Model-based visual code-free neural network construction method

Abstract

The invention discloses a visual code-free neural network construction method based on a model. The method comprises the following steps: receiving input data; identifying the data type of the data; selecting a corresponding neural network model based on the data type; extracting a model structure matched with the data type from a preset model library; setting a corresponding number of reusable modules based on the model structure; and setting corresponding parameters for the reusable module. According to the method, the threshold of using the deep learning technology is reduced, the development efficiency and the flexibility of model construction are also improved, meanwhile, the error risk of manual operation is reduced, and the data processing efficiency is also improved.

Description

Technical Field

[0001] The present invention belongs to the field of computer system engineering, and particularly relates to a model-based visual no-code neural network construction method. Background Art

[0002] With the rapid development of artificial intelligence and deep learning technologies, neural networks have become an important tool for solving complex problems. The traditional neural network construction and training processes usually require professional programming skills and rich mathematical knowledge, which pose a great threshold for non-professionals. Although deep learning frameworks such as TensorFlow, Keras, and PyTorch provide powerful functions, users still need to have certain programming capabilities and technical backgrounds. To solve this problem, many researchers and developers have begun to explore "no-code" methods, enabling users to more conveniently construct and apply neural network models. Summary of the Invention

[0003] In view of the defects existing in the above-mentioned prior art, the present invention provides a model-based visual no-code neural network construction method, including the following steps:

[0004] Step S101: Receive input data;

[0005] Step S103: Identify the data type of the data;

[0006] Step S105: Select a corresponding neural network model based on the data type;

[0007] Step S107: Extract a model structure matching the data type from a preset model library;

[0008] Step S109: Set a corresponding number of reusable modules based on the model structure;

[0009] Step S1011: Set corresponding parameters for the reusable modules.

[0010] Among them, the data types in step S103 include image data, time series data, and text data.

[0011] Among them, the preset model library in step S107 includes convolutional neural network CNN, recurrent neural network RNN, long short-term memory network LSTM, and transformer.

[0012] Among them, step S105 includes:

[0013] When the data type is image data, use CNN;

[0014] When the data type is time series data, use RNN or LSTM;

[0015] When the data type is text data, use a transformer.

[0016] Among them, the reusable module in step S107 includes an input layer module, a convolutional layer module, a recurrent layer module, a fully connected layer module, an activation function module, and a loss function module.

[0017] Among them, each module is provided with an input parameter interface, and the user sets it through a visual interface.

[0018] Among them, the parameters of the input layer module include data format, batch size, and input dimension; the parameters of the convolutional layer module include the number of convolutional kernels, the size of the convolutional kernels, the stride, and the padding method; the parameters of the recurrent layer module include the number of hidden units, the time step, and whether to use LSTM or GRU; the parameters of the fully connected layer module include the number of output nodes and whether to use Dropout; the parameters of the activation function module include the selected activation function type; and the parameters of the loss function module include the selected loss function type and weight.

[0019] Among them, the method further includes monitoring the model training process through a visual interface, and the specific steps include:

[0020] Realtime display the change trend of the loss function;

[0021] Provide visual feedback on the model accuracy;

[0022] Allow the user to perform parameter fine-tuning during the training process and observe the change of the model performance in real time.

[0023] Among them, the method further includes evaluating the model training process using the following formula:

[0024] Among them, L is the loss function, Y is the true label vector, is the predicted result vector, λ is the regularization hyperparameter, N is the number of samples, θ j is the model parameter, y i is the i-th true label, is the i-th predicted result, and M is the number of model parameters.

[0025] Among them, the method further includes: the user can save and export the constructed neural network model and its parameters.

[0026] Compared with the prior art, the present invention has the following advantages:

[0027] The core of the "code-free" neural network construction method lies in realizing the design, training, and evaluation of neural networks through a visual interface. Users can select different network modules and parameters through intuitive operations such as dragging and clicking, without writing any code. This method not only reduces the threshold for using deep learning technology but also improves development efficiency and the flexibility of model construction.

[0028] In the code-free neural network construction method, each module has clear input and output interfaces. For example, the input layer module can receive raw data, the convolutional layer module can process image data, and the recurrent layer module can process time-series data. When configuring these modules, users only need to specify relevant parameters such as the convolutional kernel size and the number of hidden units, and the system will automatically generate the corresponding code and execute it. During this process, users can view the structure and status of the model in real time to ensure that every adjustment to the model can be intuitively reflected on the visual interface.

[0029] In addition, with the continuous progress of machine learning and deep learning technologies, the complexity of models is also increasing. Therefore, code-free tools need to integrate more advanced functions such as dynamic learning rate adjustment, regularization, and cross-validation. These functions can be achieved through simple interface settings, avoiding users from manually writing complex parameter adjustment codes. Through these optimizations, users can easily adjust various hyperparameters of the model during training, further improving the performance of the model.

[0030] In terms of data preprocessing, code-free tools also provide rich functions. Steps such as data cleaning, standardization, and feature engineering can be completed through a visual interface. Users can select different preprocessing methods and observe the effects of the processed data in real time. This process not only reduces the risk of errors in manual operations but also improves the efficiency of data processing. Description of the Drawings

[0031] By referring to the following detailed description with reference to the drawings, the above and other objects, features, and advantages of the exemplary embodiments of the present disclosure will become readily understandable. In the drawings, several embodiments of the present disclosure are shown in an exemplary but non-limiting manner, and the same or corresponding reference numerals represent the same or corresponding parts, where:

[0032] Figure 1 is a flowchart showing a code-free neural network construction method based on model visualization according to an embodiment of the present invention. Detailed Embodiments

[0033] To make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings. Apparently, the described embodiments are only a part rather than all of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the scope of protection of the present invention.

[0034] The terms used in the embodiments of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The singular forms "a", "the" and "said" used in the embodiments of the present invention and the appended claims are also intended to include the plural forms unless the context clearly dictates otherwise. "Plural" generally includes at least two.

[0035] It should be understood that although the terms first, second, third, etc. may be used in the embodiments of the present invention to describe..., these... should not be limited to these terms. These terms are only used to distinguish.... For example, without departing from the scope of the embodiments of the present invention, the first... may also be referred to as the second..., and similarly, the second... may also be referred to as the first....

[0036] It should be understood that the term "and / or" used herein is only a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " herein generally represents an "or" relationship between the associated objects before and after.

[0037] Depending on the context, the words "if", "when" as used herein may be interpreted as "when" or "when...", or "in response to determining" or "in response to detecting". Similarly, depending on the context, the phrase "if determined" or "if detecting (stated condition or event)" may be interpreted as "when determined" or "in response to determining", or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)".

[0038] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a commodity or device comprising a series of elements not only includes those elements but also includes other elements not explicitly listed, or further includes elements inherent in such commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the existence of additional identical elements in the commodity or device comprising the element.

[0039] The optional embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0040] Example 1

[0041] As Figure 1 shown, the present invention discloses a method for constructing a code-free neural network based on model visualization, comprising the following steps:

[0042] Step S101, receiving the input data;

[0043] Step S103, identifying the data type of the data;

[0044] Step S105, selecting a corresponding neural network model based on the data type;

[0045] Step S107, extracting a model structure matching the data type from a preset model library;

[0046] Step S109, setting a corresponding number of reusable modules based on the model structure;

[0047] Step S1011, setting corresponding parameters for the reusable modules.

[0048] Example 2

[0049] A method for constructing a code-free neural network based on model visualization proposed by the present invention, comprising the following steps:

[0050] Step S101, receiving the input data;

[0051] Step S103, identifying the data type of the data;

[0052] Step S105, selecting a corresponding neural network model based on the data type;

[0053] Step S107, extracting a model structure matching the data type from a preset model library;

[0054] Step S109, setting a corresponding number of reusable modules based on the model structure;

[0055] Step S1011, setting corresponding parameters for the reusable modules.

[0056] Wherein, the data type in step S103 includes image data, time series data, and text data.

[0057] Wherein, the preset model library in step S107 includes convolutional neural network CNN, recurrent neural network RNN, long short-term memory network LSTM, and transformer.

[0058] Wherein, step S105 includes:

[0059] When the data type is image data, use CNN;

[0060] When the data type is time series data, use RNN or LSTM;

[0061] When the data type is text data, use a Transformer.

[0062] In practical applications, the type of raw data usually determines which deep learning model to use:

[0063] For image data, use a Convolutional Neural Network (CNN) because CNN can effectively extract spatial features.

[0064] For time series data, use a Recurrent Neural Network (RNN) or Long Short-Term Memory Network (LSTM) because these networks can capture the temporal dependencies in the sequence.

[0065] For text data, use a Transformer or RNN, which are suitable for processing natural language.

[0066] Dimensionality reduction of high-dimensional data, especially in image data where the raw data is usually high-dimensional. Extracting meaningful features can reduce computational complexity and improve the performance of subsequent models.

[0067] Information compression. By learning important features, it is possible to retain the most important information for the task, remove redundant data, and enhance the generalization ability of the model.

[0068] Conversion to a format suitable for model input. Some deep learning models require input data to have a specific format or dimension, and feature extraction helps convert the raw data into these formats.

[0069] Throughout the entire process of data processing, the logic of feature extraction can be adjusted as follows:

[0070] Data type identification. Identify the type of input data (such as image, time series data, etc.) and select a suitable deep learning model based on the type.

[0071] Model input. Directly input the raw data (such as an image or a sequence) into the model according to the selected model, rather than performing feature extraction separately.

[0072] Model training and optimization. By training the deep learning model, the model will automatically learn important features internally.

[0073] Feature representation at the output layer. At the output layer of the model, the final feature representation (such as classification results or regression values) can be obtained, and these results stem from the model's deep understanding of the raw data.

[0074] Among them, the reusable modules in the step S107 include an input layer module, a convolutional layer module, a recurrent layer module, a fully connected layer module, an activation function module, and a loss function module.

[0075] Among them, each module is provided with an input parameter interface, and the user sets it through a visual interface.

[0076] Among them, the parameters of the input layer module include data format, batch size, and input dimension; the parameters of the convolutional layer module include the number of convolutional kernels, the size of the convolutional kernels, the stride, and the padding method; the parameters of the recurrent layer module include the number of hidden units, the time step, and whether to use LSTM or GRU; the parameters of the fully connected layer module include the number of output nodes and whether to use Dropout; the parameters of the activation function module include the selected activation function type; and the parameters of the loss function module include the selected loss function type and weight.

[0077] In a certain embodiment, each module has a clear input parameter interface, and the user sets it through a visual interface, specifically including:

[0078] The parameters of the input layer module include data format, batch size, and input dimension.

[0079] Interface example: setInputParameters(dataFormat,batchSize,inputDim)

[0080] The parameters of the convolutional layer module include the number of convolutional kernels, the size of the convolutional kernels, the stride, and the padding method.

[0081] Interface example: setConvLayerParameters(numKernels,kernelSize,stride,padding)

[0082] The parameters of the recurrent layer module include the number of hidden units, the time step, and whether to use LSTM or GRU.

[0083] Interface example: setRNNLayerParameters(hiddenUnits,timeSteps,useLSTM)

[0084] The parameters of the fully connected layer module include the number of output nodes and whether to use Dropout.

[0085] Interface example: setDenseLayerParameters(outputUnits,useDropout)

[0086] The parameters of the activation function module include the selected activation function type (such as ReLU, Tanh).

[0087] Interface example: setActivationFunction(functionType)

[0088] The parameters of the loss function module include the selected loss function type and weights.

[0089] Interface example: setLossFunction(lossType,weights)

[0090] When the user adjusts the module parameters, the system will provide real-time feedback on the current state of the model, such as training accuracy, loss value, etc., to ensure that the user can intuitively understand the impact of each adjustment.

[0091] By visualizing the network architecture through a graphical interface, users can easily view the connections between layers and parameter configurations without writing any code.

[0092] Among them, the method also includes monitoring the model training process through a graphical interface, and the specific steps include:

[0093] Display the changing trend of the loss function in real time;

[0094] Provide visual feedback on the model accuracy;

[0095] Allow users to fine-tune parameters during the training process and observe the changes in model performance in real time.

[0096] In one embodiment, the model training process includes a cross-validation step, and the specific steps are as follows:

[0097] Randomly divide the data set into K subsets;

[0098] Train the model on K - 1 subsets and validate on the remaining 1 subset;

[0099] Repeat the above steps K times, and the final result is the average performance metric of K validations to improve the robustness of the model.

[0100] Among them, the method also includes evaluating the model training process using the following formula:

[0101] Among them, L is the loss function, Y is the true label vector, is the predicted result vector, λ is the regularization hyperparameter, N is the number of samples, θ j are the model parameters, y i is the i-th true label, is the i-th prediction result, and M is the number of model parameters.

[0102] Among them, the regularization hyperparameter λ is usually used to control the complexity of the model to prevent overfitting. It can be obtained by cross-validation, Bayesian optimization, grid search, or random search.

[0103] Taking Bayesian optimization as an example, the regularization hyperparameter can be obtained using the following formula:

[0104] where P(Y|X, λ) represents the probability distribution of the output Y given the input X and the hyperparameter λ 0 under which.

[0105] During the model training process, the true label Y, the prediction result and the regularization hyperparameter λ are used to calculate the loss L. The value of the loss function is used to guide the update of the model parameters, thereby optimizing the performance of the model. Through the backpropagation algorithm, the weights of the model are adjusted according to the loss value to reduce the gap between the predicted value and the true value.

[0106] The loss function L measures the prediction ability of the model. The smaller the value, the closer the prediction result of the model is to the true label. Therefore, this formula is frequently used during the training process to evaluate the learning effect of the model, that is, the value range of L is [0, +∞), where 0 represents perfect fitting, and the smaller the value, the better the model performance.

[0107] The true label vector Y is usually directly obtained from the dataset. For example, for a classification task, the label is the class number; for a regression task, the label is a continuous value.

[0108] The prediction result vector is obtained by inferring the input data through the trained model. It can be calculated through the following steps:

[0109] Forward propagation: The input data X passes through the neural network and undergoes calculations in each layer to obtain the prediction result:

[0110] where f is the forward propagation function of the model and θ are the model parameters.

[0111] The regularization hyperparameter λ is usually determined by cross-validation or hyperparameter optimization methods (such as grid search, random search). Different values can be tried during the training process to select the optimal regularization strength.

[0112] The model parameters θ j are updated through an optimization algorithm (such as gradient descent) during the training process. In each iteration, the parameter update rule can be expressed as:

[0113] Among them, η is the learning rate.

[0114] The number of parameters M is the sum of all learnable parameters in the model. It can be calculated by traversing each layer of the model.

[0115] Among them, the method further includes: the user can save and export the constructed neural network model and its parameters.

[0116] In a certain embodiment, the present invention provides a "save model" function that allows the user to store the current model configuration as a file;

[0117] The model file includes the network structure, training parameters, and training status;

[0118] The user can reload the model in subsequent use to continue training or perform inference.

[0119] In step S109, the system decomposes each function of the neural network into multiple reusable modules, such as the input layer, hidden layer, output layer, activation function, loss function, etc. Each reusable module can be independently configured and combined, and the user can perform drag-and-drop operations through the graphical interface to quickly construct the required network structure.

[0120] Among them, the reusable modules include:

[0121] The input layer module receives data and supports various formats such as images, text, and time-series data.

[0122] The convolutional layer module is used for image processing, and the parameters include the number, size, and stride of the convolutional kernels.

[0123] The recurrent layer module is used for processing time-series data, and the parameters include the number of hidden units and the time step.

[0124] The fully connected layer module connects the output of the previous layer to the current layer, and the parameter includes the number of output nodes.

[0125] The activation function module selects common activation functions (such as ReLU, Sigmoid) without manual implementation.

[0126] The loss function module allows the user to select and configure the loss function (such as mean squared error, cross entropy).

[0127] Among them, the method also provides a variety of data visualization options to help users understand data features, specifically including:

[0128] A histogram of the data distribution;

[0129] A scatter plot of the correlation between features;

[0130] Showing the importance of features in each layer of the neural network through a heat map.

[0131] The visualization interface supports users to customize model parameters, specifically including:

[0132] Select different activation functions and loss functions;

[0133] Adjust training hyperparameters such as learning rate and batch size;

[0134] Provide suggestions for parameter adjustment to optimize the training effect.

[0135] The method also supports multi-user collaboration, and the specific steps include:

[0136] Provide a user account management system to allow multiple users to share the same project;

[0137] Support version control to record the history of each modification;

[0138] Allow users to comment on and discuss the project to promote collaboration.

[0139] Embodiment III

[0140] The present disclosure embodiment provides a non-volatile computer storage medium, and the computer storage medium stores computer-executable instructions, and the computer-executable instructions can execute the method steps described in the above embodiments.

[0141] It should be noted that the above computer-readable medium in the present disclosure can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program, and this program can be used by or in combination with an instruction execution system, apparatus, or device. In the present disclosure, the computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, and this computer-readable signal medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0142] The above computer-readable medium can be included in the above electronic device; or it can exist independently and not be assembled into the electronic device.

[0143] The computer program code for performing the operations of the present disclosure can be written in one or more programming languages or combinations thereof. The above programming languages include object-oriented programming languages - such as Java, Smalltalk, C++, and also include conventional procedural programming languages - such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as an independent software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer can be connected to the user's computer through any type of network - including a local area network (LAN) or a wide area network (WAN) - or it can be connected to an external computer (for example, by using an Internet service provider to connect through the Internet).

[0144] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a part of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks may occur in a different order than that noted in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, may 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.

[0145] The units involved in the embodiments described in the present disclosure may be implemented in software or in hardware. In some cases, the name of the unit does not constitute a limitation on the unit itself.

[0146] The preferred embodiments of the present invention are described above to make the spirit of the present invention clearer and easier to understand, and are not intended to limit the present invention. Any modifications, substitutions, and improvements made within the spirit and principles of the present invention shall be included within the scope of protection defined by the appended claims of the present invention.

Claims

1. A code-free neural network construction method based on model visualization, characterized in that: The following steps are involved: Step S101, receiving input data; Step S103, identifying the data type of the data; Step S105: Select a corresponding neural network model based on the data type; Step S107: extracting a model structure matching the data type from a preset model library; Step S109: Based on the model structure, set a corresponding number of reusable modules; Step S1011: setting corresponding parameters for the reusable module.

2. The method according to claim 1, characterized in that: The data types in step S103 include image data, time series data and text data.

3. The method according to claim 2, characterized in that: The preset model library in step S107 includes a convolutional neural network CNN, a recurrent neural network RNN, a long short-term memory network LSTM, and a transformer.

4. The method according to claim 3, characterized in that: The step S105 comprises: When the data type is image data, CNN is used; When the data type is time series data, RNN or LSTM is used; When the data type is text data, a converter is used.

5. The method according to claim 1, characterized in that: The reusable modules in step S107 include an input layer module, a convolutional layer module, a circulation layer module, a fully connected layer module, an activation function module and a loss function module.

6. The method according to claim 5, characterized in that Each module has an input parameter interface, and users can set it through a visual interface.

7. The method according to claim 6, characterized in that The parameters of the input layer module include data format, batch size, and input dimension; the parameters of the convolution layer module include the number of convolution kernels, convolution kernel size, stride, and padding method; the parameters of the recurrent layer module include the number of hidden units, time step, and whether to use LSTM or GRU; the parameters of the fully connected layer module include the number of output nodes and whether to use Dropout; the parameters of the activation function module include the selected activation function type; and the parameters of the loss function module include the selected loss function type and weight.

8. The method according to claim 1, characterized in that: The method further includes monitoring the model training process through a visual interface, and the specific steps include: Display the changing trend of the loss function in real time; Provide visual feedback on model accuracy; Allows users to fine-tune parameters during training and observe changes in model performance in real time.

9. The method according to claim 1, characterized in that: The method also includes evaluating the training process of the model using the following formula: Among them, L is the loss function, Y is the true label vector, is the prediction result vector, λ is the regularization hyperparameter, N is the number of samples, θ j is the model parameter, y i is the i-th true label, is the i-th prediction result, and M is the number of model parameters.

10. The method according to claim 1, characterized in that: The method also includes: the user can save and export the constructed neural network model and its parameters.

Images