WO2020118555A1

WO2020118555A1 - Network model data access method and device and electronic device

Info

Publication number: WO2020118555A1
Application number: PCT/CN2018/120563
Authority: WO
Inventors: 徐欣
Original assignee: 深圳鲲云信息科技有限公司
Priority date: 2018-12-12
Filing date: 2018-12-12
Publication date: 2020-06-18
Also published as: CN111542818A; CN111542818B

Abstract

Disclosed are a network model data access method and device and an electronic device. The method comprises the following steps: obtaining input data and configuration parameters, the configuration parameters including configuration parameters of each network layer (101); storing the input data and the configuration parameters into an analog memory, the analog memory including an array and an array index (102); reading configuration parameters of a corresponding network layer in the analog memory, and configuring a calculation engine of the corresponding network layer according to the configuration parameters (103); reading the input data of the analog memory into the calculation engine for calculation to obtain a calculation result (104); and storing the calculation result in the analog memory (105). By storing input data and calculation results in an analog memory, the method uses index arrays to improve the hit rate of data reading, thereby increasing the speed of data reading and further improving the data processing speed of the entire network model.

Description

Network model data access method, device and electronic equipment

Technical field

The present invention relates to the field of data processing, and more specifically, to a network model data access method, device, and electronic equipment.

Background technique

With the development of AI technology, the requirements for complex deep learning network models are higher. In the deep learning network model, the data transmission volume of a layer of network is very large. In order to ensure the correct rate of hardware on-board testing, it is necessary to pass many or All network layers are simulated, including reading data and storing calculation results in memory. Because the amount of data in each layer of network is very large, each layer of network needs to transmit a large amount of data during the simulation process. The amount of data transferred in the regression test even reaches hundreds of millions of data, so there is a problem that the data reading speed is slow, which makes the verification test time very long.

Summary of the invention

The purpose of the present invention is to provide a method, device and electronic device for accessing network model data in response to the above-mentioned defects in the prior art, and to solve the problem of slow reading speed of network model data.

The purpose of the present invention is achieved by the following technical solutions:

In a first aspect, a method for accessing network model data is provided. The method includes:

Obtain input data and configuration parameters, including the configuration parameters of each network layer;

Store the input data and configuration parameters in an analog memory in the form of an index array;

Reading configuration parameters of the corresponding network layer from the analog memory, and configuring a computing engine of the corresponding network layer according to the configuration parameters;

Reading the input data in the simulation memory to the calculation engine for calculation, and obtaining a calculation result;

The calculation result is stored in the simulation memory.

Optionally, after storing the simulation result in the simulation memory, the method further includes:

The calculation result in the simulation memory is read and compared with the historical calculation result in the database corresponding to the input data and configuration parameters to obtain a verification result.

Optionally, the input data includes image data acquired by a peripheral device, and the input data in the analog memory includes image data acquired by a peripheral device or a calculation result corresponding to a previous network layer.

Optionally, the calculation result of reading the simulation memory is compared with the historical calculation result of the input data and configuration parameters in the database to obtain a verification result, including:

Obtain the historical calculation results of the input data and configuration parameters corresponding to each network layer in the database and compare and verify the calculation results of each network layer; or

Obtain the historical calculation results of the input data and configuration parameters corresponding to the entire network model in the database, and compare and verify the final calculation results of all network layers.

Optionally, the calculation engine includes any one of the following:

Calculation engine based on field programmable gate array FPGA;

ASIC-based calculation engine;

Computing engine based on graphics processor GPU.

Optionally, after storing the calculation result in the simulation memory, the method further includes:

Reading configuration parameters corresponding to the new network layer from the simulation memory, and updating the computing engine corresponding to the network layer according to the configuration parameters;

Using the calculation result of the previous network layer in the analog memory as input data and reading to the updated calculation engine for calculation to obtain a new calculation result;

Store the new calculation result in the simulation memory;

The above steps will be repeated until the final calculation results obtained after calculating all the network layers are stored in the simulation memory.

In a second aspect, a network model data access device is provided. The device includes:

An acquisition module, for acquiring input data and configuration parameters, the configuration parameters including configuration parameters of each network layer;

An analog storage module, which is used to store the input data and configuration parameters into an analog memory in the form of an index array;

The simulation configuration module is configured to read the configuration parameters of the corresponding network layer from the simulation memory, and configure the calculation engine of the corresponding network layer according to the configuration parameters;

A calculation module, configured to read the input data in the simulation memory to the calculation engine for calculation, and obtain a calculation result;

The result storage module is used to store the calculation result in the simulation memory.

According to a third aspect, an electronic device is provided, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, and when the processor executes the computer program, the implementation of the present invention is implemented Example provides steps in the network model data access method.

According to a fourth aspect, a computer-readable storage medium is provided, and a computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, a computer model storage method provided in an embodiment of the present invention is implemented. step.

The beneficial effects brought by the present invention are: obtaining input data and configuration parameters, the configuration parameters including configuration parameters of each network layer; storing the input data and configuration parameters in a simulation memory, the simulation memory including an array and an array index Reading the configuration parameters of the corresponding network layer from the simulation memory, and configuring the calculation engine of the corresponding network layer according to the configuration parameters; reading the input data in the simulation memory to the calculation engine for calculation to obtain the calculation Results; store the calculation results in the simulation memory. By storing the input data and calculation results in the simulation memory, in the simulation memory, by indexing to the array method, the hit rate of data reading can be improved, thereby increasing the speed of data reading, and thereby increasing the data processing speed of the entire network model .

BRIEF DESCRIPTION

FIG. 1 is a schematic flowchart of a network model data access method according to an embodiment of the present invention;

2 is a schematic flowchart of another network model data access method according to an embodiment of the present invention;

3 is a schematic diagram of data access of a verification network layer according to an embodiment of the invention;

4 is a schematic structural diagram of a network model data access device according to an embodiment of the present invention;

5 is a schematic structural diagram of another network model data access device according to an embodiment of the present invention;

detailed description

The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, but not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

The terms "first", "second", "third", and "fourth" in the description and claims of the present application and the accompanying drawings are used to distinguish different objects, not to describe a specific order . In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally also includes Other steps or units inherent to these processes, methods, products or equipment.

Reference herein to "embodiments" means that specific features, structures, or characteristics described in connection with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various places in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment mutually exclusive with other embodiments. Those skilled in the art understand explicitly and implicitly that the embodiments described herein can be combined with other embodiments.

Since the advent of mathematical methods for simulating human actual neural networks, people have gradually become accustomed to calling this artificial neural network directly as a neural network. Neural networks have broad and attractive prospects in the fields of system identification, pattern recognition, and intelligent control. Especially in intelligent control, people are particularly interested in the self-learning function of neural networks, and regard this important feature of neural networks as One of the key keys to solve the problem of controller adaptability in automatic control.

Neural Networks (NN) is a complex network system formed by a large number of simple processing units (called neurons) that are widely connected to each other. It reflects many basic characteristics of human brain function and is a highly complex Non-linear dynamic learning system. Neural networks have large-scale parallelism, distributed storage and processing, self-organization, self-adaptive and self-learning capabilities, and are particularly suitable for processing inaccurate and fuzzy information processing problems that need to consider many factors and conditions simultaneously. The development of neural networks is related to neuroscience, mathematical science, cognitive science, computer science, artificial intelligence, information science, cybernetics, robotics, microelectronics, psychology, optical computing, molecular biology, etc. Interdisciplinary.

The basis of neural networks is neurons.

Neurons are biological models based on nerve cells of the biological nervous system. When people study the biological nervous system to explore the mechanism of artificial intelligence, the neurons are mathematicalized to produce a mathematical model of neurons.

A large number of neurons in the same form are connected together to form a neural network. Neural network is a highly nonlinear dynamic system. Although the structure and function of each neuron are not complicated, the dynamic behavior of the neural network is very complicated; therefore, the neural network can express various phenomena in the actual physical world.

The neural network model is described based on the mathematical model of the neuron. Artificial neural network (Artificial Neural Network) is a description of the first-order characteristics of the human brain system. Simply put, it is a mathematical model. The neural network model is represented by the characteristics of network topology nodes and learning rules. The great attraction of the neural network to people mainly includes: parallel distributed processing, high robustness and fault tolerance, distributed storage and learning capabilities, and can fully approximate complex nonlinear relationships.

In the research topics in the field of control, the control of uncertain systems has long been one of the central themes of control theory research, but this problem has not been effectively solved. Using the learning ability of the neural network, it can automatically learn the characteristics of the system in the process of controlling the uncertain system, so as to automatically adapt to the characteristics of the system over time, in order to achieve optimal control of the system; obviously this is a kind of Very exciting intentions and methods.

There are dozens of models of artificial neural networks, and the typical neural network models with more applications include BP neural network, Hopfield network, ART network and Kohonen network.

The invention provides a network model data access method, device and electronic equipment.

In the first aspect, please refer to FIG. 1. FIG. 1 is a schematic flowchart of a network model data access method according to an embodiment of the present invention. As shown in FIG. 1, the method includes the following steps:

101. Obtain input data and configuration parameters, where the configuration parameters include configuration parameters of each network layer.

The aforementioned input data may be acquired by an external device, the aforementioned input data may be image data, and the aforementioned external device may be a sensor, such as a camera, or a storage device storing image data, such as a mobile hard disk or a database server; The above input data can also be generated internally, such as image software. The above configuration parameters include configuration parameters of each layer of the network. The above-mentioned each layer of the network refers to each layer of the network in the network model. The above configuration parameters may be configuration parameters used by the network model during training, such as the convolutional layer Parameters such as weight parameters, convolution kernel parameters, and step parameters. The above-mentioned network model refers to a pre-trained network model. Training data can be obtained from various databases and obtained after data processing. Data processing can be processing such as cropping, compression, and data cleaning.

102. Store the input data and configuration parameters in an analog memory, where the analog memory includes an array and an array index.

The above-mentioned analog memory may be a virtual memory, and the input data and configuration parameters are stored in the analog memory, and the speed of reading and writing can be increased when each network layer reads input data or writes calculation data. The above array is used to store data, and the above array index is used to index into the array, thereby improving the reading speed of data. Through the analog memory to access the input data and calculation data, the hit rate of data reading can be improved, thereby increasing the speed of data access. In some possible embodiments, the above input data may also be calculation results obtained by calculating the same input data based on a network with the same configuration parameters as the upper layer network, for example: suppose that the network model 01 needs to be verified and tested The configuration parameter of the first layer is A, the input data is B, and the calculated result is C1. Assuming that there is a network layer with the same configuration parameter A as the network of this layer, it belongs to a layer in a mature network model 02. This layer of network The calculation result of the input data B is C2 (C2 can be regarded as the expected calculation result), and the calculation result C2 of the layer network and the input data B are read into the above-mentioned analog memory as input data, which can be understood as, input Data B is the initial input data in the network model 01 to be verified and tested, and C1 is calculated in the first-layer network, and the next-layer network of the first-layer network, which is the second-layer network, can read C1 for calculation , Can also read C2 to calculate. In this way, it is possible to prevent the problem that the subsequent calculation deviation due to the calculation deviation of a certain layer in the verification test is too large, and then the layer in which the deviation first occurs or all layers in which the deviation occurs cannot be found.

103. Read the configuration parameters of the corresponding network layer from the simulation memory, and configure the computing engine of the corresponding network layer according to the configuration parameters.

The above configuration parameters may be preset parameters. The above configuration parameters correspond to each network layer. In the network model, the algorithm of each layer of the network may be different, and the input data requirements and processing methods may also be different. The same, therefore, each layer of the network can be configured with different parameters to achieve the algorithm of each layer of network, the specific parameters that need to be configured can be set according to different models or different algorithms. The above calculation engine is used to calculate the input data according to the configuration parameters.

104. Read the input data in the analog memory to the calculation engine for calculation, and obtain a calculation result.

The foregoing input data may be understood as input data required by each layer of the network, may be input data of the first layer of the network model, or may be data of calculation results of the network of the upper layer. When storing the input data in the above analog memory, the input data of different layers can be stored in different storage areas, that is, the input data of different layers can be stored in different arrays or different arrays for storage, and the These arrays establish corresponding indexes. The input data can be read in the analog memory through the computing engine of the corresponding network layer.

105. Store the calculation result in the simulation memory.

In step 104, after calculating the input data, the calculation result is obtained, and the obtained calculation result can be returned to the simulation memory, stored in a new storage area, and a new index is established to index the data of the calculation result. The above new storage area may be a new array or a new subscript range in an array. In this way, after the next network layer is ready, the calculation results can be quickly read through the simulation memory for calculation, thereby increasing the data access speed of the entire network model, and thus the simulation speed.

In this embodiment, input data and configuration parameters are obtained, and the configuration parameters include configuration parameters of each network layer; the input data and configuration parameters are stored in an analog memory, and the analog memory includes an array and an array index; from Reading the configuration parameters of the corresponding network layer from the simulation memory, and configuring the calculation engine of the corresponding network layer according to the configuration parameters; reading the input data from the simulation memory to the calculation engine for calculation, and obtaining the calculation result; The calculation result is stored in the simulation memory. By storing the input data and calculation results in the simulation memory, in the simulation memory, by indexing to the array method, the hit rate of data reading can be improved, thereby increasing the speed of data reading, and thereby increasing the data processing speed of the entire network model .

It should be noted that the above calculation result may be referred to as a simulation result, and the network model data access method provided in the embodiment of the present invention may be applied to devices that need to perform data access, such as computers, servers, mobile phones, etc., which can perform network model data Accessed equipment.

Please refer to FIG. 2. FIG. 2 is a schematic flowchart of another network model data access method according to an embodiment of the present invention. As shown in FIG. 2, the method includes the following steps:

201. Obtain input data and configuration parameters, where the configuration parameters include configuration parameters of each network layer;

202. Store the input data and configuration parameters in an analog memory, where the analog memory includes an array and an array index;

203. Read the configuration parameters of the corresponding network layer from the simulation memory, and configure the calculation engine of the corresponding network layer according to the configuration parameters;

204. Read the input data in the analog memory to the calculation engine for calculation, and obtain a calculation result;

205. Store the calculation result in the simulation memory;

206. Read and compare the calculation result in the simulation memory with the historical calculation result in the database corresponding to the input data and configuration parameters to obtain a verification result.

In step 205, the calculation result is stored in the analog memory, so that the next layer of network can increase the reading speed when reading the calculation result as input data. In step 206, the above calculation result may be the calculation result of a certain layer of the network or the final calculation result of the entire network model. It is understandable that the historical calculation results in the database refer to the same input data in The result obtained in the input network model under the same configuration parameter can be considered as the expected result under the configuration parameter and input data. For example, the calculation result obtained by calculating the input data A and the configuration parameter B on a certain layer of the network is C, the above historical calculation result may be a calculation result obtained by calculating the input data A and the configuration parameter B in a layer of the network, assuming C1, the above comparison verification is to compare the calculation result C with the calculation result C1; and For example, the calculation result calculated by the input data A and the configuration parameter B in the entire network model is D. The above historical calculation result may be the calculation calculated by the input data A and the configuration parameter B in a mature network model As a result, assuming that it is D1, the above comparison verification is to compare the calculation result D with the calculation result D1. The above historical calculation results are used to verify the network model. The above comparison verification results include the same but may be different. It is understandable that if the calculation results are different from the corresponding historical calculation results in the database, the hardware model of the network model is on board The test fails. If the calculation result is the same as the corresponding historical calculation result in the database, it indicates that the hardware test of the network model passed. It should be noted that the above database may be a remote database or a local database, and the calculation result in the database may be a calculation result obtained based on a mature network model, which is used to compare calculation results of data simulation detection. The above comparison and verification can be compared by the CPU processor. In some possible embodiments, in the case of comparing and verifying the calculation results of each layer of network, if the calculation results of one layer of network are different from the historical calculation results, the historical calculation results may be stored into the above-mentioned analog memory, When the next layer network starts to calculate, the historical calculation results are read as input data to the next layer network for calculation, so that the input data of each layer is expected to be input data, thereby improving the verification test Precision.

In this embodiment, by comparing the calculation result stored in the simulation memory to the processor and the historical data result in the database, the test on the hardware board is implemented, because the calculation result is read from the simulation memory to the processing Compared with the device, the data reading speed is fast, which reduces the simulation time and further reduces the verification test time.

The above-mentioned peripherals can be data collection devices equipped with memory, such as cameras, mobile phones, tablet computers, etc., or they can be hard disks for storing data, cloud databases, and other devices that provide input data, and read the data in the above devices In the simulation memory, during the verification test, the input data can be read through the simulation memory to increase the data reading speed. The above-mentioned image data data may be image data for face recognition, image data for vehicle recognition, image data for object recognition, or the like. The above-mentioned simulation memory obtains input data and configuration parameters from peripheral devices. During simulation, the network model reads input data and configuration parameters from the simulation memory. In some possible embodiments, the above-mentioned simulation memory may also be obtained from a database The historical calculation results are stored as input data. The input data obtained from the peripherals can be regarded as the initial input data. After the initial input data is calculated through a layer of networks, the obtained calculation results can be input as input data to the next layer of networks for calculation.

Optionally, the calculation result of reading the simulation memory is compared with the historical calculation result corresponding to the input data and configuration parameters in the database to obtain a verification result, including:

In this embodiment, please refer to FIG. 3, when the calculation results of each network layer in the network model need to be verified, the above database may store historical calculation results corresponding to each network layer (expected calculation results) ), in the verification test of the network model, you can store the calculation result in the simulation memory after each layer is calculated, and then read the calculation result from the simulation memory and put it into the processor, and the processor gets it into the database Compare the historical calculation results (expected calculation results) of this layer network. Each layer of network is calculated once, and a comparison is performed. In this way, the verification test of each layer of network can improve the accuracy of the verification test. When the final calculation result of the network model needs to be verified, the above database can store the historical calculation results (expected calculation results) corresponding to the entire network model, and can wait for all the network layers in the network model to complete the calculation and then calculate the calculation results. Compare it with the historical calculation results (expected calculation results) in the database to the processor.

Optionally, the calculation engine includes any one of the following:

Calculation engine based on field programmable gate array FPGA;

ASIC-based calculation engine;

Computing engine based on graphics processor GPU.

The above-mentioned field programmable gate array makes the design of digital circuit system very flexible, and significantly shortens the development cycle of the system, reduces the volume of the digital circuit system and the types of chips used, and can be used for image acquisition or image recognition. The above-mentioned dedicated integrated circuit has the characteristics of strong specificity, and can be dedicated to the verification test of the network model hardware after being designed. The above-mentioned graphics processor has a graphics acceleration function, which can process the calculation of complex images, liberate the CPU processing from the calculation, improve the calculation of image data, and also increase the speed of comparison of the calculation results by the CPU processor. Of course, the above calculation engine is only a better choice in this embodiment, and should not be construed as limiting this embodiment.

After the storing the calculation result in the simulation memory, the method further includes:

Store the new calculation result in the simulation memory;

After the calculation of one layer of network is completed, the calculation of the next layer of network will be carried out. The above new network layer refers to the next network layer. The network layer configuration parameters in the network model can be configured at one time or at the beginning of the calculation Before configuring, for example, you can start to configure the parameters of the next layer of network after the calculation of one layer of network is completed. It can be understood that the parameters of each layer of network are stored in different arrays in the above-mentioned simulation memory, which can be simulated The corresponding index in the memory is obtained. Repeat the calculation process to calculate all the network layers in the network model until the last layer of network calculations. The final calculation results are obtained and stored in the simulation memory, which is convenient for taking out the historical calculation results in the database (expected calculation results) comparing.

In a second aspect, as shown in FIG. 4, a network model data access device is provided. The device includes:

The obtaining module 301 is used to obtain input data and configuration parameters, where the configuration parameters include configuration parameters of each network layer;

The simulation storage module 302 is used to store the input data and configuration parameters in the form of an index array into the simulation memory;

The simulation configuration module 303 is configured to read the configuration parameters of the corresponding network layer from the simulation memory, and configure the calculation engine of the corresponding network layer according to the configuration parameters;

The calculation module 304 is configured to read the input data in the analog memory to the calculation engine for calculation, and obtain a calculation result;

The result storage module 305 is used to store the calculation result in the simulation memory.

Optionally, as shown in FIG. 5, the device further includes:

The verification module 306 is configured to read and compare the calculation result in the simulation memory with the historical calculation result corresponding to the input data and configuration parameters in the database to obtain a verification result.

Optionally, the verification module 306 is further configured to obtain historical calculation results of input data and configuration parameters corresponding to each network layer in the database and compare and verify the calculation results of each network layer respectively; or

The verification module 306 is also used to obtain historical calculation results of input data and configuration parameters corresponding to the entire network model in the database, and perform comparison and verification on final calculation results of all network layers.

Optionally, the calculation engine includes any one of the following:

Calculation engine based on field programmable gate array FPGA;

ASIC-based calculation engine;

Computing engine based on graphics processor GPU.

Optionally, the simulation configuration module 303 is further configured to read configuration parameters corresponding to the new network layer from the simulation memory, and update the calculation engine corresponding to the network layer according to the configuration parameters;

The calculation module 304 is further configured to use the calculation result of the previous network layer in the simulation memory as input data and read the updated calculation engine to perform calculation to obtain a new calculation result;

The result storage module 305 is also used to store the new calculation result in the simulation memory;

The result storage module 305 is also used to store the final calculation result obtained by repeating the above steps until all network layers are calculated into the simulation memory.

In a third aspect, an embodiment of the present invention provides an electronic device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, when the processor executes the computer program The steps in the installation of the container orchestration engine provided by the embodiments of the present invention are implemented.

According to a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium storing a computer program on the computer-readable storage medium, which when executed by a processor implements a container orchestration engine provided by an embodiment of the present invention Steps in installation.

It should be noted that, for the sake of simple description, the foregoing method embodiments are all expressed as a series of action combinations, but those skilled in the art should know that this application is not limited by the described action sequence, Because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the involved actions and modules are not necessarily required by this application.

In the above embodiments, the description of each embodiment has its own emphasis. For a part that is not detailed in an embodiment, you can refer to related descriptions in other embodiments.

In the several embodiments provided in this application, it should be understood that the disclosed device may be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the unit is only a logical function division. In actual implementation, there may be another division manner, for example, multiple units or components may be combined or may Integration into another system, or some features can be ignored, or not implemented. In addition, the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical or other forms.

The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit. The above integrated unit can be implemented in the form of hardware or software program modules.

If the integrated unit is implemented in the form of a software program module and sold or used as an independent product, it may be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application essentially or part of the contribution to the existing technology or all or part of the technical solution can be embodied in the form of a software product, the computer software product is stored in a memory, Several instructions are included to enable a computer device (which may be a personal computer, server, network device, etc.) to perform all or part of the steps of the methods described in the embodiments of the present application. The aforementioned memory includes: U disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

A person of ordinary skill in the art may understand that all or part of the steps in the various methods of the above embodiments may be completed by a program instructing relevant hardware. The program may be stored in a computer-readable memory, and the memory may include: a flash disk , Read-Only Memory (English: Read-Only Memory, abbreviation: ROM), Random Access Device (English: Random Access Memory, abbreviation: RAM), magnetic disk or optical disk, etc.

The embodiments of the present application are described in detail above, and specific examples are used to explain the principles and implementation of the present application. The descriptions of the above embodiments are only used to help understand the method and core idea of the present application; at the same time, Those of ordinary skill in the art, based on the ideas of the present application, will have changes in specific implementations and application scopes. In summary, the content of this specification should not be construed as limiting the present application.

Claims

A network model data access method, characterized in that the method includes:

Obtain input data and configuration parameters, including the configuration parameters of each network layer;

Store the input data and configuration parameters in an analog memory, the analog memory includes an array and an array index;

Reading configuration parameters of the corresponding network layer from the analog memory, and configuring a computing engine of the corresponding network layer according to the configuration parameters;

Reading the input data in the simulation memory to the calculation engine for calculation, and obtaining a calculation result;

The calculation result is stored in the simulation memory.
The method according to claim 1, wherein after the storing the simulation result in the simulation memory, the method further comprises:

The calculation result in the simulation memory is read and compared with the historical calculation result in the database corresponding to the input data and configuration parameters to obtain a verification result.
The method according to claim 2, wherein the input data includes image data acquired by a peripheral device, and the input data in the analog memory includes image data acquired by a peripheral device or corresponding to the previous network layer. Calculation results.
The method according to claim 3, wherein the calculation result of reading the simulation memory is compared with the historical calculation result corresponding to the input data and configuration parameters in the database to obtain the verification result, including:

Obtain the historical calculation results of the input data and configuration parameters corresponding to each network layer in the database and compare and verify the calculation results of each network layer; or

Obtain the historical calculation results of the input data and configuration parameters corresponding to the entire network model in the database, and compare and verify the final calculation results of all network layers.
The method according to any one of claims 1 to 4, wherein the calculation engine includes any one of the following:

Calculation engine based on field programmable gate array FPGA;

ASIC-based calculation engine;

Computing engine based on graphics processor GPU.
The method according to claim 2, wherein after the storing the calculation result in the simulation memory, the method further comprises:

Reading configuration parameters corresponding to the new network layer from the simulation memory, and updating the computing engine corresponding to the network layer according to the configuration parameters;

Using the calculation result of the previous network layer in the analog memory as input data and reading to the updated calculation engine for calculation to obtain a new calculation result;

Store the new calculation result in the simulation memory;

The above steps will be repeated until the final calculation results obtained after calculating all the network layers are stored in the simulation memory.
A network model data access device, characterized in that the device includes:

An acquisition module, for acquiring input data and configuration parameters, the configuration parameters including configuration parameters of each network layer;

An analog storage module, which is used to store the input data and configuration parameters into an analog memory in the form of an index array;

The simulation configuration module is configured to read the configuration parameters of the corresponding network layer from the simulation memory, and configure the calculation engine of the corresponding network layer according to the configuration parameters;

A calculation module, configured to read the input data in the simulation memory to the calculation engine for calculation, and obtain a calculation result;

The result storage module is used to store the calculation result in the simulation memory.
The device of claim 7, wherein the device further comprises:

The comparison and verification module is used for reading and comparing the calculation results in the simulation memory with the historical calculation results corresponding to the input data and configuration parameters in the database to obtain the verification results.
An electronic device, comprising: a memory, a processor, and a computer program stored on the memory and executable on the processor, and the processor implements the computer program as claimed in claim 1 Steps 6 to 6 in the network data access method based on target detection.
A computer-readable storage medium, characterized in that a computer program is stored on the computer-readable storage medium, and when the computer program is executed by a processor, the network model according to any one of claims 1 to 6 is implemented Steps in the data access method.