than the preset threshold. The training result refers to the output result after inputting the sample data into the neural network. Taking the neural network used to identify roads in remote sensing images and the sample data set of remote sensing images as an example, an optional training process of the model is illustrated. Among them, the remote sensing image sample data set includes the original remote sensing image and road information in the original remote sensing image, and the road features may include road color, texture, height, temperature, shadow, direction change and other characteristics. Select part of the original remote sensing image from the remote sensing image sample data set and input it into the neural network; extract the road features of the input remote sensing image through the neural network; determine the loss of road feature extraction at least according to the difference between the extracted road features and the road information in the original remote sensing image value; adjust the network parameters of the neural network according to the loss value. Before starting training, the neural network kernel parameters are initialized with some different small random numbers. "Small random number" is used to ensure that the network will not enter a saturated state due to excessive kernel parameter values, resulting in training failure; "different" is used to ensure that the network can learn normally. In fact, if the same number is used to initialize the kernel parameters, the network will not be able to learn. By comparing the results of network training with the real class mark, the error is corrected. The kernel parameters are continuously optimized by adjusting the kernel parameters to minimize the error.

Optionally, during the training process, progress information is output to the user's terminal, for the user to refer to the progress information to decide whether to end the training process in advance. If the user wants to stop the training, a stop training message is sent through the terminal.

Continue to refer to FIG. 3 , which is a schematic diagram of an application scenario of the method for generating a model according to this embodiment. In the application scenario of FIG. 3 , the user sends a model generation request including the user ID to the server through the terminal 300 , and the server finds a model information set matching the user ID in the model table and sends it to the terminal 300 . The terminal 300 displays a set of model information for the user to input a user-defined model name 301 and select a model category 302 and network layer number 303 . The terminal sends the user-defined model name 301, selected model category 302 and network layer number 303 to the server. The server determines the neural network according to the model type 302 and the number of network layers 303 . The model is then obtained based on the sample data set uploaded by the terminal and neural network training.

The method provided by the above-mentioned embodiments of the present application can generate a customized model without writing codes by uploading a customized sample data set by the user.

Further referring to FIG. 4 , it shows a flow 400 of still another embodiment of a method for generating a model. The flow 400 of the method for generating a model includes the following steps:

Step 401, in response to receiving a model generation request including a user ID sent by a user terminal, look up a model information set corresponding to the user ID from a preset model table, and send the model information set to the terminal.

Step 402, in response to receiving the model category and model parameters selected by the user from the model information set sent by the terminal, determine the neural network matching the model category and model parameters selected by the user.

Step 403, in response to receiving the sample data set sent by the terminal, use a machine learning method to obtain a model based on the sample data set and neural network training.

Steps 401-403 are basically the same as steps 201-203, so they are not repeated here.

Step 404, acquiring a predetermined number of input sample data and a predetermined number of output sample data corresponding to the predetermined number of input sample data from the sample data set.

In this embodiment, after the model training is completed, the model may continue to be evaluated using the sample data set. For example, for supervised learning, the sample data in the sample data set includes input sample data and output sample data. When training the model, 10,000 pairs of input sample data and output sample data are used, and 100 pairs of input sample data and output sample data can be selected from the 10,000 pairs of input sample data and output sample data when evaluating the model.

Step 405, for each piece of input sample data obtained, input the input sample data into the model to obtain an output result corresponding to the input sample data, if the similarity between the output result and the output sample data corresponding to the input sample data is greater than the predetermined similarity threshold, the number of correct verifications is accumulated.

In this embodiment, in order to evaluate the effect of the generated model, some sample data may be selected from the sample data set and input to the model generated in step 203 to obtain an output result. Compare the output result with the output sample to determine the similarity between them. The calculation method of the similarity can adopt methods such as general cosine similarity and Euclidean distance. If the similarity between the output result and the output sample data corresponding to the input sample data is greater than a predetermined similarity threshold, it is considered that the verification is passed. A predetermined number of verifications are performed using a predetermined number of input sample data, and if the verification is passed after each verification, the number of correct verification times is accumulated.

Step 406, determine the ratio of the number of correct verification times to the predetermined number as the accuracy rate.

In this embodiment, step 405 is repeatedly executed a predetermined number of times, and the number of correct verification times is accumulated. The ratio of the correct number of verifications to the predetermined number is taken as the accuracy rate. For example, 100 validations are performed, and if the validation is correct 90 times, the accuracy is 90%.

Step 407, if the accuracy rate is greater than the predetermined accuracy rate threshold, convert the model into an application, and publish the application to the target server for at least one terminal to download and use.

In this embodiment, if the accuracy rate is greater than the predetermined accuracy rate threshold, it is considered that the model has passed the evaluation and can be used by other users. RESTful (REST (Representational State Transfer) refers to a set of architectural constraints and principles. Applications or designs that meet these constraints and principles are RESTful) architectures that convert models into applications and serve them through containers Publish to the target server. In the REST name "Representation Layer State Transformation", the subject is omitted. The "presentation layer" actually refers to the "representation layer" of "Resources". The so-called "resource" is an entity on the network, or a specific piece of information on the network. A resource is an interesting conceptual entity that is exposed to clients. Examples of resources are: application objects, database records, algorithms, and so on. It can be pointed to by a URI (UniversalResource Identifier, Uniform Resource Locator), and each resource corresponds to a specific URI. To obtain this resource, just visit its URI, so the URI becomes the address or unique identifier of each resource. All resources share a uniform interface for transferring state between client and server. Container Service provides high-performance and scalable container application management services, supports life cycle management of containerized applications, provides multiple application publishing methods and continuous delivery capabilities, and supports microservice architecture. Container Service simplifies the construction of container management clusters, integrates cloud virtualization, storage, network and security capabilities, and creates the best container operating environment in the cloud.

Optionally, for models obtained through unsupervised learning, they can be released directly without evaluation.

In some optional implementations of this embodiment, the feedback data sent by the terminal downloading and using the application is received and added to the sample data set, wherein the feedback data includes the actual input entered into the application by the terminal downloading and using the application data, the actual output results corresponding to the actual input data, the expected output results input by the user who downloads and uses the terminal of the application, and the similarity between the actual output results and the expected output results is less than a predetermined similarity threshold; based on the actual input data, the actual output results, Expect output to retrain the model and republish the updated model to the target server. Wherein, the expected output result is the correct output result for the actual input data input by the user using the model. When the user judges manually or by machine that the actual output result corresponding to the actual input data does not meet the expected output result, the user feeds back the expected output result together with the actual input data and the actual output result corresponding to the actual input data to the server. The server receives feedback data and adds it to the sample data set, manages the version of the data set, retrains the model, and supports continuous integration of models and services.

Step 408, if the accuracy rate is less than or equal to the predetermined accuracy rate threshold, determine the recommended model information from the model information set according to the accuracy rate, the model category and model parameters selected by the user, and send the recommended model information to the terminal for the user to reselect Model classes and model parameters.

In this embodiment, if the accuracy rate is less than or equal to the predetermined accuracy rate threshold, it means that the model evaluation fails, and therefore the user is recommended to retrain the model. According to the difference between the accuracy rate and the predetermined accuracy rate threshold, the adjusted model information can be determined as the recommended model information, for example, whether model parameters or model types need to be adjusted. And the adjustment amount can be determined according to the characteristics of different model types. For example, if the preset accuracy rate threshold is 90%, and the actual test accuracy rate is 89% when the current number of network layers is 3, it may be recommended to adjust the number of network layers to 5. If the actual test accuracy rate is 20% when the current number of network layers is 3, the effect of improving the accuracy rate by only changing the number of network layers may not be good, so it is recommended to adjust the type of the model to other networks, such as feedforward The neural network is tuned to a convolutional neural network.

It can be seen from FIG. 4 that, compared with the embodiment corresponding to FIG. 2 , the process 400 of the method for generating a model in this embodiment highlights the step of expanding sample data. Therefore, the solution described in this embodiment can introduce more sample data, thereby improving the speed and accuracy of model generation.

Further referring to FIG. 5 , as an implementation of the methods shown in the above figures, the present application provides an embodiment of a device for generating a model, which corresponds to the method embodiment shown in FIG. 2 , the The device can be specifically applied to various electronic devices.

As shown in FIG. 5 , the apparatus 500 for generating a model in this embodiment includes: a receiving unit 501 , a determining unit 502 and a training unit 503 . Wherein, the receiving unit 501 is configured to, in response to receiving a model generation request including a user ID sent by the user's terminal, search for a model information set corresponding to the user ID from a preset model table, and send the model information set to the terminal , wherein the model information in the model information set includes model categories and model parameters, and the model table is used to characterize the correspondence between user identifiers and model information; the determining unit 502 is configured to respond to receiving the user's selection from the model information set sent by the terminal The model category and model parameters, determine the neural network that matches the model category and model parameters selected by the user; the training unit 503 is configured to respond to receiving the sample data set sent by the terminal, using a machine learning device, based on the sample data set and the neural network The network is trained to obtain a model.

In this embodiment, for the specific processing of the receiving unit 501 , the determining unit 502 and the training unit 503 of the apparatus 500 for generating a model, reference may be made to steps 201 , 202 , and 203 in the corresponding embodiment in FIG. 2 .

In some optional implementations of this embodiment, the sample data in the sample data set includes input sample data and output sample data; and the device 500 further includes a verification unit configured to: acquire a predetermined number of The input sample data and the predetermined number of output sample data corresponding to the predetermined number of input sample data; for each piece of input sample data obtained, the input sample data is input to the model, and the output result corresponding to the input sample data is obtained , if the similarity between the output result and the output sample data corresponding to the input sample data is greater than a predetermined similarity threshold, then accumulating the number of correct verification times; determining the ratio of the correct number of verification times to the predetermined number as the accuracy rate.

In some optional implementations of this embodiment, the device 500 further includes a publishing unit configured to: if the accuracy rate is greater than a predetermined accuracy rate threshold, convert the model into an application, and publish the application to the target server for at least A terminal download is used.

In some optional implementations of this embodiment, the device 500 further includes a feedback unit configured to: receive feedback data sent by a terminal that downloads and uses an application and add the feedback data to the sample data set, wherein the feedback data includes The actual input data input into the application by the terminal using the application, the actual output result corresponding to the actual input data, the expected output result input by the user who downloads the terminal using the application, and the similarity between the actual output result and the expected output result is less than the predetermined similarity Threshold; retrain the model based on actual input data, actual output, and expected output, and republish the updated model to the target server.

In some optional implementations of this embodiment, the device 500 further includes a recommending unit configured to: if the accuracy rate is less than or equal to a predetermined accuracy rate threshold, select the model from the model according to the accuracy rate and the model category and model parameters selected by the user. The information collection determines the recommended model information, and sends the recommended model information to the terminal for the user to reselect the model category and model parameters.

In some optional implementations of this embodiment, the training unit 503 is further configured to: select sample data from the sample data set to perform the following training steps: use the sample data to train the neural network; based on the training results, adjust the network parameters of the neural network ; determine whether the training completion condition has been satisfied; in response to determining that it is not satisfied and the stop training message sent by the user's terminal is not received, select other sample data from the sample data set to continue the training step; wherein, the training completion condition includes at least one of the following Item: The number of training times of the neural network reaches the preset training number threshold; in two adjacent training sessions, the loss value between the outputs of the neural network is less than the preset threshold.

Referring now to FIG. 6 , it shows a schematic structural diagram of a computer system 600 suitable for implementing the server of the embodiment of the present application. The server shown in FIG. 6 is only an example, and should not limit the functions and scope of use of this embodiment of the present application.

As shown in FIG. 6 , a computer system 600 includes a central processing unit (CPU) 601 that can be programmed according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage section 608 into a random-access memory (RAM) 603 Instead, various appropriate actions and processes are performed. In the RAM 603, various programs and data necessary for the operation of the system 600 are also stored. The CPU 601 , ROM 602 , and RAM 603 are connected to each other via a bus 604 . An input/output (I/O) interface 605 is also connected to the bus 604 .

The following components are connected to the I/O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker; a storage section 608 including a hard disk, etc. and a communication section 609 including a network interface card such as a LAN card, a modem, or the like. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I/O interface 605 as needed. A removable medium 611 such as a magnetic disk, optical disk, magneto-optical disk, semiconductor memory, etc. is mounted on the drive 610 as necessary so that a computer program read therefrom is installed into the storage section 608 as necessary.

In particular, according to an embodiment of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, where the computer program includes program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program may be downloaded and installed from a network via communication portion 609 and/or installed from removable media 611 . When the computer program is executed by the central processing unit (CPU) 601, the above-mentioned functions defined in the method of the present application are performed. It should be noted that the computer-readable medium described in this application may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. A computer readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, electrical connections with one or more wires, portable computer diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable Programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, however, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, in which computer-readable program codes are carried. Such propagated data signals may take many forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the foregoing. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. . Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

Computer program code for carrying out the operations of this application may be written in one or more programming languages, or combinations thereof, including object-oriented programming languages—such as Java, Smalltalk, C++, and conventional A procedural programming language—such as "C" or a similar programming language. The program code may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In cases involving a remote computer, the remote computer can be connected to the user computer through any kind of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (such as through an Internet service provider). Internet connection).

The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present application. In this regard, each block in a flowchart or block diagram may represent a module, program segment, or portion of code that contains one or more logical functions for implementing specified executable instructions. It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or they may sometimes be executed in the reverse order, depending upon the functionality involved. It should also be noted that each block of the block diagrams and/or flowchart illustrations, and combinations of blocks in the block diagrams and/or flowchart illustrations, can be implemented by a dedicated hardware-based system that performs the specified functions or operations , or may be implemented by a combination of dedicated hardware and computer instructions.

The units involved in the embodiments described in the present application may be implemented by means of software or by means of hardware. The described units may also be set in a processor, for example, it may be described as: a processor includes a receiving unit, a determining unit, and a training unit. Wherein, the names of these units do not constitute a limitation on the unit itself under certain circumstances. For example, the receiving unit can also be described as "responsive to receiving the model generation request including the user identification sent by the user's terminal, from the preset A unit for searching the model information set corresponding to the user identifier in the set model table, and sending the model information set to the terminal”.

As another aspect, the present application also provides a computer-readable medium. The computer-readable medium may be included in the device described in the above embodiments, or it may exist independently without being assembled into the device. The above-mentioned computer-readable medium carries one or more programs, and when the one or more programs are executed by the device, the device: in response to receiving a model generation request sent by the user's terminal and including the user identification, from the preset Look up the model information set corresponding to the user ID in the model table, and send the model information set to the terminal, where the model information in the model information set includes model categories and model parameters, and the model table is used to represent the correspondence between the user ID and model information relationship; in response to receiving the model category and model parameters selected by the user from the model information set sent by the terminal, determine the neural network that matches the model category and model parameters selected by the user; in response to receiving the sample data set sent by the terminal, use the machine The learning method is based on the sample data set and neural network training to obtain the model.

The above description is only a preferred embodiment of the present application and an illustration of the applied technical principle. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to the technical solution formed by the specific combination of the above-mentioned technical features, but should also cover the technical solution formed by the above-mentioned technical features without departing from the inventive concept. Other technical solutions formed by any combination of or equivalent features thereof. For example, a technical solution formed by replacing the above-mentioned features with technical features with similar functions disclosed in (but not limited to) this application.

Claims (14)

1. A method for generating a model comprising: In response to receiving a model generation request including a user ID sent by the user's terminal, searching for a model information set corresponding to the user ID from a preset model table, and sending the model information set to the terminal, wherein, The model information in the model information set includes model categories and model parameters, and the model table is used to represent the correspondence between user identification and model information; In response to receiving the model category and model parameters selected by the user from the model information set sent by the terminal, determine a neural network that matches the model category and model parameters selected by the user; In response to receiving the sample data set sent by the terminal, a machine learning method is used to obtain a model based on the sample data set and the neural network training. 2. The method of claim 1, the sample data in the sample data set comprising input sample data and output sample data; and The method also includes: obtaining a predetermined number of input sample data and the predetermined number of output sample data corresponding to the predetermined number of input sample data from the sample data set; For each piece of input sample data obtained, input the input sample data into the model to obtain an output result corresponding to the input sample data, if the similarity between the output result and the output sample data corresponding to the input sample data is greater than Predetermined similarity threshold, then accumulate the correct number of verifications; The ratio of the number of correct verification times to the predetermined number is determined as the accuracy rate. 3. The method of claim 2, wherein the method further comprises: If the accuracy rate is greater than a predetermined accuracy rate threshold, the model is converted into an application, and the application is published to a target server for at least one terminal to download and use. 4. The method of claim 3, wherein the method further comprises: receiving feedback data sent by a terminal that downloads and uses the application and adding the feedback data to the sample data set, wherein the feedback data includes actual input data input into the application by the terminal that downloads and uses the application , the actual output result corresponding to the actual input data, the expected output result input by the user of the terminal who downloads and uses the application, and the similarity between the actual output result and the expected output result is less than the predetermined similarity threshold; Retraining the model based on the actual input data, the actual output result, and the expected output result, and republishing the updated model to the target server. 5. The method of claim 2, wherein the method further comprises: If the accuracy rate is less than or equal to the predetermined accuracy rate threshold, determine the recommended model information from the model information set according to the accuracy rate, the model category and model parameters selected by the user, and send the Recommended model information for the user to reselect model categories and model parameters. 6. The method according to any one of claims 1-5, wherein said model obtained based on said sample data set and said neural network training comprises: Select sample data from the sample data set to perform the following training steps: use the sample data to train the neural network; adjust the network parameters of the neural network based on the training results; determine whether the training completion condition has been met; Selecting other sample data from the sample data set to continue performing the training step in response to determining that it is not satisfied and the stop training message sent by the user's terminal is not received; Wherein, the training completion conditions include at least one of the following: The training times of the neural network reaches a preset training times threshold; In two adjacent training sessions, the loss value between the outputs of the neural network is smaller than a preset threshold. 7. An apparatus for generating a model comprising: The receiving unit is configured to, in response to receiving a model generation request including a user ID sent by a user terminal, search for a model information set corresponding to the user ID from a preset model table, and send the user ID to the terminal. A model information set, wherein the model information in the model information set includes model categories and model parameters, and the model table is used to represent the correspondence between user identification and model information; The determining unit is configured to, in response to receiving the model category and model parameters selected by the user from the model information set sent by the terminal, determine a neural network that matches the model category and model parameters selected by the user; The training unit is configured to, in response to receiving the sample data set sent by the terminal, use a machine learning device to obtain a model based on the sample data set and the neural network training. 8. The apparatus of claim 7, the sample data in the set of sample data comprising input sample data and output sample data; and The device also includes a verification unit configured to: obtaining a predetermined number of input sample data and the predetermined number of output sample data corresponding to the predetermined number of input sample data from the sample data set; For each piece of input sample data obtained, input the input sample data into the model to obtain an output result corresponding to the input sample data, if the similarity between the output result and the output sample data corresponding to the input sample data is greater than Predetermined similarity threshold, then accumulate the correct number of verifications; The ratio of the number of correct verification times to the predetermined number is determined as the accuracy rate. 9. The device according to claim 8, wherein the device further comprises a publishing unit configured to: If the accuracy rate is greater than a predetermined accuracy rate threshold, the model is converted into an application, and the application is published to a target server for at least one terminal to download and use. 10. The device according to claim 8, wherein the device further comprises a feedback unit configured to: receiving feedback data sent by a terminal that downloads and uses the application and adding the feedback data to the sample data set, wherein the feedback data includes actual input data input into the application by the terminal that downloads and uses the application , the actual output result corresponding to the actual input data, the expected output result input by the user of the terminal who downloads and uses the application, and the similarity between the actual output result and the expected output result is less than the predetermined similarity threshold; Retraining the model based on the actual input data, the actual output result, and the expected output result, and republishing the updated model to the target server. 11. The device according to claim 8, wherein the device further comprises a recommendation unit configured to: If the accuracy rate is less than or equal to the predetermined accuracy rate threshold, determine the recommended model information from the model information set according to the accuracy rate, the model category and model parameters selected by the user, and send the Recommended model information for the user to reselect model categories and model parameters. 12. The device according to any one of claims 7-11, wherein the training unit is further used for: Select sample data from the sample data set to perform the following training steps: use the sample data to train the neural network; adjust the network parameters of the neural network based on the training results; determine whether the training completion condition has been met; Selecting other sample data from the sample data set to continue performing the training step in response to determining that it is not satisfied and the stop training message sent by the user's terminal is not received; Wherein, the training completion conditions include at least one of the following: The training times of the neural network reaches a preset training times threshold; In two adjacent training sessions, the loss value between the outputs of the neural network is smaller than a preset threshold. 13. A server comprising: one or more processors; storage means for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors are made to implement the method according to any one of claims 1-6. 14. A computer-readable storage medium, on which a computer program is stored, wherein, when the program is executed by a processor, the method according to any one of claims 1-6 is implemented.