A learning method, storage medium and program product for an information optimization model

Abstract

The present application relates to a learning method, storage medium and program product of an information optimization model. The method comprises: inputting the pre-processed training sample into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the convolution kernel parameters; constructing the convolution layer through the convolution kernel parameters, and using the constructed convolution layer to extract the features of the pre-processed training sample as the first feature; extracting the first feature based on the pooling layer to obtain the second feature; inputting the second feature into the positive and negative sample penalty learning module to obtain a feature group based on the positive and negative sample grouping; expanding the feature group and inputting it into the information optimization layer in the positive and negative sample penalty learning module for learning, and using the mutual information maximization objective function to obtain the learned feature transformation parameters; performing information calculation and classification learning on the grouping of positive and negative samples based on the learned feature transformation parameters. The present application can obtain the maximum amount of information even in the case of small samples.

Description

Technical Field

[0001] The present application relates to the technical field of deep learning and information optimization. More specifically, the present application relates to a learning method, storage medium and program product of an information optimization model. Background Art

[0002] In recent years, driven by big data, deep learning has achieved remarkable results in tasks such as scene classification, object recognition, and speech recognition. However, the success of existing deep learning relies heavily on large amounts of labeled data. In real-world situations, some special scenarios and recognition tasks only contain a small amount of labeled data, and labeling this data requires significant time and labor.

[0003] Furthermore, target categories in real-world scenarios exhibit a long-tail distribution, with data-rich categories accounting for only a small fraction of the total. Training data for target categories of interest is often difficult to obtain. This suggests that building models for application tasks using a small number of samples, or small sample data, and maximizing the amount of information required are pressing technical challenges for the practical application of deep learning and the exploration of new machine learning mechanisms. Summary of the Invention

[0004] Based on the above technical problems, the present invention aims to construct an information optimization model for application tasks based on small sample data, and adopt a new learning method for the information optimization model to obtain the maximum amount of information in the case of small sample data.

[0005] A first aspect of the present invention provides a learning method for an information optimization model, wherein the information optimization model includes an information optimization encoding module and a positive and negative sample penalty learning module, and the learning method includes:

[0006] Acquiring training samples and preprocessing the training samples;

[0007] The preprocessed training samples are input into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the number of convolution kernels;

[0008] Constructing a convolution layer using the number of convolution kernels, and extracting features of the preprocessed training samples using the constructed convolution layer and recording them as first features;

[0009] Extracting the first feature based on the pooling layer to obtain a second feature;

[0010] Inputting the second feature into a positive and negative sample penalty learning module to obtain a feature group based on positive and negative sample grouping;

[0011] Expanding the feature group and inputting it into the information optimization layer of the positive and negative sample penalty learning module for learning, and using the mutual information maximization objective function to obtain the learned feature transformation parameters;

[0012] Based on the learned feature transformation parameters, information calculation and classification learning are performed on the grouping of positive and negative samples to obtain the group with the largest amount of information.

[0013] Specifically, the pre-processed training samples are input into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the convolution kernel parameters, including:

[0014] The preprocessed training samples are input into the information optimization coding module to obtain one-dimensional data based on the sampling layer;

[0015] Performing mean processing and singular value processing on the one-dimensional data to obtain non-zero eigenvalues ​​and eigenvectors;

[0016] Calculate the number of eigenvalues ​​that need to be retained based on the energy ratio that needs to be retained;

[0017] The convolution kernel parameters are obtained based on the number of retained eigenvalues.

[0018] Specifically, the feature group is expanded and input into the information optimization layer in the positive and negative sample penalty learning module for learning, and the learned feature transformation parameters are obtained by using the mutual information maximization objective function, including:

[0019] Expanding the feature group into one-dimensional data is recorded as the third feature;

[0020] Inputting the third feature into the information optimization layer in the positive and negative sample penalty learning module for learning;

[0021] The feature transformation parameters of the information optimization layer are constructed using the mutual information maximization objective function.

[0022] Furthermore, the method further comprises:

[0023] Obtaining the number of convolution kernels in the information optimization layer based on the feature transformation parameters;

[0024] Constructing a convolution layer of the information optimization layer according to the number of convolution kernels;

[0025] The convolutional layer based on the constructed information optimization layer processes and outputs the result input from the information optimization encoding module.

[0026] Furthermore, the method of constructing the feature transformation parameters of the information optimization layer by using the mutual information maximization objective function includes:

[0027] Designing a mutual information maximization objective function, wherein the mutual information maximization objective function includes a minimum interval between the information amounts of positive and negative samples and a feature transformation parameter;

[0028] Updating the feature transformation parameters in the maximization objective function using a gradient descent method;

[0029] The iteration stops when the preset conditions are met, and the feature transformation parameters of the information optimization layer are obtained.

[0030] Further preferably, the performing information calculation and classification learning on the grouping of positive and negative samples based on the learned feature transformation parameters to obtain the group with the largest amount of information includes:

[0031] Use the information calculation function to obtain the amount of information corresponding to the grouping of positive and negative samples;

[0032] Find the group with the largest amount of information among the corresponding amounts of information;

[0033] The group with the largest amount of information is used as the category learned in classification learning.

[0034] A second aspect of the present invention provides a method for learning an information optimization model using the learning method described in each embodiment of the present invention, the method comprising:

[0035] Acquire the target image;

[0036] Inputting the target image into the learned information optimization model;

[0037] Get the category of the target image.

[0038] A third aspect of the present invention provides a computer device, comprising a storage medium storing computer-readable instructions, wherein when the computer-readable instructions are executed by one or more processors, the one or more processors are caused to perform the following steps:

[0039] Acquiring training samples and preprocessing the training samples;

[0040] The preprocessed training samples are input into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the number of convolution kernels;

[0041] Constructing a convolution layer using the number of convolution kernels, and extracting features of the preprocessed training samples using the constructed convolution layer and recording them as first features;

[0042] Extracting the first feature based on the pooling layer to obtain a second feature;

[0043] Inputting the second feature into a positive and negative sample penalty learning module to obtain a feature group based on positive and negative sample grouping;

[0044] Expanding the feature group and inputting it into the information optimization layer of the positive and negative sample penalty learning module for learning, and using the mutual information maximization objective function to obtain the learned feature transformation parameters;

[0045] Based on the learned feature transformation parameters, information calculation and classification learning are performed on the grouping of positive and negative samples to obtain the group with the largest amount of information.

[0046] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to perform the following steps:

[0047] Acquiring training samples and preprocessing the training samples;

[0048] The preprocessed training samples are input into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the number of convolution kernels;

[0049] Constructing a convolution layer using the number of convolution kernels, and extracting features of the preprocessed training samples using the constructed convolution layer and recording them as first features;

[0050] Extracting the first feature based on the pooling layer to obtain a second feature;

[0051] Inputting the second feature into a positive and negative sample penalty learning module to obtain a feature group based on positive and negative sample grouping;

[0052] Expanding the feature group and inputting it into the information optimization layer of the positive and negative sample penalty learning module for learning, and using the mutual information maximization objective function to obtain the learned feature transformation parameters;

[0053] Based on the learned feature transformation parameters, information calculation and classification learning are performed on the grouping of positive and negative samples to obtain the group with the largest amount of information.

[0054] A fifth aspect of the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 7:

[0055] Acquiring training samples and preprocessing the training samples;

[0056] The preprocessed training samples are input into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the number of convolution kernels;

[0057] Constructing a convolution layer using the number of convolution kernels, and extracting features of the preprocessed training samples using the constructed convolution layer and recording them as first features;

[0058] Extracting the first feature based on the pooling layer to obtain a second feature;

[0059] Inputting the second feature into a positive and negative sample penalty learning module to obtain a feature group based on positive and negative sample grouping;

[0060] Expanding the feature group and inputting it into the information optimization layer of the positive and negative sample penalty learning module for learning, and using the mutual information maximization objective function to obtain the learned feature transformation parameters;

[0061] Based on the learned feature transformation parameters, information calculation and classification learning are performed on the grouping of positive and negative samples to obtain the group with the largest amount of information.

[0062] The beneficial effects of the present application are as follows: the present application constructs an information optimization model for application tasks based on small sample data, and proposes a learning method for the information optimization model, wherein the information optimization model includes an information optimization coding module and a positive and negative sample penalty learning module, and the optimization parameters of the two modules are learned through the learning method. In the information optimization module, the preprocessed training samples are input into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the number of convolution kernels, and the convolution layer is constructed by the number of convolution kernels, and the features of the preprocessed training samples are extracted using the constructed convolution layer as the first feature, and the first feature is extracted based on the pooling layer to obtain the second feature; in the positive and negative sample penalty learning module, the second feature is input into the positive and negative sample penalty learning module to obtain a feature group based on the positive and negative sample grouping, and the feature group is expanded and input into the information optimization layer in the positive and negative sample penalty learning module for learning, and the mutual information maximization objective function is used to obtain the learned feature transformation parameters, and the grouping of positive and negative samples is subjected to information calculation and classification learning based on the learned feature transformation parameters. This learning method can obtain features that maximize information even when learning with small samples. By using all the parameters obtained by learning the model in the model to identify the target image, the maximum amount of information can be obtained and it can be applied to various application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate embodiments of the present application and, together with the description, serve to explain the principles of the present application.

[0064] The present application can be more clearly understood from the following detailed description with reference to the accompanying drawings, in which:

[0065] Figure 1 A schematic diagram showing the steps of a learning method for an information optimization model in an exemplary embodiment of the present application is shown;

[0066] Figure 2 A schematic diagram of an information optimization model learning process in an exemplary embodiment of the present application is shown;

[0067] Figure 3 A schematic diagram of the structure of an information optimization model in an exemplary embodiment of the present application is shown;

[0068] Figure 4 A schematic structural diagram of a computer device provided by an exemplary embodiment of the present application is shown;

[0069] Figure 5 A schematic diagram of a storage medium provided by an exemplary embodiment of the present application is shown. DETAILED DESCRIPTION

[0070] Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are merely exemplary and are not intended to limit the scope of the present application. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present application. It is obvious to those skilled in the art that the present application can be implemented without one or more of these details. In other examples, in order to avoid confusion with the present application, some technical features known in the art are not described.

[0071] It should be noted that the terms used herein are only for the purpose of describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of the features, wholes, steps, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, wholes, steps, operations, elements, components and / or combinations thereof.

[0072] Now, exemplary embodiments according to the present application will be described in more detail with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in a variety of different forms and should not be construed as being limited to the embodiments described herein. The accompanying drawings are not drawn to scale, and certain details may be magnified and omitted for the purpose of clarity. The shapes of the various regions and layers shown in the figures and the relative sizes and positional relationships therebetween are merely exemplary and may deviate in practice due to manufacturing tolerances or technical limitations, and those skilled in the art may further design regions / layers with different shapes, sizes, and relative positions according to actual needs.

[0073] The following is attached with the instruction manual Figure 1-5Several examples are given to describe exemplary implementations of the present application. It should be noted that the following application scenarios are only provided to facilitate understanding of the spirit and principles of the present application, and the implementations of the present application are not limited in this respect. On the contrary, the implementations of the present application can be applied to any applicable scenario.

[0074] Example 1:

[0075] This embodiment provides a learning method for an information optimization model, wherein the information optimization model includes an information optimization encoding module and a positive and negative sample penalty learning module. Figure 1 As shown, the learning method includes:

[0076] S1. Obtain training samples and preprocess the training samples;

[0077] S2. Input the preprocessed training samples into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the number of convolution kernels;

[0078] S3. Constructing a convolution layer using the number of convolution kernels, and extracting features of the preprocessed training samples using the constructed convolution layer and recording them as first features;

[0079] S4. Extract the first feature based on the pooling layer to obtain a second feature;

[0080] S5. Input the second feature into a positive and negative sample penalty learning module to obtain a feature group based on positive and negative sample grouping;

[0081] S6, expanding the feature group and inputting it into the information optimization layer of the positive and negative sample penalty learning module for learning, and using the mutual information maximization objective function to obtain the learned feature transformation parameters;

[0082] S7. Perform information calculation and classification learning on the grouping of positive and negative samples based on the learned feature transformation parameters to obtain the group with the largest amount of information.

[0083] Specifically, the preprocessed training samples are input into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the convolution kernel parameters, including: inputting the preprocessed training samples into the information optimization coding module to obtain one-dimensional data based on the sampling layer; performing mean processing and singular value processing on the one-dimensional data to obtain non-zero eigenvalues ​​and eigenvectors; calculating the number of eigenvalues ​​that need to be retained according to the energy ratio that needs to be retained; and obtaining the convolution kernel parameters based on the number of retained eigenvalues.

[0084] Specifically, the feature group is expanded and input into the information optimization layer in the positive and negative sample penalty learning module for learning, and the learned feature transformation parameters are obtained by using the mutual information maximization objective function, including: expanding the feature group into one-dimensional data recorded as the third feature; inputting the third feature into the information optimization layer in the positive and negative sample penalty learning module for learning; and using the mutual information maximization objective function to construct the feature transformation parameters of the information optimization layer.

[0085] Furthermore, the method also includes: obtaining the number of convolution kernels in the information optimization layer based on the feature transformation parameters; constructing a convolution layer of the information optimization layer according to the number of convolution kernels; and processing and outputting the result input from the information optimization encoding module based on the constructed convolution layer of the information optimization layer.

[0086] Furthermore, the feature transformation parameters of the information optimization layer are constructed using the mutual information maximization objective function, including: designing the mutual information maximization objective function, wherein the mutual information maximization objective function includes the minimum interval between the information amount of positive and negative samples and the feature transformation parameters; using the gradient descent method to update the feature transformation parameters in the maximization objective function; and stopping the iteration when the preset conditions are met to obtain the feature transformation parameters of the information optimization layer.

[0087] Further preferably, information calculation and classification learning are performed on the grouping of positive and negative samples based on the learned feature transformation parameters to obtain the group with the largest amount of information, including: using the information calculation function to obtain the amount of information corresponding to the grouping of positive and negative samples; finding the group with the largest amount of information in the corresponding amount of information; and using the group with the largest amount of information as the category learned in the classification learning.

[0088] Example 2:

[0089] This embodiment provides a learning method for an information optimization model, such as Figure 2 As shown, the information optimization model includes an information optimization encoding module and a positive and negative sample penalty learning module, and the steps of the learning method are detailed as follows.

[0090] The first step is to obtain training samples and preprocess the training samples.

[0091] Current research on small sample learning methods is mostly focused on the image field, mainly completing image classification tasks under small sample conditions. From the implementation process, the image training set for small sample learning contains many categories, and each category has multiple samples. The training samples here can be selected from continuous images in pictures or videos. Given an input training data set, which contains a preset number of categories such as Z, Z is a natural number. For example, the training sample data here is x, and it is preprocessed to obtain The character on the left of the equal sign indicates that the mathematical expectation ∑U of the data x is obtained by decomposing the covariance matrix of the data and satisfies

[0092] In the second step, the preprocessed training samples are input into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the convolution kernel parameters.

[0093] like Figure 2 As shown, the information optimization coding module includes a sampling layer, a small sample convolution kernel learning layer, a convolution layer, and a pooling layer. In a preferred embodiment, the kth convolution information block is first initialized according to the hyperparameters. The hyperparameter S of the sampling layer is used. k For s k =(w k ×h k , l k ), for input data According to the specified window size w k ×h k , specify the sliding step size l k Intercept the local neighborhood and generate sampled data Then Expand into one-dimensional N×L k , L k =d×w k ×h k Where N is the number of image blocks after sampling, and d is the number of data channels. The expanded one-dimensional data is processed by mean and singular value processing to obtain non-zero eigenvalues ​​and eigenvectors; the number of eigenvalues ​​to be retained is calculated based on the energy ratio to be retained; and the convolution kernel parameters are obtained based on the number of retained eigenvalues. The formula for calculating the number of eigenvalues ​​to be retained is:

[0094]

[0095] where ε is a hyperparameter of the denoising sparsification layer, which is included in the small sample convolution kernel learning layer ( Figure 2 Not shown), represents the amount of information retained during the information transmission process of this layer, Represents the non-zero eigenvalues ​​arranged from large to small. It should be emphasized that the selection of non-zero eigenvalues ​​is to weaken the impact of marginal distribution in the case of small samples. Then we can make the following transformation on k0:

[0096]

[0097]

[0098]

[0099] Among them, diag represents a diagonal matrix. The transformation is to obtain the output of the layer, and then the number of convolution kernels is obtained by the derivative of the activation function.

[0100] Of course, as a transformable implementation method, before performing mean processing on the expanded one-dimensional data, the data may also be weighted using a balance factor, which is not limited here.

[0101] In the third step, a convolution layer is constructed using the number of convolution kernels, and the features of the preprocessed training samples are extracted using the constructed convolution layer and recorded as the first features.

[0102] These convolutional layers can be thought of as filters, each extracting a specific feature. The general understanding of the convolutional layer's function is that the spatial connections between images are close, while pixels farther away have weaker correlations. Therefore, each neuron doesn't need to perceive the entire image; it only needs to perceive the local area, and then integrate this local information at higher levels to obtain the global picture.

[0103] In the fourth step, the first feature is extracted based on the pooling layer to obtain the second feature.

[0104] In addition to convolutional layers, convolutional networks often use pooling layers to reduce the size of the model, increase computing speed, and improve the robustness of the extracted features. By performing feature extraction through the pooling layer hyperparameters, the output of the kth convolution information block is obtained. Figure 2 As shown, it is also possible to determine whether it is the last convolution information block, and if so, it is confirmed to be the final output result of the information optimization coding module.

[0105] In the fifth step, the second feature is input into the positive and negative sample penalty learning module to obtain a feature group based on the positive and negative sample grouping.

[0106] For example, the extracted feature F is input into the positive and negative sample penalty learning module, and the features of the i-th category and the j-th category are put together to form a feature group. At this time, consider and It means that the i-th class is a positive sample, the j-th class is a negative sample, and there are a total of Z*(Z-1) feature groups.

[0107] In the sixth step, the feature group is expanded and input into the information optimization layer in the positive and negative sample penalty learning module for learning, and the learned feature transformation parameters are obtained by using the mutual information maximization objective function.

[0108] The feature transformation parameters of the information optimization layer are constructed using the mutual information maximization objective function, including: designing the mutual information maximization objective function, wherein the mutual information maximization objective function includes the minimum interval between the information amount of positive and negative samples and the feature transformation parameters; using the gradient descent method to update the feature transformation parameters in the maximization objective function; when the preset conditions are met, the iteration stops and the feature transformation parameters of the information optimization layer are obtained. The mutual information maximization objective function is:

[0109] -0.5*γ0 ln(g(f(x))-0.5*γ0 ln(g(f(x′)))-0.5*Q[C]

[0110]

[0111] g(f(x))=sigmoid(β(s(C T x+b)-m)) 2 +e0

[0112] Among them, β represents the scaling factor, C represents the feature transformation parameter, b is the bias corresponding to the model, m is the minimum interval between the amount of information of positive and negative samples, s is 1 when a positive sample is input, otherwise s is -1, and e0 is a minimum value used to prevent the calculated value of mutual information from being extremely small. The iterative method is as follows:

[0113]

[0114]

[0115] Where t represents the number of iterations, μ t Represents the learning rate parameter, which changes with the number of iterations. When the number of iterations reaches a certain value, C can be regarded as the learned feature transformation parameters. The learned feature transformation parameters are connected with the parameters learned by the previous model to obtain the parameters of P networks, where P is the total number of parameters learned by the information optimization model of this application, with a dimension of M×(M1+…+M p ), where the learned parameters are different from the hyperparameters.

[0116] In the seventh step, information calculation and classification learning are performed on the grouping of positive and negative samples based on the learned feature transformation parameters to obtain the group with the largest amount of information.

[0117] In a preferred embodiment, information calculation and classification learning are performed on the grouping of positive and negative samples based on the learned feature transformation parameters to obtain the group with the largest amount of information, including: using the information calculation function to obtain the amount of information corresponding to the grouping of positive and negative samples; finding the group with the largest amount of information in the corresponding amount of information; and using the group with the largest amount of information as the category learned in the classification learning. Using the information calculation function ln to obtain the amount of information of the corresponding group ln(y1) to ln(y z(z-1) ), then divide it into Z groups in order, and the output of group i is ln(y (i-1)*(z-1) ),ln( yi*(z-1) ), find the maximum value. Find the group with the largest amount of information. That is, assuming that the node with the largest amount of data output information belongs to the zth class, then the data label information is z, and finally achieve learning and classification under small sample data conditions.

[0118] This application constructs an information optimization model for application tasks based on small sample data, and proposes a learning method for the information optimization model. The information optimization model includes an information optimization coding module and a positive and negative sample penalty learning module. The optimization parameters of these two modules are learned through the learning method. In the information optimization module, the preprocessed training samples are input into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the number of convolution kernels. The convolution layer is constructed by the number of convolution kernels, and the features of the preprocessed training samples are extracted using the constructed convolution layer as the first feature. The first feature is extracted based on the pooling layer to obtain the second feature. In the positive and negative sample penalty learning module, the second feature is input into the positive and negative sample penalty learning module to obtain a feature group based on the positive and negative sample grouping. The feature group is expanded and input into the information optimization layer in the positive and negative sample penalty learning module for learning, and the mutual information maximization objective function is used to obtain the learned feature transformation parameters. Based on the learned feature transformation parameters, information calculation and classification learning are performed on the grouping of positive and negative samples. This learning method can obtain features that maximize information even when learning with small samples. By using all the parameters obtained by learning the model in the model to identify the target image, the maximum amount of information can be obtained and it can be applied to various application scenarios.

[0119] Example 3:

[0120] This embodiment provides a method for learning an information optimization model using the learning method described in each embodiment, the method comprising:

[0121] Acquire the target image;

[0122] Inputting the target image into the learned information optimization model;

[0123] Get the category of the target image.

[0124] Here, as Figure 3 As shown, the information optimization model includes an information optimization coding module 301 and a positive and negative sample penalty learning module 302, and also includes a calculation and classification module 303. The calculation and classification module is used to perform information calculation and classification learning on the grouping of positive and negative samples based on the learned feature transformation parameters to obtain the group with the largest amount of information. The specific implementation is: using the information calculation function to obtain the amount of information corresponding to the grouping of positive and negative samples; finding the group with the largest amount of information in the corresponding amount of information; and using the group with the largest amount of information as the category learned in the classification learning.

[0125] Please refer to the following Figure 4 , which shows a schematic diagram of a computer device provided by some embodiments of the present application. Figure 4 As shown, the computer device 2 includes: a processor 200, a memory 201, a bus 202 and a communication interface 203, and the processor 200, the communication interface 203 and the memory 201 are connected via the bus 202; the memory 201 stores a computer program that can be run on the processor 200, and when the processor 200 runs the computer program, it executes the learning method of the information optimization model provided in any of the aforementioned embodiments of the present application. The computer device can be a computer device with a touch-sensitive display.

[0126] The memory 201 may include high-speed random access memory (RAM) and may also include non-volatile memory, such as at least one disk storage. The system network element communicates with at least one other network element via at least one communication interface 203 (which may be wired or wireless), and may utilize the Internet, a wide area network, a local area network, a metropolitan area network, or the like.

[0127] The bus 202 may be an ISA bus, a PCI bus, or an EISA bus. The bus may be divided into an address bus, a data bus, a control bus, etc. The memory 201 is used to store programs, and the processor 200 executes the programs after receiving execution instructions. The information optimization model learning method disclosed in any of the aforementioned embodiments of the present application may be applied to the processor 200 or implemented by the processor 200.

[0128] The processor 200 may be an integrated circuit chip with signal processing capabilities. During implementation, each step of the above method can be completed by hardware integrated logic circuits in the processor 200 or by software instructions. The above processor 200 may be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), an off-the-shelf field programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components. It can implement or execute the various methods, steps, and logic block diagrams disclosed in the embodiments of this application. The general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in conjunction with the embodiments of this application can be directly implemented and executed by a hardware decoding processor, or by a combination of hardware and software modules in the decoding processor. The software module can be located in a storage medium mature in the art, such as random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, etc. The storage medium is located in the memory 201 , and the processor 200 reads the information in the memory 201 and completes the steps of the above method in combination with its hardware.

[0129] The computer device provided in the embodiment of the present application and the learning method of the information optimization model provided in the embodiment of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, operated or implemented therein.

[0130] The present application also provides a computer-readable storage medium corresponding to the learning method of the information optimization model provided in the above embodiment. Figure 5 , Figure 5 The computer-readable storage medium shown is an optical disc 30 on which a computer program (ie, a program product) is stored. When the computer program is executed by a processor, the learning method of the information optimization model provided by any of the aforementioned embodiments is executed.

[0131] In addition, examples of the computer-readable storage medium may also include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other optical or magnetic storage media, which are not listed here one by one.

[0132] The computer-readable storage medium provided in the above-mentioned embodiments of the present application and the method for allocating quantum key distribution channels in a space-division multiplexed optical network provided in the embodiments of the present application are based on the same inventive concept and have the same beneficial effects as the methods adopted, run or implemented by the application programs stored therein.

[0133] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the learning method of the information optimization model provided by any of the aforementioned embodiments, the steps of the method including: obtaining training samples and preprocessing the training samples; inputting the preprocessed training samples into the information optimization coding module for learning based on the sampling layer and the small sample convolution kernel learning layer to obtain the number of convolution kernels; constructing a convolution layer using the number of convolution kernels, and using the constructed convolution layer to extract the features of the preprocessed training samples as the first feature; extracting the first feature based on the pooling layer to obtain the second feature; inputting the second feature into the positive and negative sample penalty learning module to obtain a feature group based on the grouping of positive and negative samples; expanding the feature group and inputting it into the information optimization layer in the positive and negative sample penalty learning module for learning, and using the mutual information maximization objective function to obtain the learned feature transformation parameters; performing information calculation and classification learning on the grouping of positive and negative samples based on the learned feature transformation parameters to obtain the group with the largest amount of information.

[0134] It should be noted that the algorithms and displays provided herein are not inherently related to any particular computer, virtual device, or other device. Various general-purpose devices may also be used in conjunction with the teachings herein. Based on the above description, the structure required to construct such devices is apparent. Furthermore, this application is not directed to any particular programming language. It should be understood that the content of this application described herein can be implemented using various programming languages, and the above description of specific languages ​​is intended to disclose the best implementation methods of this application. In the description provided herein, a large number of specific details are described. However, it is understood that the embodiments of the present application can be practiced without these specific details. In some instances, well-known methods, structures, and techniques are not shown in detail so as not to obscure the understanding of this specification.

[0135] Similarly, it should be understood that in order to streamline the present application and aid understanding of one or more of the various inventive aspects, in the above description of the exemplary embodiments of the present application, various features of the present application are sometimes grouped together into a single embodiment, figure, or description thereof. However, this disclosed method should not be interpreted as reflecting an intention that the claimed application requires more features than are expressly recited in each claim. Rather, as reflected in the claims below, inventive aspects lie in less than all the features of the individual embodiments disclosed above. Accordingly, the claims following the detailed description are hereby expressly incorporated into this detailed description, with each claim standing on its own as a separate embodiment of the present application.

[0136] The various component embodiments of the present application can be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art will appreciate that a microprocessor or digital signal processor (DSP) can be used in practice to implement some or all of the functions of some or all of the components in the creation device of the virtual machine according to the embodiment of the present application. The application can also be implemented as a device or device program for executing part or all of the methods described herein. The program implementing the application can be stored on a computer-readable medium, or can have the form of one or more signals. Such a signal can be downloaded from an Internet website, or provided on a carrier signal, or provided in any other form.

[0137] The above description is merely a preferred embodiment of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A learning method for an information optimization model, characterized in that: The information optimization model includes an information optimization encoding module and a positive and negative sample penalty learning module, and the learning method includes: Acquire training samples and preprocess the training samples, wherein the training samples are continuous images in pictures or videos; The preprocessed training samples are input into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the number of convolution kernels; Constructing a convolution layer using the number of convolution kernels, and extracting features of the preprocessed training samples using the constructed convolution layer and recording them as first features; Extracting the first feature based on the pooling layer to obtain a second feature; Inputting the second feature into a positive and negative sample penalty learning module to obtain a feature group based on positive and negative sample grouping; Expanding the feature group and inputting it into the information optimization layer of the positive and negative sample penalty learning module for learning, and using the mutual information maximization objective function to obtain the learned feature transformation parameters; Based on the learned feature transformation parameters, information calculation and classification learning are performed on the grouping of positive and negative samples to obtain the group with the largest amount of information; The information calculation and classification learning of the grouping of positive and negative samples based on the learned feature transformation parameters to obtain the group with the largest amount of information includes: Use the information calculation function to obtain the amount of information corresponding to the grouping of positive and negative samples; Find the group with the largest amount of information among the corresponding amounts of information; The group with the largest amount of information is used as the category of the continuous images in the picture or video learned in the classification learning.

2. The information optimization model learning method according to claim 1, characterized in that: The feature group is expanded and input into the information optimization layer of the positive and negative sample penalty learning module for learning, and the learned feature transformation parameters are obtained by using the mutual information maximization objective function, including: Expanding the feature group into one-dimensional data is recorded as the third feature; Inputting the third feature into the information optimization layer in the positive and negative sample penalty learning module for learning; The feature transformation parameters of the information optimization layer are constructed using the mutual information maximization objective function.

3. The learning method of the information optimization model according to claim 2, characterized in that: The method further comprises: Obtaining the number of convolution kernels in the information optimization layer based on the feature transformation parameters; Constructing a convolution layer of the information optimization layer according to the number of convolution kernels; The convolutional layer based on the constructed information optimization layer processes and outputs the result input from the information optimization encoding module.

4. The learning method of the information optimization model according to claim 2, characterized in that: The method of constructing the feature transformation parameters of the information optimization layer by utilizing the mutual information maximization objective function includes: Designing a mutual information maximization objective function, wherein the mutual information maximization objective function includes a minimum interval between the information amounts of positive and negative samples and a feature transformation parameter; Updating the feature transformation parameters in the maximization objective function using a gradient descent method; The iteration stops when the preset conditions are met, and the feature transformation parameters of the information optimization layer are obtained.

5. The learning method of the information optimization model according to any one of claims 1 to 4, characterized in that: The pre-processed training samples are input into the information optimization coding module to learn based on the sampling layer and the small sample convolution kernel learning layer to obtain the convolution kernel parameters, including: The preprocessed training samples are input into the information optimization coding module to obtain one-dimensional data based on the sampling layer; Performing mean processing and singular value processing on the one-dimensional data to obtain non-zero eigenvalues ​​and eigenvectors; Calculate the number of eigenvalues ​​that need to be retained based on the energy ratio that needs to be retained; The convolution kernel parameters are obtained based on the number of retained eigenvalues.

6. A method for learning an information optimization model using the learning method according to claim 1, characterized in that: The method comprises: Acquire the target image; Inputting the target image into the learned information optimization model; Get the category of the target image.

7. A computer device having a storage medium storing computer-readable instructions, characterized in that: When the computer-readable instructions are executed by one or more processors, the one or more processors are caused to perform the steps of any one of the methods of claims 1-6.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

9. A computer program product comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images