Data processing method and device based on neural cluster, storage medium, and processor

Abstract

The application discloses a data processing method and device based on a neural cluster, a storage medium and a processor. The method comprises the following steps: obtaining original data, performing a common space mode transformation on the original data to obtain transformed data; obtaining a first target function containing a first matrix based on the transformed data, wherein the first target function is a target function of a neural cluster network model of the original data, and the first matrix is a weight parameter of the target function of the neural cluster network model; updating the first matrix based on a preset gradient descent update rule to obtain a second matrix; and updating the first target function according to the second matrix. The application solves the technical problems of low training efficiency and poor robustness of a supervised learning model in the prior art.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of machine learning, in particular to a data processing method and device based on neural cluster, storage medium and processor. BACKGROUND

[0002] Machine learning has been widely applied in data mining, computer vision, natural language processing, biometric identification and other fields. The key of machine learning is to discover unknown structure in data, and to learn a good feature representation from observed data, which can help to reveal the underlying structure of data. At present, machine learning mainly includes two methods of supervised learning and unsupervised learning. Among them, supervised learning is a machine learning task of inferring a function from labeled training data, and the training data includes a set of training examples. In supervised learning, each example is composed of an input object (usually a vector) and an expected output value (also known as a supervision signal), and the supervised learning algorithm analyzes the training data and generates an inferred function, which can be used to map new instances.

[0003] At present, supervised representation learning mainly applies support vector machine (SVM) suitable for shallow model and back propagation (BP) algorithm suitable for deep learning model. At present, SVM is only suitable for shallow model and small sample, and is difficult to extend to deep model; BP algorithm is the main basic algorithm of deep learning at present, but a large number of training samples are required to achieve good results, and has the disadvantages of low training efficiency and poor robustness.

[0004] At present, there is no effective solution to the problem of low training efficiency and poor robustness of the supervised learning model in the prior art. SUMMARY

[0005] The embodiments of the present application provide a data processing method and device based on neural cluster, storage medium and processor, to at least solve the technical problem of low training efficiency and poor robustness of the supervised learning model in the prior art.

[0006] According to an aspect of an embodiment of the present application, a data processing method based on neural cluster is provided, comprising: obtaining original data, and performing common space mode transformation on the original data to obtain transformed data; obtaining a first target function containing a first matrix based on the transformed data, wherein the first target function is a target function of a neural cluster network model, and the first matrix is a weight parameter of the target function of the neural cluster network model; updating the first matrix based on a preset gradient descent update rule to obtain a second matrix; and updating the first target function according to the second matrix.

[0007] Further, the original data is acquired, and the original data is subjected to a common space mode transformation to obtain transformed data, including: acquiring an input vector and a neuron output vector representing the original data; determining an interaction information formula according to the input vector and the neuron output vector of the original data; determining a second target function including a covariance matrix and a transformation matrix; obtaining the transformation matrix according to the interaction information formula and the second target function; and transforming the original data into the transformed data according to the transformation matrix.

[0008] Further, under the condition that the number of neuron output vectors is greater than the vector dimension of the original data, the transformation matrix is obtained according to the interaction information formula and the second target function, including: obtaining a high approximation formula of the interaction information formula; and obtaining the transformation matrix according to the high approximation formula and the second target function.

[0009] Further, the first matrix is updated based on a preset gradient descent update rule to obtain a second matrix, including: updating the first matrix based on the preset gradient descent update rule to obtain a third matrix; determining an iteration number, wherein the iteration number is used to represent the number of times of updating the first matrix based on the preset gradient descent update rule; determining whether the iteration number reaches a preset number, if yes, taking the third matrix as the second matrix and outputting, and if not, assigning the third matrix to the first matrix and returning to the step of updating the first matrix based on the preset gradient descent update rule to obtain the third matrix.

[0010] Further, before the first matrix is updated based on the preset gradient descent update rule to obtain the third matrix, the method further includes: calculating a derivative of the first target function with respect to the first matrix.

[0011] Further, the first target function is updated according to the second matrix, including: performing an orthogonalization transformation on the second matrix to obtain an orthogonalization result; and updating the value of the first target function according to the orthogonalization result.

[0012] Further, the orthogonalization transformation is a Gram-Schmidt orthogonalization transformation.

[0013] According to another aspect of the embodiment of the present application, a data processing device based on a neural cluster is also provided, including: a transformation module configured to acquire original data and perform a common space mode transformation on the original data to obtain transformed data; a function acquisition module configured to obtain a first target function including a first matrix based on the transformed data, wherein the first target function is a target function of a neural cluster network model, and the first matrix is a weight parameter of the target function of the neural cluster network model; a matrix update module configured to update the first matrix based on a preset gradient descent update rule to obtain a second matrix; and a function update module configured to update the first target function according to the second matrix.

[0014] According to another aspect of the embodiments of the present application, there is also provided a storage medium comprising a stored program, wherein the program, when executed by a device in which the storage medium is located, controls the device to perform the neural cluster-based data processing method.

[0015] According to another aspect of the embodiments of the present application, there is also provided a processor configured to execute a program, wherein the program, when executed, performs the neural cluster-based data processing method.

[0016] In the embodiments of the present application, the neural cluster-based supervised representation learning algorithm according to the above steps is used to obtain transformed data by performing CSP transformation on the obtained raw data, construct a supervised learning objective function of a neural cluster network model according to the transformed data, and update the weight parameter matrix in the model according to a preset gradient descent update rule, so as to realize the fast optimization of the weight parameters of the neural cluster network model. Based on the supervised representation learning algorithm, the efficiency and performance of the training learning of the neural cluster network model are improved, and the problems of low training efficiency and poor robustness of the supervised learning model in the prior art are solved. BRIEF DESCRIPTION OF DRAWINGS

[0017] The accompanying drawings, which are included to provide a further understanding of the application and are incorporated in and constitute a part of this application, illustrate embodiments of the application and together with the description serve to explain the application. In the drawings:

[0018] Figure 1 is a flowchart of a neural cluster-based data processing method according to an embodiment of the present application;

[0019] Figure 2 is a flowchart of an optional neural cluster-based data processing method according to an embodiment of the present application;

[0020] Figure 3 is an example diagram of a handwritten digit dataset MNIST;

[0021] Figure 4 is a schematic diagram of the weight parameter C learned after processing the dataset according to an embodiment of the present application; Figure 3

[0022] Figure 5 is a schematic diagram of a neural cluster-based data processing device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] ​In the following, the technical solutions in the embodiments of the present application will be described clearly and completely with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments of the present application. Based on the embodiments in the present application, all the other embodiments obtained by a person of ordinary skill in the art without creative effort should belong to the protection scope of the present application.

[0024] It should be noted that the terms "first", "second" and the like in the description and claims of the present application and the above-described drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the data thus used can be interchanged under appropriate circumstances, so that the embodiments of the present application described herein can be implemented in other than the order illustrated or described herein. In addition, the terms "comprise" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a list of steps or units does not necessarily limit to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0025] Embodiment 1

[0026] According to the embodiments of the present application, a neural cluster-based data processing method embodiment is provided. It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0027] Figure 1 is a neural cluster-based data processing method according to the embodiments of the present application, as shown in Figure 1 The method comprises the following steps:

[0028] Step S101, obtaining original data, and performing common spatial pattern transformation on the original data to obtain transformed data.

[0029] The original data is image data, voice data or signal data and the like from image recognition, natural language processing, speech recognition, signal analysis and the like.

[0030] Common spatial pattern transformation, i.e. CSP (Common spatial pattern) transformation, can transform the original data x to obtain transformed data Wherein, is the transpose matrix of the transformation matrix V; the CSP transformation can initially expand the differences between different classes of original data, facilitating further learning and training for classification to improve learning efficiency.

[0031] In step S102, a first target function including a first matrix can be obtained based on the transformed data, wherein the first target function is a target function of the neural cluster network model, and the first matrix is a weight parameter of the target function of the neural cluster network model.

[0032] The first target function is a supervised learning target function of the neural cluster network model. In an optional embodiment, the first target function is Q[C], the first matrix is C, the first matrix C is a weight parameter of the first target function Q[C], and the expression of the first target function can be:

[0033]

[0034] Wherein β and m are non-negative constants, and m can be regarded as a margin parameter.

[0035] In an optional embodiment, in order to distinguish the first matrix and the second matrix, the first matrix C in step S102 is denoted as C t , and the second matrix obtained after updating is denoted as C t+1 .

[0036]

[0037] Wherein, the learning rate parameter μ t = v t / κ t , 0 < v < 1, t = 1, …, t max . ||▽C t (:,k)|| represents the modulus value of the gradient vector of the first matrix Ct.

[0038] In step S104, the first target function is updated according to the second matrix.

[0039] The second matrix is obtained by iterative updating of the first matrix, and therefore the second matrix is also a weight parameter in the above-mentioned first target function. The obtained second matrix C t+1 is substituted into the above-mentioned first target function Q[C] (i.e., C t+1 in the above-mentioned first target function is replaced by C), to obtain the updated first target function Q[C], and the optimization of the first target function is realized by updating the weight parameter in the first target function.

[0040] According to the neural cluster-based supervised representation learning algorithm proposed in the above steps, the obtained original data is preprocessed by CSP transformation to obtain transformed data, a supervised learning objective function of a neural cluster network model is constructed according to the transformed data, and the weight parameter matrix in the model is updated according to the preset gradient descent update rule, so that the weight parameters of the neural cluster network model are quickly optimized. Based on the supervised representation learning algorithm, it can not only be used for training and learning of large data samples, but also be used for training and learning of small data samples. Through the above CSP transformation, the noise of the original data can be filtered out, and the difference between different classes of the original data is enlarged, thereby improving the efficiency and performance and robustness of the neural cluster network model in training and learning without increasing the computational complexity, and solving the problems of low training efficiency and poor robustness of the supervised learning model in the prior art.

[0041] As an optional embodiment, the above step S101 of obtaining the original data and performing common spatial pattern transformation on the original data to obtain transformed data comprises: obtaining an input vector representing the original data and a neuron output vector; determining an interaction information formula according to the input vector and the neuron output vector of the original data; determining a second objective function comprising a covariance matrix and a transformation matrix; obtaining the transformation matrix according to the interaction information formula and the second objective function; and transforming the original data into transformed data according to the transformation matrix.

[0042] Since each neuron of the brain neural system is connected to more than several thousand other neurons, the encoding of the brain nerve involves large-scale neuron cluster coding, and the neural cluster network model is established based on the neuron inspiration of the brain neural system. Conditional mutual information (i.e. interaction information) is understood as the amount of information about another random variable contained in a certain condition constraint.

[0043] The CSP transformation process of the original data is described as follows: an input vector representing the original data and a neuron output vector are obtained, the input vector x is a K-dimensional vector, and the input vector x can be denoted as x=(x1,…,x K ) T , and the corresponding data label is t. The neuron output vector contains N neurons, which can be denoted as r=(r1,…,r N ) T , and the corresponding random variables are represented by capital letters X, T and R. The interaction information I between the input vector x and the output vector r is denoted as:

[0044]

[0045] , p(r|t) and p(x|t) represent conditional probability density functions, <·> r,x,tdenotes an expected value for the probability density function p(x, r, t).

[0046] If the corresponding label data t is set to have only two categories, i.e., t e {1, -1}, the covariance matrices of the two categories of label data are denoted as Σ1 and Σ2, respectively, and the covariance matrices are normalized to obtain:

[0047]

[0048] wherein Tr denotes a trace of a matrix. The transformation matrix V is obtained by minimizing the following objective function L(V):

[0049] Minimizing satisfies:

[0050] Solving the above objective function L(V) obtains: wherein U is an eigenvector matrix and D is a diagonal matrix of eigenvalues.

[0051] After obtaining the transformation matrix V, the transformed data of the input vector x after CSP transformation is The expression of the transformed data is:

[0052] Through the above steps, the common spatial pattern (CSP) transformation preprocessing of the original data is realized. After the CSP transformation is completed, the subsequent parameter training and learning of the supervised learning objective function of the neural cluster network model constructed based on the obtained transformed data are performed. Compared with the supervised learning method in the prior art in which the original data is simply normalized and then learned, the training and learning efficiency and effect are improved.

[0053] As an optional embodiment, under the condition that the number of neuron output vectors is greater than the vector dimension of the original data, the transformation matrix is obtained according to the mutual information formula and the second objective function, including: obtaining a high approximation formula of the mutual information formula; and obtaining the transformation matrix according to the high approximation formula and the second objective function.

[0054] Under the condition that the number N of neuron output vectors is greater than the vector dimension K of the original data, for example, N is much greater than K, the mutual information I (whose random variables include X, T and R, and the mutual information I is denoted as I(R; X|T)) can be highly approximated by the following formula: G The expression of the high approximation formula I

[0055]

[0056] wherein det(·) denotes a matrix determinant, and H(X|T) = -<lnp(x|t) x,trepresents the conditional entropy of X under condition T, where the expression of G(x, t) is as follows:

[0057]

[0058] I G is substituted into the following CSP transformation formula as the mutual information I:

[0059] minimize satisfying:

[0060] Solving the above objective function L(V) obtains the transformation matrix V, and after obtaining the transformation matrix V, the transformed data of the input vector x after CSP transformation is expressed as:

[0061] According to the above steps, the objective function based on the maximization of conditional mutual information is constructed. Compared with the objective function based on error square and the objective function based on cross-entropy in the prior art, the embodiment can significantly improve the learning training efficiency and performance of the neural cluster network model.

[0062] As an optional embodiment, the first matrix is updated based on a preset gradient descent update rule to obtain a second matrix, including: the first matrix is updated based on the preset gradient descent update rule to obtain a third matrix; the number of iterations is determined, wherein the number of iterations is used to represent the number of times that the first matrix is updated based on the preset gradient descent update rule; it is judged whether the number of iterations reaches a preset number, if yes, the third matrix is taken as the second matrix and output, if not, the third matrix is assigned to the first matrix, and the step of updating the first matrix based on the preset gradient descent update rule to obtain the third matrix is returned.

[0063] The preset gradient descent update rule can be:

[0064]

[0065] wherein the data label t is the number of iterations, the learning rate parameter μ t = v t / κ t With the change of the number of iterations t, 0 < v1 < 1, t = 1, …, t max , ||▽C t (:,k)|| represents the modulus value of the gradient vector of the first matrix C.

[0066] The preset number of times is t max , that is, the maximum number of iterations of the first matrix. According to the above gradient descent update rule, the first matrix C tUpdated to the third matrix C t+1 Determine if the iteration number t+1 of the third matrix is ​​equal to t. max If so, then the third matrix C t+1 C tmax That is, C t After t max After the second iteration, the final optimized weight parameter C was obtained. tmax (i.e., C) opt Finally, the weight parameter C is optimized. opt This is the output of the second matrix mentioned above. If the iteration count t+1 has not reached t... max Then, continue iterating the first matrix according to the gradient descent update rule described above until the number of iterations reaches the preset maximum to obtain the final optimized weight parameter C. opt For example, if the preset number of iterations is 3, according to the gradient descent update rule, C... 1 Get C 2 Continue iterating, from C 2 Get C 3 C 3 If the number of iterations reaches the preset number, then C 3 This is the output of the second matrix, which serves as the final optimization weight parameters.

[0067] This embodiment proposes an adaptive gradient descent method, which has higher training efficiency compared to the existing stochastic gradient descent method. Furthermore, the optimized parameter C is obtained through the above method. opt The system can also be used directly for classification and recognition. It can determine the category of the input by calculating the amount of output information after the input stimulus is transformed by the neural cluster.

[0068] As an optional embodiment, before updating the first matrix based on a preset gradient descent update rule to obtain the third matrix, the method further includes: calculating the derivative of the first objective function with respect to the first matrix.

[0069] Specifically, the derivative expression of the first objective function Q[C] with respect to C is as follows:

[0070]

[0071] in, k = 1, 2, ..., E, where E represents the number of output features.

[0072] It should be noted that the derivative expression of the first objective function Q[C] with respect to C is part of the gradient descent update rule mentioned above.

[0073] As an optional embodiment, the first objective function is updated according to the second matrix, comprising: performing an orthogonal transformation on the second matrix to obtain an orthogonal result; and updating the value of the first objective function according to the orthogonal result.

[0074] As an optional embodiment, the orthogonal transformation is a Gram-Schmidt orthogonal transformation.

[0075] By performing the CSP transformation on the original data, the noise in the original data can be filtered out, and the second matrix is constrained to be orthogonal, thereby greatly increasing the robustness of the neural cluster network model and the efficiency of learning and training.

[0076] Figure 2 A flowchart of an optional neural cluster-based data processing method according to an embodiment of the present application is provided, and a handwritten digital MNIST data set is used as the data set (as shown in Figure 3 The data set MNIST example diagram is shown), which has a total of 60,000 gray handwritten sample pictures, a total of 10 categories (digits 0 to 9), and the image size is 28*28. In this embodiment, 60,000 training sample pictures are used as the original training data set as input, as shown in Figure 2 The method comprises the following steps:

[0077] Step S201, performing a common spatial pattern transformation preprocessing on the original data set x to obtain transformed data Wherein, V is a transformation matrix obtained according to the common spatial pattern transformation.

[0078] Step S202, initializing the matrix C and other parameters, and calculating the objective function Q:

[0079]

[0080] Wherein β and m are non-negative constants, and m can be regarded as a margin parameter.

[0081] Step S203, setting the maximum number of iterations as t max = 50.

[0082] Step S204, determining whether the maximum number of iterations is reached; if yes, entering step S208 to output the last optimized parameter matrix C and other parameters; if no, entering step S205.

[0083] Step S205, calculating the derivative of Q with respect to C:

[0084]

[0085] Wherein, k = 1, 2, … E, E represents the size of the number of output features.

[0086] Step S206, update the matrix C using the adaptive gradient descent method, and perform Gram-Schmidt orthogonalization processing on the matrix C:

[0087]

[0088] where t is the iteration number learning rate parameter μ t = v t / κ t As the iteration number t changes, 0 < v1 < 1, t = 1, …, t max , ||▽C t (:,k)|| represents the modulus of the gradient vector of the first matrix C;

[0089] Gram-Schmidt orthogonalization processing is performed on the matrix C t+1 After t max iterations, the final optimized parameter C opt is obtained.

[0090] Step S207, update the value of the objective function Q, and return to step S204 to determine whether the iteration number reaches the maximum iteration number.

[0091] After t max iterations of the matrix C, the optimized weight parameter C opt in this embodiment is obtained, and the visualization schematic diagram of the weight parameter C opt is shown in Figure 4 The objective function Q is updated according to the optimized weight parameter C opt . This embodiment directly classifies 10000 test sample sets in the MNIST data set through the feature parameters learned by the single-layer network, and the recognition accuracy reaches 98.4%, while the recognition accuracy of the best SVM method of the single-layer neural network structure is 94.5%.

[0092] The embodiment utilizes a neural cluster and an asymptotic formula of conditional mutual information, proposes a neural cluster network model and a learning algorithm based on a conditional mutual information maximization principle, and proposes a supervised learning objective function based on conditional mutual information maximization and a method for quickly optimizing model parameters, which can be applied to image recognition, natural language processing, speech recognition, signal analysis and other product and application scenarios. The learning effect and efficiency obtained by the supervised representation learning algorithm proposed in the embodiment are far superior to the effect and efficiency of other methods (such as the SVM method), which can not only learn large data samples well, but also learn small sample data well, and can significantly improve the efficiency, performance and robustness of supervised representation learning without significantly increasing the computational complexity.

[0093] Embodiment 2

[0094] According to the embodiment of the present application, a neural cluster-based data processing device embodiment is provided. Figure 5 is a schematic diagram of a neural cluster-based data processing device according to an embodiment of the present application, as shown in Figure 5 the device comprises: a transformation module 51 configured to obtain original data, and perform common spatial pattern transformation on the original data to obtain transformed data; a function obtaining module 52 configured to obtain a first target function comprising a first matrix based on the transformed data, wherein the first target function is a target function of a neural cluster network model, and the first matrix is a weight parameter of the target function of the neural cluster network model; a matrix updating module 53 configured to update the first matrix based on a preset gradient descent updating rule to obtain a second matrix; and a function updating module 54 configured to update the first target function according to the second matrix.

[0095] The above device further comprises a module for performing other method steps of the neural cluster-based data processing method in Embodiment 1.

[0096] Embodiment 3

[0097] According to the embodiment of the present application, a storage medium embodiment is provided, comprising a stored program, wherein the program controls the device where the storage medium is located to perform the above-mentioned neural cluster-based data processing method when the program is running.

[0098] According to the embodiment of the present application, a processor is provided, which is used to run a program, wherein the program performs the above-mentioned neural cluster-based data processing method when the program is running.

[0099] The above-mentioned embodiment numbers of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0100] In the above-mentioned embodiments of the present application, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0101] In several embodiments provided in the present application, it should be understood that the disclosed technology can be implemented in other manners. For example, the described unit embodiments are merely schematic, and the division of units can be different from the above. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections can be indirect couplings or communication connections through some interfaces, and electrical or other forms.

[0102] The units described as separate components can or can not be physically separate, and the components shown as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple units. Some or all of the units can be selected according to actual needs to achieve the purposes of the embodiments.

[0103] In addition, each functional unit in the various embodiments of the present application can be integrated in a processing unit, or each unit can be physically present separately, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0104] The integrated unit, if realized in the form of a software functional unit and sold or used as an independent product, can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application, essentially or the part that contributes to the prior art, or all or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes several instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a mobile hard disk, a magnetic disk or an optical disk, and various media that can store program codes.

[0105] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, several improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application.

Claims

1. A neural cluster-based data processing method, characterized in that, obtaining original data, and performing a common space pattern transformation on the original data to obtain transformed data, wherein the original data includes image data, voice data, or signal data of image recognition, natural language processing, speech recognition, and signal analysis applications; obtaining a first target function including a first matrix based on the transformed data, wherein the first target function is a target function of a neural cluster network model, and the first matrix is a weight parameter of the target function of the neural cluster network model; updating the first matrix based on a preset gradient descent update rule to obtain a second matrix; updating the first target function of the neural cluster network model according to the second matrix, and identifying and classifying an original data set by using the updated neural cluster network model; the obtaining of the original data and the performing of the common space pattern transformation on the original data to obtain the transformed data comprises: obtaining an input vector representing the original data and a neuron output vector; determining an interaction information formula according to the input vector of the original data and the neuron output vector, wherein the expression of the interaction information formula is: where p(r, x | t), p(r | t), and p(x | t) represent conditional probability density functions, <·> r,x,t denotes an expected value for the probability density function p(x, r, t), x is an input vector, t is a data label corresponding to the input vector, and r is a neuron output vector; determining a second target function including a covariance matrix and a transformation matrix, wherein the second target function is: satisfies: wherein V is the transformation matrix, the label data t has two categories, t, the covariance matrices of the two categories of data labels are Σ1 and Σ2, the covariance matrix is normalized to obtain: wherein Tr represents the trace of a matrix; obtaining the transformation matrix according to the interaction information formula and by minimizing the second target function; transforming the original data into the transformed data according to the transformation matrix.

2. The method of claim 1, wherein, the updating of the first matrix based on the preset gradient descent update rule to obtain the second matrix comprises: updating the first matrix based on the preset gradient descent update rule to obtain a third matrix; determining an iteration number, wherein the iteration number is used to represent the number of times of updating the first matrix based on the preset gradient descent update rule; determining whether the iteration number reaches a preset number of times, if yes, taking the third matrix as the second matrix and outputting, and if no, assigning the third matrix to the first matrix and returning to the step of updating the first matrix based on the preset gradient descent update rule to obtain the third matrix.

3. The method of claim 2, wherein, before the updating of the first matrix based on the preset gradient descent update rule to obtain the third matrix, the method further comprises: calculating a derivative of the first target function with respect to the first matrix.

4. The method of claim 1, wherein, the updating of the first target function according to the second matrix comprises: performing an orthogonalization transformation on the second matrix to obtain an orthogonalization result; updating the value of the first target function according to the orthogonalization result.

5. The method of claim 4, characterized in that, the orthogonalization transformation is a Gram-Schmidt orthogonalization transformation.

6. A neural cluster based data processing apparatus, characterized by the device comprises: The transformation module is configured to obtain original data, and perform a common space pattern transformation on the original data to obtain transformed data, wherein the original data includes image data, voice data, or signal data of image recognition, natural language processing, voice recognition, or signal analysis application. The function obtaining module is configured to obtain a first target function including a first matrix based on the transformed data, wherein the first target function is a target function of a neural cluster network model, and the first matrix is a weight parameter of the target function of the neural cluster network model. The matrix updating module is configured to update and orthogonalize the first matrix based on a preset gradient descent update rule to obtain a second matrix. The function updating module is configured to update the first target function of the neural cluster network model according to the second matrix, and to recognize and classify an original data set by using the updated neural cluster network model. The transformation module is further configured to obtain an input vector representing the original data and a neuron output vector. An interaction information formula is determined according to the input vector of the original data and the neuron output vector, wherein an expression of the interaction information formula is: where p(r, x | t), p(r | t), and p(x | t) represent conditional probability density functions, <·> r,x,t denotes an expected value for the probability density function p(x, r, t), x is an input vector, t is a data label corresponding to the input vector, and r is a neuron output vector; A second target function including a covariance matrix and a transformation matrix is determined, wherein the second target function is: satisfies: wherein V is the transformation matrix, the label data t has two categories, t, covariance matrices of the two categories of data labels are Σ1 and Σ2, the covariance matrix is normalized to obtain: wherein Tr represents a trace of a matrix. The transformation matrix is obtained according to the interaction information formula and by minimizing the second target function. The original data is transformed into the transformed data according to the transformation matrix.

7. A storage medium, characterized by The storage medium includes a stored program, wherein the program controls a device in which the storage medium is located to perform the neural cluster-based data processing method in any one of claims 1 to 5 when the program is running.

8. A processor, comprising: The processor is configured to run a program, wherein the program performs the neural cluster-based data processing method in any one of claims 1 to 5 when the program is running.

Images