CN115204356A - Data processing method and device based on pulse rearrangement deep residual neural network - Google Patents

Abstract

The present application relates to the field of pulse neural networks and data processing technologies, and more particularly, to a data processing method and apparatus based on a pulse rearrangement depth residual error neural network. The method comprises the following steps: configuring a plurality of immediately adjacent pulse rearrangement residual modules in a pulse rearrangement-based depth residual neural network; taking the pulse rearrangement-based depth residual error neural network after the pulse rearrangement residual block is configured as a first pulse rearrangement model; training the first pulse rearrangement model; and inputting the target data into the trained first pulse rearrangement model for processing to obtain a target processing result. The rearrangement processing greatly reduces the number of parameters, reduces the over-fitting risk, also reduces the storage and calculation overhead, and further improves the data processing efficiency. Meanwhile, the method and the device can obtain the category corresponding to the target data, and can obtain a regression sequence or a regression single vector, so that the method and the device are suitable for various application scenarios.

Description

Data processing method and device based on pulse rearrangement deep residual neural network

technical field

The present application relates to the technical field of spiking neural networks and data processing, and more particularly, the present application relates to a data processing method and apparatus based on a spiking rearrangement deep residual neural network.

Background technique

Artificial Neural Network (ANN) has made breakthroughs in many fields thanks to deep learning. The network depth has a great impact on the performance of the network, and deeper network structures have also been proposed one after another, and have better performance than shallow networks. In deep ANN, the most common structure is residual connections, which successfully solves the training problem of deep ANN.

Spiking Neural Network (SNN) is known as the third-generation neural network, which has the advantages of event-driven and low power consumption. However, since SNN uses discrete pulses for communication, the loss of information is relatively large. To improve the performance of SNNs, a natural idea is to use a residual network-like structure to gain performance gains by increasing depth. However, directly using the traditional residual network structure in ANN, the obtained SNN performance is still poor.

SUMMARY OF THE INVENTION

Based on the above technical problems, the present invention aims to configure a plurality of adjacent pulse rearrangement residual modules based on the pulse rearrangement deep residual neural network, and configure the pulse rearrangement based deep residual neural network after configuring the pulse rearrangement residual block. As the first pulse rearrangement model, the target data is processed using the first pulse rearrangement model.

A first aspect of the present invention provides a data processing method based on a pulse rearrangement deep residual neural network, the method comprising:

Configure multiple adjacent pulse rearrangement residual modules in a deep residual neural network based on pulse rearrangement;

Using the pulse rearrangement-based deep residual neural network after configuring the pulse rearrangement residual block as the first pulse rearrangement model;

training the first pulse rearrangement model;

Input the target data into the trained first pulse rearrangement model for processing, and obtain the target processing result.

In some embodiments of the present invention, the first pulse rearrangement model is further provided with a first convolution layer, a pooling layer and a fully connected layer; the target data is input into the trained first pulse rearrangement model for processing include:

input the target data into the first convolutional layer for downsampling;

inputting the down-sampled target data into the plurality of adjacent pulse rearrangement residual modules to perform pulse rearrangement residual processing to obtain a first processing result;

The first processing result is sequentially input to the pooling layer and the fully connected layer to obtain a second processing result.

In some embodiments of the present invention, the obtaining the target processing result includes:

Obtain the category corresponding to the target data according to the second processing result;

obtaining a regression sequence and/or a regression single vector based on the second processing result;

The category corresponding to the target data, the regression sequence and/or the regression single vector are used as the target processing result.

In some embodiments of the present invention, the spike rearrangement residual module is sequentially configured with a spike rearrangement layer, a second convolution layer, a normalization layer, a spike neuron layer, and a spike anti-rearrangement layer; the downsampling The latter target data is input into the plurality of adjacent pulse rearrangement residual modules for pulse rearrangement residual processing, and a first processing result is obtained, including:

Input the down-sampled target data into the pulse rearrangement layer to obtain the pulse rearrangement result;

The pulse rearrangement result is sequentially processed by the second convolution layer, the normalization layer, the spiking neuron layer and the pulse anti-rearrangement layer to obtain the first processing result.

In some embodiments of the present invention, the down-sampled target data is input into the pulse rearrangement layer to obtain a pulse rearrangement result, and the formula is:

表示向下取整。Among them, Y represents the rearrangement operation performed by the pulse rearrangement layer, X represents the down-sampled target data; n, z, y, and x represent the batch size, channel number, height and width of the down-sampled target data in turn; M represents the total number of channels, r represents the rearrangement coefficient, % represents the remainder,

Indicates rounded down.

In some embodiments of the present invention, the first pulse rearrangement residual module in the plurality of immediately adjacent pulse rearrangement residual modules is further configured with a downsampling function.

In some embodiments of the present invention, before the target data is input into the trained first pulse rearrangement model for processing and the target processing result is obtained, the method further includes:

Get the data to be entered;

If the input data is a single number, repeat the single number for a preset number of times to obtain a sequence with a preset length; if the input data is a sequence with a length of T composed of multiple numbers, where T is greater than 1, no need to repeat;

The sequence with the preset length and the sequence with the length T are used as target data.

A second aspect of the present invention provides a data processing device based on a pulse rearrangement deep residual neural network, the device comprising:

a configuration module for configuring a plurality of adjacent pulse rearrangement residual modules in a deep residual neural network based on pulse rearrangement;

The rearrangement module is used to use the pulse rearrangement-based deep residual neural network after configuring the pulse rearrangement residual block as the first pulse rearrangement model;

a training module for training the first pulse rearrangement model;

The processing module is used for inputting the target data into the trained first pulse rearrangement model for processing to obtain the target processing result.

A third aspect of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor running the computer program to implement the following steps :

Configure multiple adjacent pulse rearrangement residual modules in a deep residual neural network based on pulse rearrangement;

Using the pulse rearrangement-based deep residual neural network after configuring the pulse rearrangement residual block as the first pulse rearrangement model;

training the first pulse rearrangement model;

Input the target data into the trained first pulse rearrangement model for processing, and obtain the target processing result.

A fourth aspect of the present invention provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

Configure multiple adjacent pulse rearrangement residual modules in a deep residual neural network based on pulse rearrangement;

Using the pulse rearrangement-based deep residual neural network after configuring the pulse rearrangement residual block as the first pulse rearrangement model;

training the first pulse rearrangement model;

Input the target data into the trained first pulse rearrangement model for processing, and obtain the target processing result.

The technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages:

In this application, a plurality of adjacent pulse rearrangement residual modules are first configured in the pulse rearrangement-based deep residual neural network, and the pulse rearrangement-based deep residual neural network after configuring the pulse rearrangement residual blocks is used as the first pulse rearrangement. Arrangement model, train the first pulse rearrangement model, input the target data into the trained first pulse rearrangement model for processing, and obtain the target processing result, and the rearrangement processing greatly reduces the amount of parameters and reduces overfitting It also reduces the storage and computing overhead, thereby improving the efficiency of data processing. At the same time, the present application can obtain the category corresponding to the target data, and obtain a regression sequence or a single regression vector, which is suitable for various application scenarios.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention.

Description of drawings

Various other advantages and benefits will become apparent to those of ordinary skill in the art upon reading the following detailed description of the preferred embodiments. The drawings are for purposes of illustrating preferred embodiments only and are not to be considered limiting of the application. Also, the same components are denoted by the same reference numerals throughout the drawings. In the attached image:

1 shows a schematic diagram of steps of a data processing method based on a pulse rearrangement deep residual neural network in an exemplary embodiment of the present application;

FIG. 2 shows a flowchart of a data processing method based on a pulse rearrangement deep residual neural network in an exemplary embodiment of the present application;

FIG. 3 shows a schematic diagram of the pulse matrix rearrangement and anti-rearrangement process in an exemplary embodiment of the present application;

FIG. 4 shows a schematic structural diagram of a data processing apparatus based on a pulse rearrangement deep residual neural network in an exemplary embodiment of the present application;

FIG. 5 shows a schematic structural diagram of a computer device provided by an exemplary embodiment of the present application;

FIG. 6 shows a schematic diagram of a storage medium provided by an exemplary embodiment of the present application.

Detailed ways

Hereinafter, embodiments of the present application will be described with reference to the accompanying drawings. It should be understood, however, that these descriptions are exemplary only, and are not intended to limit the scope of the application. Also, in the following description, descriptions of well-known structures and techniques are omitted to avoid unnecessarily obscuring the concepts of the present application. It will be apparent to those skilled in the art that the present application may be practiced without one or more of these details. In other instances, some technical features known in the art have not been described in order to avoid confusion with the present application.

It should be noted that the terminology used herein is for the purpose of describing specific embodiments only, and is not intended to limit the exemplary embodiments in accordance with the present application. As used herein, the singular forms are also intended to include the plural forms unless the context clearly dictates otherwise. Furthermore, it should also be understood that when the terms "comprising" and/or "comprising" are used in this specification, they indicate the presence of stated features, integers, steps, operations, elements and/or components, but do not exclude the presence or Addition of one or more other features, integers, steps, operations, elements, components and/or combinations thereof.

Now, exemplary embodiments according to the present application will be described in more detail with reference to the accompanying drawings. These exemplary embodiments may, however, be embodied in many different forms and should not be construed as limited to the embodiments set forth herein. The drawings are not to scale, some details may be exaggerated and some details may be omitted for clarity. The shapes of the various regions and layers shown in the figures, as well as their relative sizes and positional relationships are only exemplary, and in practice, there may be deviations due to manufacturing tolerances or technical limitations, and those skilled in the art should Regions/layers with different shapes, sizes, relative positions can be additionally designed as desired.

Exemplary embodiments according to the present application are described below with reference to the accompanying drawings 1 to 6 of the specification. It should be noted that the following application scenarios are only shown to facilitate understanding of the spirit and principles of the present application, and the embodiments of the present application are not limited in this respect. Rather, the embodiments of the present application can be applied to any scenario where applicable.

Example 1:

This embodiment provides a data processing method based on a pulse rearrangement deep residual neural network, as shown in FIG. 1 , the method includes:

S1. Configure a plurality of adjacent pulse rearrangement residual modules in a deep residual neural network based on pulse rearrangement;

S2, using the pulse rearrangement-based deep residual neural network after configuring the pulse rearrangement residual block as the first pulse rearrangement model;

S3, training the first pulse rearrangement model;

S4, input the target data into the trained first pulse rearrangement model for processing, and obtain the target processing result.

In a specific implementation, referring to FIG. 2 , the first pulse rearrangement model is further provided with a first convolution layer, a pooling layer and a fully connected layer; the target data is input into the trained first pulse rearrangement model. The processing includes: inputting the target data into the first convolutional layer for downsampling; inputting the downsampled target data into a plurality of adjacent pulse rearrangement residual modules for pulse rearrangement residual processing to obtain a first processing result; The first processing result is sequentially input to the pooling layer and the fully connected layer to obtain the second processing result.

In a specific implementation manner, obtaining the target processing result includes: obtaining a category corresponding to the target data according to the second processing result; obtaining a regression sequence and/or a regression single vector based on the second processing result; Regress a sequence and/or regress a single vector as the target processing result.

In a specific implementation manner, the pulse rearrangement residual module is sequentially configured with a pulse rearrangement layer, a second convolution layer, a normalization layer, a pulse neuron layer and a pulse anti-rearrangement layer; the down-sampled target data is Inputting a plurality of adjacent pulse rearrangement residual modules to perform pulse rearrangement residual processing to obtain a first processing result, including: inputting the down-sampled target data into the pulse rearrangement layer to obtain a pulse rearrangement result; The result is sequentially processed by the second convolution layer, the normalization layer, the spiking neuron layer and the spiking anti-rearrangement layer to obtain the first processing result.

In a specific implementation manner, the down-sampled target data is input into the pulse rearrangement layer, and the pulse rearrangement result is obtained, which is represented by formula (1):

表示向下取整。Among them, Y represents the rearrangement operation performed by the pulse rearrangement layer, X represents the down-sampled target data; n, z, y, and x represent the batch size, channel number, height and width of the down-sampled target data in turn; M represents the total number of channels, r represents the rearrangement coefficient, % represents the remainder,

Indicates rounded down.

In a specific implementation manner, the first pulse rearrangement residual module among the plurality of immediately adjacent pulse rearrangement residual modules is further configured with a downsampling function.

In a specific implementation manner, before inputting the target data into the trained first pulse rearrangement model for processing, and obtaining the target processing result, the method further includes: acquiring the data to be input; if the data to be input is a single number, Repeat the single number for a preset number of times to obtain a sequence with a preset length; if the input data is a sequence with a length of T composed of multiple numbers, where T is greater than 1, there is no need to repeat; set the length to the preset length A sequence of and a sequence of length T are used as target data.

Example 2:

This embodiment provides a data processing method based on a pulse rearrangement deep residual neural network, and the steps included in the method are described in detail below.

The first step is to configure multiple adjacent pulse rearrangement residual modules in the deep residual neural network based on pulse rearrangement.

In a specific implementation, as shown in Figure 2, a deep residual neural network based on pulse rearrangement is provided with a first convolution layer, a pooling layer and a fully connected layer, and the pooling layer is pooled to 1*1 size, and then fully connected through the fully connected layer. The spike rearrangement residual module (referred to as the spike rearrangement element-wise residual block in Figure 2) is sequentially configured with a spike rearrangement layer, a second convolutional layer, a normalization layer, a spike neuron layer, and a spike anti-rearrangement layer. The rearrangement layer, the second convolutional layer, the normalization layer, the spiking neuron layer, and the spiking anti-rearrangement layer can be viewed as stacked. There is no essential difference between the first convolutional layer and the second convolutional layer structure here, just to distinguish the convolutions at different positions. Multiple pulse-rearranged residual modules are set up next to each other as a stage, and the entire pulse-rearranged-based deep residual neural network can be configured with i (i>0) stages. The structure in which multiple pulse rearrangement residual modules are placed next to each other can be expressed as:

是第i个{脉冲重排-卷积-规范化-脉冲神经元层-脉冲反重排}堆叠；此处的n是堆叠的总数量(不同于第四步中的n)。公式(2)中的

即为经过多个{脉冲重排-卷积-规范化-脉冲神经元层-脉冲反重排}作用后得到的输出，与原始的输入S[t]一起，经过连接函数g的作用，得到输出O[t]。连接函数g的完整形式为s_o＝g(s_a,s_b)，本质上是一个逐元素的逻辑函数。为了加以区分，用s_o的取值组成的二进制转换为十进制来表示对应的g。例如表1所示的示例中，s_o分别取值为1,0,1,1,转换为十进制为11,因此对应的g记作g₁₁。where S[t] is the input of the entire pulse rearrangement residual module at time step t,

is the ith {spike rearrangement-convolution-normalization-spike neuron layer-spike anti-rearrangement} stack; n here is the total number of stacks (different from n in step 4). in formula (2)

That is, the output obtained after multiple actions of {pulse rearrangement-convolution-normalization-spike neuron layer-pulse anti-rearrangement}, together with the original input S[t], through the action of the connection function g, the output is obtained O[t]. The full form of the connection function g is s _o =g(s _a ,s _b ), which is essentially an element-wise logistic function. In order to distinguish, the binary conversion composed of the values of s _o is converted to decimal to represent the corresponding g. For example, in the example shown in Table 1, the values of s _o are 1, 0, 1, and 1 respectively, which are converted to 11 in decimal, so the corresponding g is denoted as g ₁₁ .

Table 1 Element-by-element logical function value corresponding table

可以使用

来求解。According to the truth table composed of (s _a , s _b , s _o ), 16 logical functions can be taken, and the corresponding g is {g ₀ , g ₁ ,…,g ₁₅ }. It should be noted here that when training the network in the third step, the gradient of g can be defined by numerical gradient. For example, using g ₁₁ in Table 1, if you want to solve

can use

to solve.

In the second step, the deep residual neural network based on pulse rearrangement after configuring the pulse rearrangement residual block is used as the first pulse rearrangement model.

The third step is to train the first pulse rearrangement model.

及其参数θ，含有N个数据的训练集D，损失函数

训练总轮数E。During specific training, the parameter is the learning rate ∈, the network

and its parameter θ, the training set D containing N data, the loss function

The total number of training rounds E.

[1] Let e=1

[2] Let i=1

[3] Take out (X, Y)=D[i] from the data set, and obtain the length of the input sequence as T

[4] Let t=1

[5] Input X[t] into the network, get

[6] Let t=t+1

[7] If t>, enter [8]; otherwise, return to [5]

[8] Calculate the loss

[9] Backpropagation, Gradient Descent, Updating Parameters

[10] Let i=i+1

[11] If i>, enter [12]; otherwise, go back to [3]

[12] Let e=e+1

[13] If e>, exit; otherwise, return to [2]

或交叉熵

或者是其他度量距离的损失。若为数据回归任务，当回归目标是一个序列时，损失函数可以直接是

当回归目标是单个元素Y时，损失函数需要根据输出的类型做相应的更改。当使用所有时刻的平均值作为回归结果时，损失函数可以为

当使用最后一个时刻的输出或最后一层脉冲神经元在最后一个时刻的膜电位作为回归结果，损失损失函数可以为

If it is a data classification task, when the real class is j, for any t, Y[t][j]=1; and for any k≠j, Y[t][k]=0. The loss function can be the mean squared error

or cross entropy

Or the loss of other metric distances. For data regression tasks, when the regression target is a sequence, the loss function can be directly

When the regression target is a single element Y, the loss function needs to be changed according to the type of output. When using the mean of all moments as the regression result, the loss function can be

When using the output of the last moment or the membrane potential of the last layer of spiking neurons at the last moment as the regression result, the loss loss function can be

In the fourth step, the target data is input into the trained first pulse rearrangement model for processing, and the target processing result is obtained.

In an implementation manner, inputting the target data into the trained first pulse rearrangement model includes: inputting the target data into the first convolutional layer for downsampling; inputting the downsampled target data into a plurality of adjacent pulse rearrangements The row residual module performs pulse rearrangement residual processing to obtain the first processing result; the first processing result is input into the pooling layer and the fully connected layer in turn to obtain the second processing result. The pulse rearrangement residual module is sequentially configured with a pulse rearrangement layer, a second convolution layer, a normalization layer, a pulse neuron layer and a pulse anti-rearrangement layer; the down-sampled target data is input into multiple adjacent pulse rearrangement residuals. The difference module performs pulse rearrangement residual processing to obtain the first processing result, including: inputting the down-sampled target data into the pulse rearrangement layer to obtain the pulse rearrangement result; passing the pulse rearrangement result through the second convolution layer, The normalization layer, the spiking neuron layer and the spiking anti-rearrangement layer are processed to obtain the first processing result.

In a specific implementation manner, the down-sampled target data is input into the pulse rearrangement layer, and the pulse rearrangement result is obtained, which is represented by formula (1):

表示向下取整。Among them, Y represents the rearrangement operation performed by the pulse rearrangement layer, X represents the down-sampled target data; n, z, y, and x represent the batch size, channel number, height and width of the down-sampled target data in turn; M represents the total number of channels, r represents the rearrangement coefficient, % represents the remainder,

Indicates rounded down.

脉冲反重排则是做逆向操作，将[N,M,H,W]的形状重塑为

在这里参考图3，目标数据为四个脉冲矩阵，分别是a，b，c，d，尺寸为2x2的脉冲矩阵组成的4通道脉冲矩阵，形状为[4,2,2]；经过脉冲重排后得到一个大的4x4的脉冲矩阵，尺寸为[1,4,4]。将[1,4,4]的大脉冲矩阵拆分成4通道得到[4,2,2]的4通道脉冲矩阵，即为脉冲反重排。作为可变换的实施方式，若目标数据为图像，则经过脉冲重排残差模块，会经过像素重排处理，再经卷积处理、规范化处理、脉冲神经元层处理后是像素反重排处理，最后再池化和全连接。另外，若欲输入数据为单个数字时，将该单个数字重复预设次得到长度为预设长的序列；若欲输入数据为多个数字组成的长度为T的序列，其中T大于1，则不需要进行重复；将长度为预设长的序列和长度为T的序列作为目标数据。Pulse rearrangement refers to splitting the pulse data on the channel into width and height in sequence, while pulse reverse rearrangement is an inverse operation, which is a special size transformation. For example, the size is [N, M, H ,W] (where N is the batch size, M is the number of channels, H is the height, and W is the width), and the rearrangement coefficient is r ² (where r is a positive integer), then change its shape to

Pulse anti-rearrangement is to do the reverse operation and reshape the shape of [N, M, H, W] as

Referring to Figure 3 here, the target data is four pulse matrices, namely a, b, c, d, a 4-channel pulse matrix composed of pulse matrices with a size of 2x2, and the shape is [4, 2, 2]; After rowing, we get a large 4x4 impulse matrix of size [1, 4, 4]. The large pulse matrix of [1,4,4] is divided into 4 channels to obtain the 4-channel pulse matrix of [4,2,2], which is the pulse anti-rearrangement. As a transformable embodiment, if the target data is an image, it will undergo pixel rearrangement processing through the pulse rearrangement residual module, and then undergo pixel reverse rearrangement processing after convolution processing, normalization processing, and pulse neuron layer processing. , and finally pooled and fully connected. In addition, if the data to be input is a single number, repeat the single number a preset number of times to obtain a sequence with a preset length; if the data to be input is a sequence of length T composed of multiple numbers, where T is greater than 1, then There is no need to repeat; the sequence of the preset length and the sequence of length T are used as the target data.

The above normalization layer function is to complete batch normalization or layer normalization. If batch normalization is used, the parameters of batch normalization can be combined with the convolution layer to reduce the amount of network parameters and increase the calculation speed. Specifically, the weight of the convolution is written as W _conv , and the bias is generally set to 0, because batch normalization has its own bias; the weight of batch normalization is written as W _bn , the bias term is B _bn , and the statistical data mean X _m , the variance X _v , the weights and biases of the combined convolution are:

A spiking neuron layer refers to a layer composed of spiking neurons. The behavior of the spiking neuron layer can be described using the charge, discharge, reset 3 equations. Equation (3) indicates that the first equation is the charging equation:

H[t]=f(V[t-1],X[t]) (3)

where X[t] is the input at time t. To avoid confusion, use H[t] to denote the voltage after charging and V[t] to denote the voltage after discharge. where f represents the charging equation, and different neurons have different charging equations. The charging equation is obtained by discretizing the continuous-time differential equation. For example, the subthreshold dynamics for LIF neurons described using continuous-time differential equations is represented by equation (4):

After discretization, the subthreshold discrete time difference equation is obtained, which is the charging equation, which is represented by formula (5):

where _Vrest is the resting potential and τ is the membrane time constant. The second equation is the discharge equation, expressed by Equation (6):

S[t]=Θ(H[t]-V _th ) (6)

S[t] is the impulse released by the neuron, and Θ(x) is the Heaviside step function that outputs 1 if and only if x ≥ 0, and 0 otherwise. The discharge equation states that when the charged voltage of the neuron exceeds the threshold V _th , it will release a pulse of 1, otherwise it will output a 0. The third equation is the reset equation expressed by equation (7):

Among them, V _reset represents the reset potential. Hard reset means hard reset, which means that the voltage is directly reset to V _reset after the neuron releases the pulse. Soft reset means soft reset, which means that after the neuron releases the pulse, the voltage will decrease V _th .

It should be noted that the derivative of Θ(x) in the discharge equation is infinite when x=0, and 0 when x≠0. Such derivatives are directly used for gradient descent, which will make the network unable to train. In order to solve this problem, the gradient substitution method is used, and Θ(x) is still used in the forward propagation to ensure that the output of the spiking neuron is a pulse; and the derivative σ' of the substitution function σ(x) is used in the backward propagation ( x). σ(x) is usually selected as a continuous function whose value range is (0,1), such as the common sigmoid function.

In a specific implementation manner, referring to FIG. 2 again, the obtaining of the target processing result includes: obtaining a category corresponding to the target data according to the second processing result; obtaining a regression sequence and/or a regression sequence based on the second processing result Or regress a single vector; take the category corresponding to the target data, the regression sequence and/or the regression single vector as the target processing result. Among them, as shown in Figure 2, the output of the model at the last moment or the membrane potential of the last layer of spiking neurons at the last moment can be used as a single vector for regression. It can be seen that the method described in this application is suitable for various application scenarios.

参数量大为减少，降低了过拟合风险，也减轻了存储和计算开销。Compared with ordinary convolution without pulse rearrangement and anti-rearrangement in this application, the amount of parameters is reduced from M _in · M _out · K _h · K _w to

The number of parameters is greatly reduced, the risk of overfitting is reduced, and the storage and computational overhead is also reduced.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention.

Example 3:

This embodiment provides a data processing apparatus based on a pulse rearrangement deep residual neural network. As shown in FIG. 4 , the apparatus includes:

A configuration module 401 is used to configure a plurality of adjacent pulse rearrangement residual modules in the pulse rearrangement-based deep residual neural network;

The rearrangement module 402 is configured to use the pulse rearrangement-based deep residual neural network after configuring the pulse rearrangement residual block as the first pulse rearrangement model;

A training module 403, configured to train the first pulse rearrangement model;

The processing module 404 is configured to input the target data into the trained first pulse rearrangement model for processing to obtain the target processing result.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the invention.

It should also be emphasized that the system provided in the embodiments of the present application can acquire and process related data based on artificial intelligence technology. Among them, artificial intelligence (AI) is a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. . The basic technologies of artificial intelligence generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation/interaction systems, and mechatronics. Artificial intelligence software technology mainly includes computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning/deep learning.

Please refer to FIG. 5 below, which shows a schematic diagram of a computer device provided by some embodiments of the present application. As shown in FIG. 5 , 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 through the bus 202 ; There is a computer program that can be run on the processor 200. When the processor 200 runs the computer program, the data processing method based on the pulse rearrangement deep residual neural network provided by any of the foregoing embodiments of the present application is executed.

The memory 201 may include a high-speed random access memory (RAM: Random Access Memory), and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory. The communication connection between the network element of the system and at least one other network element is realized through at least one communication interface 203 (which may be wired or wireless), which may use the Internet, a wide area network, a local network, a metropolitan area network, and the like.

The bus 202 may be an ISA bus, a PCI bus, an EISA bus, or the like. The bus can be divided into an address bus, a data bus, a control bus, and the like. The memory 201 is used to store a program, and the processor 200 executes the program after receiving the execution instruction, and the data of the pulse rearrangement-based deep residual neural network disclosed in any of the foregoing embodiments of the present application The processing method may be applied in the processor 200 or implemented by the processor 200 .

The processor 200 may be an integrated circuit chip with signal processing capability. In the implementation process, each step of the above-mentioned method can be completed by an integrated logic circuit of hardware in the processor 200 or an instruction in the form of software. The above-mentioned processor 200 may be a general-purpose processor, including a central processing unit (Central Processing Unit, CPU for short), a network processor (Network Processor, NP for short), etc.; it may also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic device, discrete hardware component. The methods, steps, and logic block diagrams disclosed in the embodiments of this application can be implemented or executed. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like. The steps of the method disclosed in conjunction with the embodiments of the present application may be directly embodied as executed by a hardware decoding processor, or executed by a combination of hardware and software modules in the decoding processor. The software modules may be located in random access memory, flash memory, read-only memory, programmable read-only memory or electrically erasable programmable memory, registers and other storage media mature in the art. 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.

The embodiments of the present application also provide a computer-readable storage medium corresponding to the data processing method based on the pulse rearrangement deep residual neural network provided by the foregoing embodiments, please refer to FIG. 6 , the computer-readable storage medium shown in FIG. 6 The medium is an optical disc 30, on which a computer program (ie, a program product) is stored. When the computer program is run by the processor, the computer program will execute the data processing method based on the pulse rearrangement deep residual neural network provided by any of the foregoing embodiments. .

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 and magnetic storage media, which will not be described in detail here.

The computer-readable storage medium provided by the above-mentioned embodiments of the present application and the quantum key distribution channel allocation method in the space-division multiplexing optical network provided by the embodiments of the present application are based on the same inventive concept, and have the same inventive concept as the application program stored in the computer-readable storage medium. run or achieve the same beneficial effect as the method.

Embodiments of the present application further provide a computer program product, including a computer program, when the computer program is executed by a processor, the computer program implements the steps of the data processing method based on the pulse rearrangement deep residual neural network provided by any of the foregoing embodiments, including: Configure a plurality of adjacent pulse rearrangement residual modules in the pulse rearrangement-based deep residual neural network; use the pulse rearrangement-based deep residual neural network after configuring the pulse rearrangement residual block as the first pulse rearrangement model; The first pulse rearrangement model is trained; the target data is input into the trained first pulse rearrangement model for processing, and the target processing result is obtained.

It should be noted that the algorithms and displays provided herein are not inherently related to any particular computer, virtual appliance or other device. Various general-purpose devices can also be used with the teachings based on this. The structure required to construct such a device is apparent from the above description. Furthermore, this application is not directed to any particular programming language. It should be understood that the content of the application described herein can be implemented using a variety of programming languages and that the descriptions of specific languages above are intended to disclose the best mode of the application. In the description provided herein, numerous specific details are set forth. It will be understood, however, that the embodiments of the present application may be practiced without these specific details. In some instances, well-known methods, structures and techniques have not been shown in detail in order not to obscure an understanding of this description.

Similarly, it is to be understood that in the above description of exemplary embodiments of the present application, various features of the present application are sometimes grouped together into a single embodiment, figure or its description. This disclosure, however, should not be interpreted as reflecting an intention that the claimed application requires more features than are expressly recited in each claim. Rather, as the following claims reflect, inventive aspects lie in less than all features of a single foregoing disclosed embodiment. Thus, 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 this application.

Those skilled in the art will understand that the modules in the device in the embodiment can be adaptively changed and arranged in one or more devices different from the embodiment. The modules or units or components in the embodiments may be combined into one module or unit or component, and further they may be divided into multiple sub-modules or sub-units or sub-assemblies. All features disclosed in this specification and all processes or elements of any method or apparatus so disclosed may be combined in any combination, except that at least some of such features and/or procedures or elements are mutually exclusive. Unless expressly stated otherwise, each feature disclosed in this specification may be replaced by alternative features serving the same, equivalent or similar purpose.

Various component embodiments of the present application may be implemented in hardware, or in software modules running on one or more processors, or in a combination thereof. Those skilled in the art should understand that, in practice, a microprocessor or a digital signal processor (DSP) may be used to implement some or all functions of some or all components in the apparatus for creating a virtual machine according to the embodiments of the present application. The present application can also be implemented as an apparatus or apparatus program for performing part or all of the methods described herein. A program implementing the present application may be stored on a computer-readable medium, or may be in the form of one or more signals. Such signals may be downloaded from Internet sites, or provided on carrier signals, or in any other form.

The above description is only a preferred embodiment of the present application, but the protection scope of the present application is not limited to this. Substitutions should be covered within the protection scope of this application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims (10)

1. A data processing method based on a pulse rearrangement depth residual error neural network is characterized by comprising the following steps:

configuring a plurality of immediately adjacent pulse rearrangement residual modules in a pulse rearrangement based depth residual neural network;

taking the pulse rearrangement-based depth residual error neural network after the pulse rearrangement residual block is configured as a first pulse rearrangement model;

training the first pulse rearrangement model;

and inputting the target data into the trained first pulse rearrangement model for processing to obtain a target processing result.

2. The data processing method based on the pulse rearrangement depth residual neural network according to claim 1, characterized in that the first pulse rearrangement model is further provided with a first convolution layer, a pooling layer and a full-link layer; inputting the target data into the trained first pulse rearrangement model for processing, wherein the processing comprises the following steps:

inputting target data into the first convolution layer for down-sampling;

inputting the down-sampled target data into the plurality of adjacent pulse rearrangement residual modules for pulse rearrangement residual processing to obtain a first processing result;

and inputting the first processing result into the pooling layer and the full-connection layer in sequence to obtain a second processing result.

3. The method according to claim 2, wherein the obtaining the target processing result comprises:

obtaining the category corresponding to the target data according to the second processing result;

obtaining a regression sequence and/or a regression single vector based on the second processing result;

and taking the category corresponding to the target data, the regression sequence and/or the regression single vector as a target processing result.

4. The data processing method based on the pulse rearrangement depth residual neural network according to claim 2, wherein the pulse rearrangement residual module is configured with a pulse rearrangement layer, a second convolution layer, a normalization layer, a pulse neuron layer and a pulse rearrangement layer in sequence; the step of inputting the downsampled target data into the plurality of adjacent pulse rearrangement residual error modules to perform pulse rearrangement residual error processing to obtain a first processing result includes:

inputting the target data after down sampling into a pulse rearrangement layer to obtain a pulse rearrangement result;

and processing the pulse rearrangement result by a second convolution layer, a normalization layer, a pulse neuron layer and a pulse reverse rearrangement layer in sequence to obtain a first processing result.

5. The data processing method based on the pulse rearrangement depth residual neural network of claim 4, wherein the target data after down-sampling is input into a pulse rearrangement layer to obtain a pulse rearrangement result, and the formula is as follows:

wherein, Y represents the rearrangement operation performed by the pulse rearrangement layer, and X represents the target data after down sampling; n, z, y and x sequentially represent the batch size, channel number, height and width of the target data after down-sampling; m represents the total number of channels, r represents the rearrangement coefficient,% represents the remainder,

indicating a rounding down.

6. The method of claim 2, wherein a first of the plurality of immediately adjacent pulse rearrangement residual modules is further configured with a downsampling function.

7. The data processing method based on the pulse rearrangement deep residual error neural network of claim 1, wherein before the target data is input into the trained first pulse rearrangement model for processing, and a target processing result is obtained, the method further comprises:

acquiring data to be input;

if the data to be input is a single number, repeating the single number for a preset time to obtain a sequence with a preset length; if the data to be input is a sequence with the length of T and formed by a plurality of numbers, wherein T is more than 1, the data does not need to be repeated;

and taking the sequence with the length being a preset length and the sequence with the length being T as target data.

8. A data processing apparatus based on a pulse rearrangement depth residual neural network, the apparatus comprising:

a configuration module for configuring a plurality of immediately adjacent pulse rearrangement residual modules in a pulse rearrangement-based depth residual neural network;

the rearrangement module is used for taking the pulse rearrangement-depth-based residual error neural network after the pulse rearrangement residual block is configured as a first pulse rearrangement model;

a training module for training the first pulse rearrangement model;

and the processing module is used for inputting the target data into the trained first pulse rearrangement model for processing to obtain a target processing result.

9. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method according to any of claims 1-7.

10. A computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.

Images