A reserve pool data processing method and device, electronic equipment and storage medium

Abstract

Embodiments of the present application provide a reservoir pool data processing method and device, electronic equipment and storage medium. The reservoir pool data processing method comprises: in the test verification stage, obtaining input layer data; based on the random weight of the reservoir pool, performing time evolution calculation on the input layer data to generate output layer data; performing ridge regression on the input layer data to obtain input fitting value, and performing the ridge regression on the output layer data to obtain output fitting value; determining a fitting difference value according to the input fitting value and the output fitting value; when the fitting difference value is greater than a preset overfitting threshold, determining that the reservoir pool is overfitting, adjusting the random weight until the fitting difference value is not greater than the overfitting threshold. Through the embodiments of the present application, the output layer of the reservoir pool can be optimized, and the operation efficiency of the reservoir pool is improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a data processing method for a data storage pool, a data processing device for a data storage pool, an electronic device, and a storage medium. Background Technology

[0002] As the AI ​​(Artificial Intelligence) craze sweeps across all industries, "AI chips," the core of artificial intelligence, have become highly sought after. They are essential components in all smart devices, specifically designed to handle AI-related computational tasks. The AI ​​chip field is not only a battleground for semiconductor chip companies, but also attracts internet companies and cloud computing companies to announce their own chip development plans.

[0003] AI chips encompass two fields: computer science (software), which focuses on designing highly efficient intelligent algorithms, and semiconductor chips (hardware), which studies how to effectively implement these algorithms on silicon wafers, transforming them into final products that can be integrated with supporting software. Among these, the use of buffer pools has emerged as a solution for processing time-series data using dynamic systems in AI chips. However, existing buffer pool output layer overfitting determination methods are complex, leading to large errors in the output layer. This necessitates repeated processing by the buffer pool, resulting in low operational efficiency. Summary of the Invention

[0004] In view of the above problems, embodiments of the present invention are proposed to provide a reservoir data processing method, a reservoir data processing apparatus, an electronic device, and a storage medium that overcome or at least partially solve the above problems.

[0005] In a first aspect, embodiments of the present invention disclose a method for processing data from a reserve pool, comprising:

[0006] During the testing and verification phase, input layer data is acquired.

[0007] Based on the random weights of the reserve pool, the input layer data is subjected to time evolution calculation to generate output layer data;

[0008] Ridge regression is performed on the input layer data to obtain the input fitted value, and ridge regression is performed on the output layer data to obtain the output fitted value;

[0009] The fitting difference is determined based on the input fitting value and the output fitting value;

[0010] When the fitting difference is greater than a preset overfitting threshold, the reserve pool is determined to be overfitted, and the random weights are adjusted until the fitting difference is no greater than the overfitting threshold.

[0011] Optionally, the random weights of the reservoir include an input connectivity weight matrix, a reservoir connectivity weight matrix, and an output weight matrix; the step of performing time evolution calculations on the input layer data based on the random weights of the reservoir to generate output layer data includes:

[0012] Obtain the input time vector;

[0013] Determine the first product value of the time vector and the input connectivity weight matrix;

[0014] Determine the second product value between the input layer data and the reservoir connectivity weight matrix;

[0015] Based on the first product value and the first product value, determine the state vector of the storage pool;

[0016] The output layer data is generated based on the state vector of the reserve pool and the output weight matrix.

[0017] Optionally, the step of generating the output layer data based on the reservoir state vector and the output weight matrix includes:

[0018] Calculate the third product of the reserve pool state vector and the output weight matrix;

[0019] The third product value is determined to be the output layer data.

[0020] Optionally, the step of performing ridge regression on the input layer data to obtain the input fitted value includes:

[0021] The input layer data is regularized to obtain the input regularization value;

[0022] Calculate the sum of the input regularization value and the preset first regularization term, and determine the sum of the input regularization value and the preset first regularization term as the input fitted value.

[0023] Optionally, the step of performing ridge regression on the output layer data to obtain the output fitted value includes:

[0024] The output layer data is regularized to obtain the output regularized value;

[0025] Calculate the sum of the output regularization value and the preset second regularization term, and determine the sum of the output regularization value and the preset second regularization term as the output fitted value.

[0026] Optionally, the step of determining the fitting difference based on the input fitted value and the output fitted value includes:

[0027] Calculate the first difference between the input fitted value and the output fitted value;

[0028] The absolute value of the first difference is determined to be the fitting difference.

[0029] Optionally, the method further includes:

[0030] When the fitting difference is not greater than the overfitting threshold, the reserve pool is determined to have been successfully verified.

[0031] In a second aspect, embodiments of the present invention also disclose a reservoir data processing apparatus, comprising:

[0032] The acquisition module is used to acquire input layer data during the testing and verification phase.

[0033] The evolution module is used to perform time evolution calculations on the input layer data based on the random weights of the reserve pool to generate output layer data.

[0034] The fitting module is used to perform ridge regression on the input layer data to obtain input fitting values, and to perform the same ridge regression on the output layer data to obtain output fitting values.

[0035] The difference module is used to determine the fitting difference based on the input fitting value and the output fitting value;

[0036] An overfitting determination module is used to determine that the reserve pool is overfitted when the fitting difference is greater than a preset overfitting threshold, and to adjust the random weights until the fitting difference is not greater than the overfitting threshold.

[0037] In a third aspect, an embodiment of the present invention discloses an electronic device including a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program, when executed by the processor, implements the steps of the reservoir data processing method as described above.

[0038] In a fourth aspect, embodiments of the present invention disclose a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the reservoir data processing method as described above.

[0039] The embodiments of the present invention have the following advantages:

[0040] This invention, in its embodiment, acquires input layer data during the testing and verification phase; performs time evolution calculations on the input layer data based on random weights in the reserve pool to generate output layer data; performs ridge regression on the input layer data to obtain input fitted values, and performs the same ridge regression on the output layer data to obtain output fitted values; determines the fitting difference based on the input fitted values ​​and the output fitted values; and when the fitting difference exceeds a preset overfitting threshold, determines that the reserve pool is overfitted, adjusts the random weights, and continues until the fitting difference is no greater than the overfitting threshold. By using the time evolution of the reserve pool to calculate the output layer data during the testing and verification phase, and then determining the overfitting value based on ridge regression using the input and output layer data, and adjusting the random weights of the reserve pool when the fitting value is too high, the output layer of the reserve pool is optimized, improving the operating efficiency of the reserve pool. Attached Figure Description

[0041] Figure 1 This is a flowchart illustrating the steps of an embodiment of a data processing method for a storage pool according to the present invention.

[0042] Figure 2 This is a flowchart illustrating the steps of another embodiment of the data processing method for the storage pool of the present invention;

[0043] Figure 3 This is a schematic diagram illustrating the execution of an example of a data processing method for a reserve pool according to the present invention;

[0044] Figure 4 This is a structural block diagram of an embodiment of a data processing device for a storage pool according to the present invention. Detailed Implementation

[0045] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0046] A reservoir is a randomly generated, large-scale, sparsely connected (typically maintaining 1%-5% connections) recursive structure. Originally defined as a lightweight method for training recurrent neural networks (RNNs), it has evolved into a method for computationally processing time-series data using dynamic systems. The readout layer is typically a linear combination of reservoir states. Supervised learning techniques are used to determine readout weights, where the network is driven by teacher sample inputs and its output is compared to the corresponding teacher data output to estimate error. Weights can then be computed using any closed-form regression technique during offline training, or adaptive techniques can be used for online training. However, calculating the weights does not preclude determining overfitting in the reservoir output layer. This leads to complex methods for overfitting the reservoir output layer, large errors in the reservoir output layer, and the need for repeated runs to obtain satisfactory results, resulting in low operational efficiency.

[0047] One of the core concepts of this invention is to first perform time evolution on the input layer data based on the random weights of the reserve pool to generate output layer data, and then use the ridge regression algorithm to process the output layer data to obtain the output fitted value. Next, the input layer data is processed to obtain the input fitted value. When the difference is greater than a preset overfitting threshold, overfitting is determined, and the random weights of the reserve pool are adjusted.

[0048] Reference Figure 1 The diagram illustrates a flowchart of an embodiment of a data processing method for a reserve pool according to the present invention. The data processing method for the reserve pool may specifically include the following steps:

[0049] Step 101: In the testing and verification phase, obtain the input layer data;

[0050] The neural network model built on the reservoir will be trained using a training set. Once the neural network model built on the reservoir is trained, it will enter the testing and verification phase. In the testing and verification phase, the neural network model built on the reservoir will be verified to determine whether it meets the usage requirements.

[0051] During the testing and verification phase, the verification dataset, i.e., the input layer data, is obtained from the preset storage address. The preset storage address can be a storage address (range) corresponding to the device's own storage space, a storage address (range) corresponding to the storage space of a third-party database connected to the device, or a storage address corresponding to the cloud storage space connected to the device.

[0052] Step 102: Based on the random weights of the reserve pool, perform time evolution calculation on the input layer data to generate output layer data;

[0053] Since the nodes within the reserve pool are fixed, the input layer data can be used as input based on the random weights between each node in the reserve pool. The time evolution of the input layer data can be calculated, and the dynamics of the input layer data can be analyzed and calculated to generate the output layer data of the input layer under dynamic conditions after time evolution.

[0054] Step 103: Perform ridge regression on the input layer data to obtain the input fitted value, and perform the ridge regression on the output layer data to obtain the output fitted value;

[0055] After obtaining the output layer data, the output layer data and input layer data can be fitted to determine the corresponding fitted values. Therefore, ridge regression can be performed on the input layer data, and the weights in the input layer data can be regressed to determine the input fitted values. Correspondingly, ridge regression can be performed on the output layer data, and the weights in the output layer data can be regressed to determine the output fitted values. The input and output fitted values ​​can be expressed as percentages or fractions, and those skilled in the art can set other representation methods according to needs; this embodiment of the invention does not limit this. It should be noted that the input and output fitted values ​​must use the same representation method.

[0056] Step 104: Determine the fitting difference based on the input fitting value and the output fitting value;

[0057] After obtaining the input and output fitted values, the fitting difference between them can be determined. This fitting difference characterizes the difference between the fitted values ​​and the output fitted values. The fitting difference uses the same representation as the input and output fitted values. If both the input and output fitted values ​​are expressed as percentages, the fitting difference should also be expressed as a percentage.

[0058] Step 105: When the fitting difference is greater than the preset overfitting threshold, the reserve pool is determined to be overfitted, and the random weights are adjusted until the fitting difference is not greater than the overfitting threshold.

[0059] When the fitting difference exceeds the preset overfit threshold, indicating a significant discrepancy between the training and validation results of the neural network model built based on the reserve pool, it can be determined that the reserve pool is overfitting. In this case, the random weight determination method for the reserve pool can be adjusted, and the above steps can be repeated for testing and validation until the fitting difference is no greater than the overfit threshold, at which point the testing ends.

[0060] The preset fitting threshold should be represented in the same way as the fitting difference, i.e., the sum of the fitting difference, input fitting value, and output fitting value. The preset fitting threshold should use the same representation. The size of the preset fitting threshold can be determined according to the actual situation, and this embodiment of the invention does not limit it.

[0061] In one example of the present invention, the preset fitting threshold is 10%, which allows the reserve pool to quickly adjust the random weights so that the error of the reserve pool can meet the usage requirements.

[0062] This invention, in its embodiment, acquires input layer data during the testing and verification phase; performs time evolution calculations on the input layer data based on random weights in the reserve pool to generate output layer data; performs ridge regression on the input layer data to obtain input fitted values, and performs the same ridge regression on the output layer data to obtain output fitted values; determines the fitting difference based on the input fitted values ​​and the output fitted values; and when the fitting difference exceeds a preset overfitting threshold, determines that the reserve pool is overfitted, adjusts the random weights, and continues until the fitting difference is no greater than the overfitting threshold. By using the time evolution of the reserve pool to calculate the output layer data during the testing and verification phase, and then determining the overfitting value based on ridge regression using the input and output layer data, and adjusting the random weights of the reserve pool when the fitting value is too high, the output layer of the reserve pool is optimized, improving the operating efficiency of the reserve pool.

[0063] Reference Figure 2 The diagram illustrates a flowchart of another embodiment of the data processing method for a reserve pool according to the present invention. The data processing method for a reserve pool may specifically include the following steps:

[0064] Step 201: In the testing and verification phase, obtain the input layer data;

[0065] When the neural network model built based on the reserve pool completes training and enters the testing and verification phase, the input layer data for verifying the neural network model is obtained. The input layer data is the test dataset used for verifying the neural network model.

[0066] It can be obtained from the storage space address (range) where the input layer data is stored.

[0067] Step 202: Based on the random weights of the reserve pool, perform time evolution calculation on the input layer data to generate output layer data;

[0068] Once the input layer data enters the storage pool, random weights are used between the computation nodes within the pool, and the nodes are interconnected using random sparse connectivity. The connection between the input and the storage pool is fixed after initialization. Based on the random weights in the storage pool, time evolution calculations can be performed on the input layer data, introducing time vectors to analyze the dynamic system and generate output layer data.

[0069] In one embodiment of the present invention, the random weights of the reservoir include an input connectivity weight matrix, a reservoir connectivity weight matrix, and an output weight matrix; the step of performing time evolution calculations on the input layer data based on the random weights of the reservoir to generate output layer data specifically includes the following sub-steps:

[0070] Sub-step S2021: Obtain the input time vector;

[0071] In this embodiment of the invention, the random weights of the reserve pool may include the input connectivity weight matrix W. in The connectivity weight matrix W of the reservoir res and output weight matrix W out The input connectivity weight matrix is ​​used as the time random weight, the reservoir connectivity weight matrix is ​​used as the calculated random weight of each node in the input process, and the output weight matrix is ​​used as the random weight of each node in the output layer.

[0072] When performing time evolution calculations, an input time vector can be obtained from the reservoir, and this input time vector can be used to describe the dynamics.

[0073] Sub-step S2022: Determine the first product value of the time vector and the input connectivity weight matrix;

[0074] In this embodiment of the invention, the first product value of the time vector and the input connectivity weight matrix can be determined. That is, the first product value is obtained by multiplying the time vector by the input connectivity weight matrix.

[0075] Sub-step S2023: Determine the second product value of the input layer data and the reservoir connectivity weight matrix;

[0076] In this embodiment of the invention, a second product value is determined by multiplying the input layer data and the reservoir connectivity weight matrix. This product is then used to calculate the value for each computation node in the input layer.

[0077] Sub-step S2024: Determine the state vector of the storage pool based on the first product value and the first product value;

[0078] After obtaining the first product value and the first product value, the transfer function of the reservoir in the reservoir layer is used. Based on the first product value and the first product value, the dependent variable output by the transfer function is the reservoir state vector.

[0079] Sub-step S2025: Generate the output layer data based on the reserve pool state vector and the output weight matrix.

[0080] The state vector of the reservoir is combined with the output weight matrix to process the computation nodes of the output layer and generate the output layer data.

[0081] Specifically, the step of generating the output layer data based on the reserve pool state vector and the output weight matrix may include the following sub-steps:

[0082] Sub-step S20251: Calculate the third product value of the reserve pool state vector and the output weight matrix;

[0083] In practical applications, the reservoir state vector can be multiplied by the output weight matrix to calculate the third product value, which is the product of the reservoir state vector and the output weight matrix.

[0084] Sub-step S20252: Determine the third product value as the output layer data.

[0085] After obtaining the third product value, it is determined that the third product value is the output layer data.

[0086] In summary, sub-steps S2021 to S2025 can be described using the following formula:

[0087] x(t+1)=f(W res ·x(t)+W in ·u(t))

[0088] y(t)=W out ·x′(t)

[0089] Among them, W in The input is the connectivity weight matrix;

[0090] W res This is the connectivity weight matrix for the reserve pool;

[0091] W out To output the weight matrix;

[0092] f is the transfer function of the reservoir node, which is usually hyperbolic tangent, but sigmoid or linear functions can also be used;

[0093] x ′ (t) represents the constant 1 extended reservoir state vector;

[0094] x(t) represents the input layer data;

[0095] u(t) is the time vector.

[0096] Step 203: Perform ridge regression on the input layer data to obtain the input fitted value, and perform the ridge regression on the output layer data to obtain the output fitted value;

[0097] After obtaining the output layer data, regression can be performed on the input and output layer data. Ridge regression is performed on the input layer data to obtain the input fit value, and ridge regression is performed on the output layer data to obtain the output fit value. By analyzing the input and output layer data separately, the training status of the neural network model built in the reserve pool can be determined.

[0098] In practical applications, ridge regression can be used to construct weighted regression models to prevent overfitting of the pool output data to the simulated input data during machine learning.

[0099] Specifically, the step of performing ridge regression on the input layer data to obtain the input fitted value may include the following sub-steps:

[0100] Sub-step S2031: Regularize the input layer data to obtain input regular values;

[0101] When performing ridge regression on the input layer data, the input layer data can be regularized first to obtain the input regularization value. This allows the input layer data to filter out noise caused by direction, making the input fitting result more accurate.

[0102] Sub-step S2032: Calculate the sum of the input regularization value and the preset first regularization term, and determine the sum of the input regularization value and the preset first regularization term as the input fitted value.

[0103] After obtaining the input regularization value, the sum of the input regularization value and the preset first regularization term can be calculated. This sum is then determined as the input fitted value.

[0104] It should be noted that the first regularization term is the compensation value corresponding to the penalty loss function applied to the input layer data during the regression process. Those skilled in the art can determine this based on the actual calculation process; this embodiment of the invention does not impose any limitations on it.

[0105] Specifically, the step of performing ridge regression on the output layer data to obtain the output fitted value may include the following sub-steps:

[0106] Sub-step S2033: Regularize the output layer data to obtain the output regularization value;

[0107] Accordingly, when performing ridge regression on the output layer data, the output layer data can first be regularized to obtain the output regularization value. This allows the output layer data to filter out noise caused by direction, making the output fitting result more accurate.

[0108] Sub-step S2034: Calculate the sum of the output regularization value and the preset second regularization term, and determine the sum of the output regularization value and the preset second regularization term as the output fitted value.

[0109] After obtaining the input regularization value, the sum of the input regularization value and the preset first regularization term can be calculated. This sum is then determined as the input fitted value.

[0110] It should be noted that the first regularization term is the compensation value corresponding to the penalty loss function applied to the input layer data during the regression process. Those skilled in the art can determine this based on the actual calculation process; this embodiment of the invention does not impose any limitations on it.

[0111] In summary, sub-steps S2031 to S2034 employ the ridge regression method to construct a weighted regression model, i.e.:

[0112] ||Xθ-y|| 2 -||rθ|| 2

[0113] r = aI

[0114] θ(a)=(X T X+aI) -1 X T y

[0115] Where X = input;

[0116] y = Output;

[0117] || = regular expression;

[0118] θ is the fitting hyperparameter;

[0119] T is a weight constant;

[0120] a is the weight of the identity matrix;

[0121] θ(a) is used to find θ when a is determined;

[0122] l is the identity matrix.

[0123] Step 204: Determine the fitting difference based on the input fitting value and the output fitting value;

[0124] After obtaining the input fit values ​​of the input layer and the output fit values ​​of the output layer, the fitting difference can be calculated based on the input fit values ​​and the output fit values. The fitting difference can be used to determine whether overfitting exists.

[0125] Specifically, the step of determining the fitting difference based on the input fitting value and the output fitting value may include the following sub-steps:

[0126] Sub-step S2041: Calculate the first difference between the input fitted value and the output fitted value;

[0127] In practical applications, the first difference between the input fitted value and the output fitted value can be calculated. That is, the first difference can be obtained by subtracting the output fitted value from the input fitted value, or by subtracting the input fitted value from the output fitted value.

[0128] Sub-step S2042: Determine the absolute value of the first difference as the fitted difference.

[0129] Calculate the absolute value of the first difference to avoid errors caused by the direction of the input fitted value and the output fitted value; determine the absolute value of the first difference as the fitted difference.

[0130] Step 205: When the fitting difference is greater than the preset overfitting threshold, the reserve pool is determined to be overfitted, and the random weights are adjusted until the fitting difference is not greater than the overfitting threshold.

[0131] In this embodiment of the invention, the fitting difference can be compared with a preset overfitting threshold. When the fitting difference is greater than the preset overfitting threshold, it can be determined that the reserve pool is overfitted and needs to be adjusted. The random weight determination method of the reserve pool can be adjusted until the fitting difference is not greater than the overfitting threshold.

[0132] Step 206: When the fitting difference is not greater than the overfitting threshold, the reserve pool is determined to be successfully verified.

[0133] When the fitting difference is not greater than the overfitting threshold, it indicates that there is no overfitting in the reserve pool, the error of the neural network model formed by the reserve pool is within the usable range, the reserve pool is successfully verified, and the reserve pool can be used.

[0134] This invention, in its verification phase, acquires input layer data; performs time evolution calculations on the input layer data based on random weights in a reserve pool to generate output layer data; performs ridge regression on the input layer data to obtain input fitted values, and performs the same ridge regression on the output layer data to obtain output fitted values; determines the fitting difference based on the input and output fitted values; when the fitting difference exceeds a preset overfitting threshold, the reserve pool is determined to be overfitted, and the random weights are adjusted until the fitting difference is no greater than the overfitting threshold; when the fitting difference is no greater than the overfitting threshold, the reserve pool is deemed successfully verified. By using the time evolution of the reserve pool to calculate the output layer data during the verification phase, and then determining the overfitting value based on ridge regression using the input and output layer data, the random weights of the reserve pool are adjusted when the fitting value is too high, thus optimizing the output layer of the reserve pool and improving its operating efficiency. Furthermore, determining successful verification of the reserve pool when its overfitting value meets the requirements facilitates the determination of the reserve pool's verification status.

[0135] To enable those skilled in the art to better understand the embodiments of the present invention, an example is provided below to illustrate the embodiments of the present invention:

[0136] Reference Figure 3 The diagram illustrates an example of a data processing method for a storage pool according to the present invention.

[0137] In this example, the neural network model built on the reservoir consists of three parts: an input layer, a reservoir layer, and an output layer.

[0138] First, obtain the input layer data;

[0139] The input layer data in the reservoir undergoes time evolution calculations across multiple batches. Once the input layer data enters the reservoir, random weights are used between computation nodes, and nodes are interconnected using random sparse connectivity. The connections between the input layer and the reservoir are fixed after initialization and typically require no modification. Output layer data is obtained through calculations in the input layer and the reservoir.

[0140] Before performing ridge regression, a weighted regression model can be constructed based on the ridge regression method. The weighted regression model can be used to predict whether the output data of the reserve pool is overfitted to the simulated input data during the machine learning process.

[0141] The input layer data is fed into the weighted regression model to obtain the input fitted values;

[0142] The output layer data is input into the weighted regression model to obtain the output fitted value;

[0143] Determine whether the difference between the input fitted value and the output fitted value is greater than 10% (preset overfit threshold).

[0144] If the difference is greater than 10%, then overfitting is determined, and the nodes in the pool calculation need to be changed using random weights.

[0145] It should be noted that, for the sake of simplicity, the method embodiments are all described as a series of actions. However, those skilled in the art should understand that the embodiments of the present invention are not limited to the described order of actions, because according to the embodiments of the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions involved are not necessarily essential to the embodiments of the present invention.

[0146] Reference Figure 4 The diagram illustrates a structural block diagram of an embodiment of a data processing device for a reservoir according to the present invention. Specifically, the data processing device for a reservoir may include the following modules:

[0147] The acquisition module 401 is used to acquire input layer data during the testing and verification phase.

[0148] Evolution module 402 is used to perform time evolution calculation on the input layer data based on the random weights of the reserve pool to generate output layer data;

[0149] The fitting module 403 is used to perform ridge regression on the input layer data to obtain input fitting values, and to perform the ridge regression on the output layer data to obtain output fitting values.

[0150] Difference module 404 is used to determine the fitting difference based on the input fitting value and the output fitting value;

[0151] The overfitting determination module 405 is used to determine that the reserve pool is overfitted when the fitting difference is greater than a preset overfitting threshold, and to adjust the random weights until the fitting difference is not greater than the overfitting threshold.

[0152] In an optional embodiment of the present invention, the random weights of the reservoir include an input connectivity weight matrix, a reservoir connectivity weight matrix, and an output weight matrix; the evolution module 402 includes:

[0153] The first acquisition submodule is used to acquire the input time vector;

[0154] The first determining submodule is used to determine the first product value of the time vector and the input connectivity weight matrix;

[0155] The second determining submodule is used to determine the second product value of the input layer data and the reservoir connectivity weight matrix;

[0156] The third determining submodule is used to determine the state vector of the storage pool based on the first product value and the first product value;

[0157] The first output submodule is used to generate the output layer data based on the reserve pool state vector and the output weight matrix.

[0158] In an optional embodiment of the present invention, the first output submodule includes:

[0159] A calculation unit is used to calculate the third product value of the reserve pool state vector and the output weight matrix;

[0160] An output unit is used to determine that the third product value is the output layer data.

[0161] In an optional embodiment of the present invention, the fitting module 403 includes:

[0162] The first regularization submodule is used to regularize the input layer data to obtain input regularization values;

[0163] The first calculation submodule is used to calculate the sum of the input regularization value and the preset first regularization term, and to determine the sum of the input regularization value and the preset first regularization term as the input fitted value.

[0164] In an optional embodiment of the present invention, the fitting module 403 further includes:

[0165] The second regularization submodule is used to regularize the output layer data to obtain output regularization values;

[0166] The second calculation submodule is used to calculate the sum of the output regularization value and the preset second regularization term, and to determine the sum of the output regularization value and the preset second regularization term as the output fitted value.

[0167] In an optional embodiment of the present invention, the difference module 404 includes:

[0168] The third calculation submodule is used to calculate the first difference between the input fitted value and the output fitted value;

[0169] The fourth determining submodule is used to determine that the absolute value of the first difference is the fitted difference.

[0170] In an optional embodiment of the present invention, the device further includes:

[0171] The verification module is used to determine that the reserve pool has been successfully verified when the fitting difference is not greater than the overfitting threshold.

[0172] In this embodiment of the invention, the output layer data is calculated by utilizing the time evolution of the reservoir during the testing and verification phase. Then, the overfitting value is determined by ridge regression based on the input layer data and the output layer data. When the overfitting value is too high, the random weights of the reservoir are adjusted, thereby optimizing the output layer of the reservoir and improving the operating efficiency of the reservoir.

[0173] As the device embodiment is basically similar to the method embodiment, the description is relatively simple, and relevant parts can be found in the description of the method embodiment.

[0174] This invention also provides an electronic device, comprising:

[0175] The device includes a processor and a storage medium, the storage medium storing a computer program executable by the processor. When the electronic device is running, the processor executes the computer program to perform the method described in any of the embodiments of the present invention. The specific implementation and technical effects are similar to those in the method embodiments, and will not be repeated here.

[0176] The storage medium may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the storage medium may also be at least one storage device located remotely from the aforementioned processor.

[0177] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0178] This invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, performs the method described in any one of the embodiments of this invention. The specific implementation and technical effects are similar to those in the method embodiments, and will not be repeated here.

[0179] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0180] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, apparatus, or computer program products. Therefore, embodiments of the present invention can take the form of entirely hardware embodiments, entirely software embodiments, or embodiments combining software and hardware aspects. Furthermore, embodiments of the present invention can take the form of computer program products implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0181] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, terminal devices (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0182] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0183] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0184] Although preferred embodiments of the present invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present invention.

[0185] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.

[0186] The above provides a detailed description of the data processing method, apparatus, electronic device, and storage medium for a data storage pool provided by the present invention. Specific examples have been used to illustrate the principles and implementation methods of the present invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of the present invention. At the same time, those skilled in the art will recognize that there will be changes in the specific implementation methods and application scope based on the ideas of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A data processing method for a reserve pool, characterized in that, include: During the testing and verification phase, input layer data is acquired. Based on the random weights of the reserve pool, the input layer data is subjected to time evolution calculation to generate output layer data; Ridge regression is performed on the input layer data to obtain the input fitted value, and ridge regression is performed on the output layer data to obtain the output fitted value; The fitting difference is determined based on the input fitting value and the output fitting value; When the fitting difference is greater than a preset overfitting threshold, the reserve pool is determined to be overfitted, and the random weights are adjusted until the fitting difference is no greater than the overfitting threshold.

2. The method according to claim 1, characterized in that, The random weights of the reservoir include an input connectivity weight matrix, a reservoir connectivity weight matrix, and an output weight matrix; the step of performing time evolution calculations on the input layer data based on the random weights of the reservoir to generate output layer data includes: Obtain the input time vector; Determine the first product value of the time vector and the input connectivity weight matrix; Determine the second product value between the input layer data and the reservoir connectivity weight matrix; Based on the first product value and the first product value, determine the state vector of the storage pool; The output layer data is generated based on the state vector of the reserve pool and the output weight matrix.

3. The method according to claim 2, characterized in that, The step of generating the output layer data based on the reserve pool state vector and the output weight matrix includes: Calculate the third product of the reserve pool state vector and the output weight matrix; The third product value is determined to be the output layer data.

4. The method according to claim 1, characterized in that, The step of performing ridge regression on the input layer data to obtain the input fitted value includes: The input layer data is regularized to obtain the input regularization value; Calculate the sum of the input regularization value and the preset first regularization term, and determine the sum of the input regularization value and the preset first regularization term as the input fitted value.

5. The method according to claim 4, characterized in that, The step of performing ridge regression on the output layer data to obtain the output fitted value includes: The output layer data is regularized to obtain the output regularized value; Calculate the sum of the output regularization value and the preset second regularization term, and determine the sum of the output regularization value and the preset second regularization term as the output fitted value.

6. The method according to claim 1, characterized in that, The step of determining the fitting difference based on the input fitting value and the output fitting value includes: Calculate the first difference between the input fitted value and the output fitted value; The absolute value of the first difference is determined to be the fitting difference.

7. The method according to claim 1, characterized in that, The method further includes: When the fitting difference is not greater than the overfitting threshold, the reserve pool is determined to have been successfully verified.

8. A data processing device for a storage pool, characterized in that, include: The acquisition module is used to acquire input layer data during the testing and verification phase. The evolution module is used to perform time evolution calculations on the input layer data based on the random weights of the reserve pool to generate output layer data. The fitting module is used to perform ridge regression on the input layer data to obtain input fitting values, and to perform the same ridge regression on the output layer data to obtain output fitting values. The difference module is used to determine the fitting difference based on the input fitting value and the output fitting value; An overfitting determination module is used to determine that the reserve pool is overfitted when the fitting difference is greater than a preset overfitting threshold, and to adjust the random weights until the fitting difference is not greater than the overfitting threshold.

9. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and capable of running on the processor, wherein the computer program, when executed by the processor, implements the steps of the reservoir data processing method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, A computer program is stored on the computer-readable storage medium, which, when executed by a processor, implements the steps of the reservoir data processing method as described in any one of claims 1 to 7.

Images