A storage pool computing network optimization method and related device

Abstract

The present application discloses a method for optimizing a storage pool computing network and a related device, the method comprising: sampling an input signal to obtain a sampled signal; quantizing the sampled signal in at least two quantization methods to obtain at least two digital signals, wherein the values ​​of elements in different digital signals are different; inputting voltage pulses corresponding to the elements in the different digital signals into storage pools constructed by different numbers of virtual nodes, so that different storage pools extract signal features of the input signal in different quantization methods. By quantizing the signal in different methods and inputting it into storage pools constructed by different numbers of virtual nodes, the internal state richness of the storage pool can be improved, and the accuracy of signal recognition by the storage pool system can be further improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a storage pool computing network optimization method and related devices. Background Art

[0002] With the rapid development of artificial intelligence technology, neuromorphic computing, which aims to achieve artificial intelligence by imitating the mechanisms of neurons and synapses that constitute the human brain, has become a research focus. Among them, reservoir computing (RC), which is derived from the traditional recurrent neural network (RNN), has been widely used in dynamic system identification, time series detection and other fields due to its low training cost and simple hardware implementation.

[0003] The accuracy of the signal recognition of the reservoir system is closely related to the richness of the internal state of the reservoir. In the existing memristor-based reservoir computing technology, the inherent differences between memristors can be relied on to improve the richness of the internal state of the reservoir. However, when the process conditions and parameters of the device are determined, the richness of the reservoir is determined accordingly and cannot be adjusted according to specific task types. If the device is redesigned every time according to different task types, the hardware overhead is too high. Summary of the invention

[0004] Based on the above problems, the present application provides a storage pool computing network optimization method and related devices to improve the accuracy of storage pool system recognition signals.

[0005] The embodiments of the present application disclose the following technical solutions:

[0006] In a first aspect, the present application provides a method for optimizing a storage pool computing network, the method comprising:

[0007] Sampling the input signal to obtain a sampled signal;

[0008] Quantizing the sampled signal in at least two quantization modes to obtain at least two digital signals, wherein the values ​​of elements in different digital signals are different;

[0009] The voltage pulse signals corresponding to the elements in the different digital signals are input into the reserve pools constructed by different numbers of virtual nodes, so that the different reserve pools can extract the signal characteristics of the input signals in different quantization modes.

[0010] Optionally, the at least two digital signals include at least: a first digital signal and a second digital signal, and the step of inputting voltage pulse signals corresponding to elements in the different digital signals into a reserve pool constructed by different numbers of virtual nodes includes:

[0011] sequentially inputting voltage pulse signals corresponding to a first number of elements in the first digital signal into a first reserve pool constructed by a first number of virtual nodes; and,

[0012] The voltage pulse signals corresponding to the second number of elements in the second digital signal are sequentially input into a second reserve pool constructed by the second number of virtual nodes.

[0013] Optionally, the at least two quantization modes include at least a first quantization mode and a second quantization mode, the first quantization mode corresponds to a first number of bits, the second quantization mode corresponds to a second number of bits, and the quantization processing of the sampled signal by the at least two quantization modes to obtain at least two digital signals includes:

[0014] Determine a first value interval according to the maximum value, the minimum value and the first number of bits of the sampled signal;

[0015] Obtaining a first digital signal according to a corresponding relationship between the amplitude of the sampling signal and the first value interval;

[0016] Determine a second value interval according to the maximum value, the minimum value, and the second number of bits of the sampled signal;

[0017] A second digital signal is obtained according to the corresponding relationship between the amplitude of the sampling signal and the second value interval.

[0018] Optionally, the number of bits used in the quantization method is related to the performance of the memristor.

[0019] Optionally, the digital signal is a digital signal represented by binary code.

[0020] In a second aspect, an embodiment of the present application provides a storage pool computing network optimization device, the device comprising:

[0021] Sampling module, quantization module and input module;

[0022] The sampling module samples the input signal to obtain a sampled signal;

[0023] The quantization module is used to perform quantization processing on the sampled signal in at least two quantization modes to obtain at least two digital signals, where the values ​​of elements in different digital signals are different;

[0024] The input module is used to input the voltage pulse signals corresponding to the elements in the different digital signals into the reserve pools constructed by different numbers of virtual nodes, so that different reserve pools can extract the signal characteristics of the input signals in different quantization modes.

[0025] Optionally, the at least two digital signals include: a first digital signal and a second digital signal, and the input module is specifically configured to:

[0026] sequentially inputting voltage pulse signals corresponding to a first number of elements in the first digital signal into a first reserve pool constructed by a first number of virtual nodes; and,

[0027] The voltage pulse signals corresponding to the second number of elements in the second digital signal are sequentially input into a second reserve pool constructed by the second number of virtual nodes.

[0028] Optionally, the at least two quantization modes include at least a first quantization mode and a second quantization mode, the first quantization mode corresponds to a first number of bits, the second quantization mode corresponds to a second number of bits, the sampling signal is quantized by the at least two quantization modes to obtain at least two digital signals, and the quantization module is specifically used to:

[0029] Determine a first value interval according to the maximum value, the minimum value and the first number of bits of the sampled signal;

[0030] Obtaining a first digital signal according to a corresponding relationship between the amplitude of the sampling signal and the first value interval;

[0031] Determine a second value interval according to the maximum value, the minimum value, and the second number of bits of the sampled signal;

[0032] A second digital signal is obtained according to the corresponding relationship between the amplitude of the sampling signal and the second value interval.

[0033] In a third aspect, an embodiment of the present application provides a storage pool computing network optimization device, the device comprising:

[0034] Memory, processor;

[0035] The memory is used to store computer programs;

[0036] The processor is used to implement the steps of the reserve pool computing network optimization method described in any one of the first aspects when executing the computer program.

[0037] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium,

[0038] The computer-readable storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the reserve pool computing network optimization method described in any one of the first aspects are implemented.

[0039] Compared with the prior art, this application has the following beneficial effects:

[0040] The signal processing method provided by the present application is to sample an input signal to obtain a sampled signal; quantize the sampled signal by at least two quantization methods to obtain at least two digital signals, and the values ​​of elements in different digital signals are different; and input the voltage pulse signals corresponding to the elements in the different digital signals into reserve pools constructed by different numbers of virtual nodes, so that different reserve pools can extract the signal characteristics of the input signal under different quantization methods.

[0041] This application mainly constructs different reserve pools through different numbers of virtual nodes, and obtains different digital signals by performing different quantization processing on the sampled signals using different quantization methods. The different digital signals are input into different reserve pools constructed by different numbers of virtual nodes, which can improve the richness of the reserve pools and thus improve the accuracy of the reserve pool system in identifying signals. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0043] Figure 1 A flowchart of a method for optimizing a storage pool computing network provided in an embodiment of the present application;

[0044] Figure 2 A schematic diagram of digital signal quantization provided in an embodiment of the present application;

[0045] Figure 3 A schematic diagram of a conventional reserve pool and a virtual node reserve pool constructed in accordance with an embodiment of the present application;

[0046] Figure 4 A schematic diagram of the results of identifying an input signal under the conditions of different quantization methods and reserve pools constructed with different virtual nodes provided in an embodiment of the present application;

[0047] Figure 5 A flowchart of a method for optimizing a reserve pool computing network constructed by a virtual node provided in an embodiment of the present application;

[0048] Figure 6 A schematic diagram of a storage pool computing network optimization method for constructing a virtual node provided in an embodiment of the present application;

[0049] Figure 7 A schematic diagram of the structure of a reserve pool computing network optimization device constructed by a virtual node provided in an embodiment of the present application. DETAILED DESCRIPTION

[0050] As described above, reservoir computing is an efficient artificial neural network suitable for processing time series signals. It is derived from the traditional recurrent neural network RNN, but has lower training costs and simpler hardware implementation. It has been widely used in dynamic system identification, time series prediction and other fields. The computing power of the reservoir system is closely related to the richness of the internal state of the reservoir.

[0051] After research, the inventors found that the current reserve pool system can use the differences between memristors to improve the richness of the internal state of the reserve pool. However, once the process conditions of the device are determined, the richness of the internal state of the reserve pool is determined accordingly and cannot be adjusted according to specific tasks. If the process conditions are adjusted according to specific tasks each time, the hardware overhead will be too large and the cost-effectiveness will be low.

[0052] In view of this, the present application provides a method for optimizing a reserve pool computing network, which can improve the differences between devices through software design, thereby constructing different reserve pools and enriching the internal richness of the reserve pools. The optimization method includes: sampling the input signal to obtain a sampled signal; quantizing the sampled signal through at least two quantization methods to obtain at least two digital signals, and the values ​​of elements in different digital signals are different; inputting the voltage pulse signals corresponding to the elements in different digital signals into reserve pools constructed by different numbers of virtual nodes, so that different reserve pools can extract signal characteristics of the input signal under different quantization methods.

[0053] It can be seen that this method processes the input signal differently, converts the digital signals generated under different quantization methods into input pulse sequences and inputs them into the reserve pool, so that the same memristor can also construct different reserve pools due to the different numbers or values ​​of virtual nodes, thereby memorizing different characteristics of the signal, improving the internal richness of the reserve pool system, and then improving the accuracy of identifying the input signal.

[0054] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0055] See also Figure 1 , which is a flow chart of a reserve pool computing network optimization method provided in an embodiment of the present application.

[0056] like Figure 1 As shown, the storage pool computing network optimization method includes:

[0057] S101: Sample the input signal to obtain a sampled signal.

[0058] The input signal can be various analog signals in production and life. Analog signal refers to information represented by continuously changing physical quantities. The amplitude, frequency, or phase of the signal changes continuously with time, or within a continuous time interval, the characteristic quantity representing the information can be presented as a signal of any numerical value at any moment, such as: the image in the camera, any sound recorded by the recorder, any photo in the camera, and the pressure and speed recorded in the workshop control room.

[0059] In addition, the input signal may also be any digital signal. It should be noted that the embodiment of the present application does not impose any limitation on the form and content of the input signal.

[0060] Sampling is the discretization of continuous signals in time, that is, taking the instantaneous value of the analog signal point by point at a certain time interval. Of course, sampling can also be performed on digital signals. Generally speaking, the higher the sampling frequency, the denser the number of sampling points, and the closer the obtained discrete signal is to the original signal. However, too high a sampling frequency is not desirable. For a signal of fixed length (T), too much data (N = T / △t) is collected, which adds unnecessary computing workload and storage space to the computer; if the amount of data (N) is limited, the sampling time is too short, which will cause some data information to be excluded. If the sampling frequency is too low and the sampling points are too far apart, the discrete signal is not enough to reflect the waveform characteristics of the original signal, and the signal cannot be restored, causing signal confusion. Intuitively speaking, signal aliasing is mistaking a signal that should be a high-frequency signal for a low-frequency signal. According to the sampling theorem, when the sampling frequency is greater than twice the maximum component frequency in the original signal, the signal can be restored relatively well. If it is less than twice the maximum component frequency, it is under-sampling, and signal aliasing will occur.

[0061] When the input signal is an analog signal, the sampled signal can be a set of sample points that are discrete in time and continuous in amplitude. The sampled signal is actually still an analog signal.

[0062] Sampling can be done with professional data acquisition equipment or with a computer system equipped with a data acquisition card.

[0063] S102: quantizing the sampled signal in at least two quantization modes to obtain at least two digital signals, where the values ​​of elements in different digital signals are different.

[0064] Quantization can be a process of converting a sampled analog signal into a digital signal by rounding. From the sampling in S101, it can be seen that when the input signal is an analog signal, the signal after the input signal is sampled is a step signal. Although it is discrete on the time axis, the amplitude of the step signal is still continuous. If this signal is accurately represented by binary digits, an infinite number of binary codes are required. Therefore, a rounding method is used to merge each sampled value into a certain adjacent integer, so that a binary code of a certain word length can be used to represent the sampled value. This process of taking a finite number of values ​​to approximately represent a continuously changing signal is called quantization. Quantization can be divided into uniform quantization and non-uniform quantization, in which uniform quantization is to divide the dynamic range of the input signal uniformly; non-uniform quantization is to divide the dynamic range of the input signal unevenly, and generally quantization is performed using a curve similar to an exponential. Non-uniform quantization is proposed for uniform quantization. For example, in general speech signals, most of them are small-amplitude signals, and human hearing follows the exponential law. In order to ensure that the signal can be restored more accurately, more bits should be used to represent small signals.

[0065] The quantization method in the embodiment of the present application can select an appropriate number of bits according to the characteristics of the memristor. The higher the number of bits, the closer the quantized result is to the original input signal, that is, the lower the distortion. The quantization method and the number of virtual nodes jointly determine the richness of the internal state of the reserve pool. The specific relationship is: when the quantization method is x bits and the number of virtual nodes is n, the internal state of the reserve pool is (2 x ) n That is, the memristors used to construct the reserve pool are required to be (2 x ) n The states can be distinguished, and within the range allowed by the memristor characteristics, the most appropriate values ​​of x and n are determined by simulation according to the specific task to be performed.

[0066] To facilitate understanding of the process of sampling signal quantization, Figure 2 The quantization is explained in two ways: 1 bit and 2 bits. The upper figure uses 1 bit, and the lower figure uses 2 bits.

[0067] When quantization is performed using a 1-bit quantization method, a set of digital signals can be obtained, which can be expressed as {1,1,0,1,0,1,1,1};

[0068] When quantization is performed using a 2-bit quantization method, a set of digital signals can be obtained, which can be expressed as {2, 2, 1, 2, 0, 3, 2, 3}.

[0069] S103: Inputting voltage pulse signals corresponding to elements in different digital signals into storage pools constructed by different numbers of virtual nodes, so that different storage pools extract signal features of the input signals under different quantization methods.

[0070] The reserve pool calculation RC, which is derived from the traditional recurrent neural network RNN, has been widely used in dynamic system identification, time series detection and other fields due to its low training cost and simple hardware implementation.

[0071] The reserve pool can extract the signal characteristics of the input signal under different quantization methods through the memristor.

[0072] Memristor, the full name of which is memory resistor (Memristor). It is a circuit device that represents the relationship between magnetic flux and charge. Memristor has the dimension of resistance, but unlike resistor, the resistance of memristor is determined by the charge flowing through it. Therefore, by measuring the resistance of memristor, we can know the amount of charge flowing through it, thus it has the function of memorizing charge.

[0073] For ease of understanding, the construction of the reserve pool will be described in detail below. In the embodiments of the present application, no limitation is imposed on the number of virtual nodes for constructing the reserve pool and the number of reserve pools.

[0074] Traditional reservoir computing can be composed of three layers: input layer, reservoir layer, and output layer. The input layer inputs the input signal to the reservoir layer through fixed random weight connections. The reservoir layer is usually composed of a large number of randomly connected nonlinear nodes, forming a recurrent network, that is, a network with an internal feedback loop. Under the influence of the input signal, the network exhibits transient responses, which are read out in the output layer through the linear weighted sum of the states of individual nodes. The goal of RC is to achieve a specific nonlinear transformation of the input signal or to classify the input signal. Classification involves the distinction between a set of input data, such as identifying features such as images, sounds, and time series.

[0075] The input signal is nonlinearly transformed into a high-dimensional state space in which the signal is represented. This is achieved by using a large number of reservoir nodes, which are interconnected through the periodic nonlinear dynamics of the reservoir. In practice, conventional RC structures employ hundreds / thousands of nonlinear reservoir nodes to achieve good performance.

[0076] In the embodiments of the present application, a reservoir computer is implemented, in which the structure of multi-connected nodes is replaced by a dynamic system, which includes a non-linear node subject to delayed feedback. Mathematically, a key feature of continuous time-delay systems is that their state space becomes infinite-dimensional. This is because their state at time t depends on the non-linear node in the continuous time interval [t - τ, t], where τ is the delay time. In practice, the dynamics of the delay system are still finite-dimensional, but exhibit the characteristics of high-dimensional and short-term memory. Therefore, the delay system can meet the requirements for the normal operation of the reservoir.

[0077] Combined with the accompanying drawings of the specification Figure 3 shows the basic principle of constructing a reservoir in the embodiments of the present application. In a delay interval of length τ, we define N equally spaced points in time with θ = τ / N. We call these N equally spaced points "virtual nodes" because they act like one of the nodes in a traditional reservoir. The value of the delay variable at every N points defines the state of the virtual node. These states characterize the transient response of the reservoir to a certain input at a given time. The separation time θ between virtual nodes plays an important role and can be used to optimize the reservoir dynamics. We choose θ < T, where T is the characteristic time scale of the non-linear node. With this choice, the state of the virtual node becomes dependent on the states of neighboring nodes. Connected in this way, the virtual nodes simulate a network that acts as a storage.

[0078] Generally, the richness of the internal state of the reservoir is reflected by the number of virtual nodes in the reservoir and the input signal of the reservoir calculation. If the reservoir system consists of three reservoirs, where Reservoir A and Reservoir C are both constructed by 2 virtual nodes, and Reservoir B is constructed by 3 virtual nodes. The voltage pulse signal corresponding to the 1-bit quantized digital signal is input into Reservoir A, then the internal state of Reservoir A has 2 2 kinds; the voltage pulse signal corresponding to the 1-bit quantized digital signal is input into Reservoir B, then the internal state of Reservoir B has 2 3 kinds; the 2-bit quantized digital signal is input into Reservoir C, then the internal state of Reservoir C has (2 2 ) 2 kinds.

[0079] By inputting the voltage pulses corresponding to different digital signals into the reservoirs constructed with different numbers of virtual nodes, more accurate features at the same time point can be extracted. For example, by inputting the 1-bit quantized digital signal {1, 1, 0, 1, 0, 1, 1, 1} and the 2-bit quantized digital signal {2, 2, 1, 2, 0, 3, 2, 3} into Reservoir A and Reservoir C constructed by 2 virtual nodes respectively, obviously, the 2-bit quantized digital signal is more accurate than the 1-bit quantized digital signal. In addition, the internal state of Reservoir B has 22 , the internal state of the reservoir C is (2 2 ) 2 , the internal state of reserve pool C is different from the internal state of reserve pool B, so for the reserve pool system, different features of the same signal can be extracted.

[0080] First, the input signal is sampled to obtain a sampled signal; then, the sampled signal is quantized by at least two quantization methods to obtain at least two digital signals, and the values ​​of elements in different digital signals are different; finally, the voltage pulse signals corresponding to the elements in different digital signals are input into a reserve pool constructed by different numbers of virtual nodes, so that the memristors in different reserve pools record the signal characteristics of the input signal under different quantization methods.

[0081] The present application mainly uses different reserve pools with different numbers of virtual nodes, obtains different digital signals by performing different quantization processing on the sampled signals using different quantization methods, and inputs the different digital signals into different reserve pools constructed with different numbers of virtual nodes, thereby improving the richness of the reserve pools and further improving the accuracy of the reserve pool system in identifying the input signal.

[0082] In conjunction with the instruction manual Figure 4 The beneficial effects are explained, among which, "F1" adopts 1-bit quantization and 2 virtual nodes to build a reserve pool; "F2" adopts 1-bit quantization and 3 virtual nodes to build a reserve pool; "F3" adopts 2-bit quantization and 2 virtual nodes to build a reserve pool. According to the above preprocessing method and reserve pool network construction, the dynamic gesture recognition task of the mobile phone is selected for verification. The final system recognition accuracy shows that when only "F3" is used for task verification, the accuracy is 82%. When "F2" and "F3" are used for task verification together, the accuracy is increased by 5% to 87%; when "F1", "F2" and "F3" are used for task verification together, the accuracy reaches 90%.

[0083] See also Figure 5 , which is a flowchart of a method for optimizing a reserve pool computing network constructed by virtual nodes provided in an embodiment of the present application.

[0084] like Figure 5 As shown, the storage pool computing network optimization method includes:

[0085] S501: Sample the input signal according to the Nyquist theorem to obtain a sampled signal.

[0086] Nyquist sampling theorem, that is, to restore the original signal from the sampled signal without distortion, the sampling frequency should be greater than 2 times the highest frequency of the signal. When the sampling frequency is less than 2 times the highest frequency of the spectrum, the spectrum of the signal has aliasing. When the sampling frequency is greater than 2 times the highest frequency of the spectrum, the spectrum of the signal has no aliasing.

[0087] S502: quantizing the sampled signal in at least two quantization modes to obtain at least two digital signals, where the values ​​of elements in different digital signals are different.

[0088] Quantization can be the process of converting a sampled analog signal into a digital signal by rounding. From the sampling in S501, it can be seen that the signal after the input signal is sampled is a step signal. Although it is discrete on the time axis, the amplitude of this step signal is still continuous. If this signal is accurately represented by binary digits, an infinite number of binary codes are required. Therefore, the rounding method is used to merge each sampled value into a certain adjacent integer, so that the sampled value can be represented by a binary code of a certain word length. This process of taking a finite number of values ​​to approximately represent a continuously changing signal is called quantization.

[0089] Quantization can be divided into uniform quantization and non-uniform quantization. Uniform quantization is to divide the dynamic range of the input signal uniformly; non-uniform quantization is to divide the dynamic range of the input signal unevenly, and generally use an exponential curve for quantization. Non-uniform quantization is proposed for uniform quantization. For example, most of the general speech signals are small-amplitude signals, and human hearing follows the exponential law. In order to ensure that the signal can be restored more accurately, more bits should be used to represent small signals.

[0090] For ease of understanding, the following is a Figure 2 The quantization is explained in two ways: 1 bit and 2 bits. The upper figure uses 1 bit, and the lower figure uses 2 bits.

[0091] When quantization is performed using a 1-bit quantization method, a set of digital signals can be obtained, which can be expressed as {1,1,0,1,0,1,1,1};

[0092] When quantization is performed using a 2-bit quantization method, a set of digital signals can be obtained, which can be expressed as {2, 2, 1, 2, 0, 3, 2, 3}.

[0093] S503: If the digital signal includes a first digital signal and a second digital signal, according to a reserve pool corresponding to the digital signal, the first digital signal and the second digital signal are cut according to the number of virtual nodes in the corresponding reserve pool.

[0094] For ease of understanding, the segmentation process is described by taking an example.

[0095] Assume that the first digital signal is {1,1,0,1,0,1,1,1}, which is input into the reserve pool A constructed by 2 virtual nodes and the reserve pool B constructed by 3 virtual nodes, and the second digital signal is {2,2,1,2,0,3,2,3}, which is input into the reserve pool B constructed by 2 virtual nodes.

[0096] The first digital signal is cut according to the number of virtual nodes in reserve pools A and C. After cutting, the digital signals are {[1,1],[0,1],[0,1],[1,1]}, {[1,1,0],[1,0,1],[1,1.]}; the first digital signal is cut according to the number of virtual nodes in reserve pool B. After cutting, the digital signals are {[2,2],[1,2],[0,3],[2,3]}.

[0097] S504: Convert the cut digital signal into a corresponding voltage pulse signal, and input the corresponding voltage pulse signal into a corresponding storage pool.

[0098] Since the signal is ultimately to be output to the memristor, the digital signal needs to be converted into a corresponding voltage pulse signal before that.

[0099] The internal state of pool A is 2 2 There are 2 internal states of pool B. 3 The internal states of the reservoir C are (2 2 ) 2 kind.

[0100] By inputting the voltage pulse signal corresponding to the cut digital signal into the corresponding reserve pool, it has two main effects: first, by inputting the voltage pulses corresponding to the same group of digital signals into the reserve pools constructed by different virtual nodes, the memristors in different reserve pools can memorize the characteristics of the signals in different time dimensions (for example, the voltage pulses corresponding to the first digital signal are respectively input into reserve pool A constructed by 2 virtual nodes and reserve pool B constructed by 3 virtual nodes); second, by inputting different digital signals into the reserve pool constructed by the same virtual node, the different characteristics presented by the signals at the same time point can be targeted (for example, the voltage pulse corresponding to the first digital signal is input into reserve pool A constructed by 2 virtual nodes, and the voltage pulse corresponding to the second digital signal is input into reserve pool C constructed by 2 virtual nodes).

[0101] It can be seen from the above two embodiments that the method adopted in this application is to adjust the quantization method and the virtual number of the reserve pool according to the input signal (recognition task), which will not increase the hardware overhead and is a relatively flexible method.

[0102] See also Figure 7 , this figure is a structural diagram of a reserve pool computing network optimization device constructed by a virtual node provided in an embodiment of the present application.

[0103] like Figure 7 As shown, the storage pool computing network optimization device includes: a sampling module 701, a quantization module 702 and an input module 703;

[0104] The sampling module 701 samples the input signal to obtain a sampled signal;

[0105] A quantization module 702 is used to quantize the sampled signal in at least two quantization modes to obtain at least two digital signals, where the values ​​of elements in different digital signals are different;

[0106] The input module 703 is used to input the voltage pulse signals corresponding to the elements in different digital signals into the reserve pools constructed by different numbers of virtual nodes, so that the memristors in different reserve pools record the signal characteristics of the input signals in different quantization modes.

[0107] The at least two digital signals include: a first digital signal and a second digital signal, and the input module 703 is specifically used to:

[0108] sequentially inputting voltage pulse signals corresponding to a first number of elements in the first digital signal into a first reserve pool constructed by a first number of virtual nodes; and,

[0109] The voltage pulse signals corresponding to the second number of elements in the second digital signal are sequentially input into a second reserve pool constructed by the second number of virtual nodes.

[0110] The at least two groups of quantization modes include at least a first quantization mode and a second quantization mode, the first quantization mode corresponds to a first number of bits, and the second quantization mode corresponds to a second number of bits. The sampled signal is quantized by at least two quantization modes to obtain at least two digital signals. The quantization module 702 is specifically used for:

[0111] Determine a first value interval according to a maximum value, a minimum value, and a first number of bits of the sampled signal;

[0112] Obtaining a first digital signal according to a corresponding relationship between the amplitude of the sampling signal and the first value interval;

[0113] Determine a second value interval according to the maximum value, the minimum value, and the second number of bits of the sampled signal;

[0114] A second digital signal is obtained according to the corresponding relationship between the amplitude of the sampling signal and the second value interval.

[0115] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components indicated as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0116] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for optimizing a storage pool computing network, characterized in that: The method comprises: Sampling the input signal to obtain a sampled signal; Quantizing the sampled signal in at least two quantization modes to obtain at least two digital signals, wherein the values ​​of elements in different digital signals are different; The voltage pulse signals corresponding to the elements in the different digital signals are input into the reserve pools constructed by different numbers of virtual nodes, so that the different reserve pools can extract the signal characteristics of the input signals in different quantization modes.

2. The method according to claim 1, characterized in that The at least two digital signals include at least: a first digital signal and a second digital signal, and the step of inputting voltage pulse signals corresponding to elements in the different digital signals into a reserve pool constructed by different numbers of virtual nodes includes: sequentially inputting voltage pulse signals corresponding to a first number of elements in the first digital signal into a first reserve pool constructed by a first number of virtual nodes; and, The voltage pulse signals corresponding to the second number of elements in the second digital signal are sequentially input into a second reserve pool constructed by a second number of virtual nodes.

3. The method according to claim 1, characterized in that The at least two quantization modes include at least a first quantization mode and a second quantization mode, the first quantization mode corresponds to a first bit number, the second quantization mode corresponds to a second bit number, and the at least two digital signals are obtained by quantizing the sampled signal using the at least two quantization modes, including: Determine a first value interval according to the maximum value, the minimum value and the first number of bits of the sampled signal; Obtaining a first digital signal according to a corresponding relationship between the amplitude of the sampling signal and the first value interval; Determine a second value interval according to the maximum value, the minimum value, and the second number of bits of the sampled signal; A second digital signal is obtained according to the corresponding relationship between the amplitude of the sampling signal and the second value interval.

4. The method according to claim 1, characterized in that The number of bits used in the quantization method is related to the performance of the memristor.

5. The method according to claim 1, characterized in that The digital signal is a digital signal represented by binary code.

6. A storage pool computing network optimization device, characterized in that: The device comprises: a sampling module, a quantization module and an input module; The sampling module samples the input signal to obtain a sampled signal; The quantization module is used to perform quantization processing on the sampled signal in at least two quantization modes to obtain at least two digital signals, where the values ​​of elements in different digital signals are different; The input module is used to input the voltage pulse signals corresponding to the elements in the different digital signals into the reserve pools constructed by different numbers of virtual nodes, so that different reserve pools can extract the signal characteristics of the input signals in different quantization modes.

7. The device according to claim 6, characterized in that The at least two digital signals include: a first digital signal and a second digital signal, and the input module is specifically used for: sequentially inputting voltage pulse signals corresponding to a first number of elements in the first digital signal into a first reserve pool constructed by a first number of virtual nodes; and, The voltage pulse signals corresponding to the second number of elements in the second digital signal are sequentially input into a second reserve pool constructed by a second number of virtual nodes.

8. The device according to claim 6, characterized in that The at least two quantization modes include at least a first quantization mode and a second quantization mode, the first quantization mode corresponds to a first bit number, the second quantization mode corresponds to a second bit number, the sampling signal is quantized by the at least two quantization modes to obtain at least two digital signals, and the quantization module is specifically used for: Determine a first value interval according to the maximum value, the minimum value and the first number of bits of the sampled signal; Obtaining a first digital signal according to a corresponding relationship between the amplitude of the sampling signal and the first value interval; Determine a second value interval according to the maximum value, the minimum value, and the second number of bits of the sampled signal; A second digital signal is obtained according to the corresponding relationship between the amplitude of the sampling signal and the second value interval.

9. A storage pool computing network optimization device, characterized in that: The device comprises: a memory and a processor; The memory is used to store computer programs; The processor is used to implement the steps of the reserve pool computing network optimization method as described in any one of claims 1 to 5 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the reserve pool computing network optimization method according to any one of claims 1 to 5 are implemented.

Images