Neural network structure search method, terminal, server and readable storage medium

Abstract

The present invention discloses a method for searching a neural network structure, a terminal, a server and a readable storage medium, the method comprising: the server determines whether a first parameter corresponding to the neural network structure is stored; if it is determined that the first parameter is stored, a training instruction carrying the first parameter is sent to the terminal, so that the terminal searches for a target neural network structure according to the first parameter and trains the target neural network structure using training data. The present invention improves the resource utilization rate of the method for searching a neural network structure.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a method, terminal, server and readable storage medium for searching a neural network structure. Background Art

[0002] Existing methods for searching neural network structures all perform training on the terminal, and then send the training results to the server for searching the neural network structure. For example, the server needs to search the neural network structure for each terminal separately. While the server is searching for the neural network structure, some terminals will wait for the server to send the neural network structure, resulting in unreasonable utilization of terminal resources. Summary of the invention

[0003] The embodiments of the present invention provide a search method, terminal, server and readable storage medium for a neural network structure, aiming to solve the technical problem of unreasonable resource utilization of the terminal.

[0004] An embodiment of the present invention provides a method for searching a neural network structure, wherein the method for searching a neural network structure is applied to a terminal, wherein the terminal and a server are in the same distributed architecture, and the method for searching a neural network structure includes:

[0005] Receiving a training instruction sent by the server;

[0006] Acquire a first parameter corresponding to the neural network structure according to the training instruction;

[0007] Searching for a neural network structure according to the first parameter to obtain a target neural network structure corresponding to the first parameter;

[0008] The target neural network structure is trained using training data.

[0009] In one embodiment, after the step of receiving the training instruction sent by the server, the method further includes:

[0010] Parsing the training instruction to obtain data information;

[0011] If the data information includes a first parameter corresponding to the neural network structure, executing the step of acquiring the first parameter corresponding to the neural network structure according to the training instruction;

[0012] If the data information includes a preset neural network structure, the preset neural network structure is trained using training data.

[0013] In one embodiment, if the data information includes a preset neural network structure, after the step of training the preset neural network structure using training data, the method further includes:

[0014] Obtaining a second parameter corresponding to the trained target neural network structure;

[0015] The second parameter is sent to the server, so that the server generates a training instruction based on the second parameter.

[0016] In one embodiment, if the data information includes a preset neural network structure, after the step of training the preset neural network structure using training data, the method further includes:

[0017] The trained target neural network structure is obtained and sent to the server and the shared storage queue, wherein the terminal searches the shared storage queue for the target neural network corresponding to the first parameter.

[0018] An embodiment of the present invention provides a method for searching a neural network structure. The method for searching a neural network structure is applied to a server. The server and a terminal are in the same distributed architecture. The method for searching a neural network structure includes:

[0019] Obtaining a first parameter corresponding to a preset neural network structure;

[0020] generating a training instruction according to the first parameter;

[0021] The training instruction is sent to each terminal, wherein the terminal obtains the first parameter according to the training instruction, searches for the neural network structure according to the first parameter to obtain the target neural network structure corresponding to the first parameter, and uses the training data to train the target neural network structure.

[0022] In one embodiment, after the step of obtaining the first parameter corresponding to the preset neural network structure, the method further includes:

[0023] Determining whether to store a first parameter corresponding to a preset neural network structure;

[0024] If it is determined that the first parameter corresponding to the neural network structure is not stored, generating a training instruction according to the preset neural network structure, and sending the training instruction to the terminal, so that the terminal uses the training data to train the preset neural network structure;

[0025] If it is determined that the first parameter corresponding to the neural network structure is not stored, the step of obtaining the first parameter corresponding to the neural network structure is performed.

[0026] In one embodiment, after the step of sending the training instruction to each terminal, the method further includes:

[0027] Receiving the trained target neural network structure and a second parameter corresponding to the trained target neural network structure sent by the terminal;

[0028] The preset neural network and the first parameters are updated according to the trained target neural network structure and the second parameters.

[0029] An embodiment of the present invention also provides a terminal, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the various steps of the method for searching the neural network structure as described above when executing the computer program.

[0030] An embodiment of the present invention also provides a server, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the various steps of the method for searching the neural network structure as described above when executing the computer program.

[0031] An embodiment of the present invention further provides a readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the various steps of the method for searching a neural network structure as described above are implemented.

[0032] In the technical solution of this embodiment, the terminal receives the training instruction sent by the server; obtains the first parameter corresponding to the neural network structure according to the training instruction; searches for the neural network structure according to the first parameter to obtain the target neural network structure corresponding to the first parameter; and trains the target neural network structure using training data. Since the terminal carries the work of searching the neural network that should be performed on the server, the server only needs to send the first parameter corresponding to the target neural network structure to realize the search of the neural network structure, so that the terminal resources that should be idle are reasonably utilized. Therefore, the present invention improves the resource utilization rate of the method for searching the neural network structure. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention 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 invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0034] Figure 1 A schematic diagram of the hardware architecture of a terminal involved in an embodiment of the present invention;

[0035] Figure 2 A schematic diagram of the hardware architecture of a server involved in an embodiment of the present invention;

[0036] Figure 3A schematic diagram of a flow chart of a first embodiment of a method for searching a neural network structure according to the present invention;

[0037] Figure 4 A schematic diagram of a flow chart of a second embodiment of a method for searching a neural network structure according to the present invention;

[0038] Figure 5 A schematic diagram of a flow chart of a third embodiment of a method for searching a neural network structure according to the present invention;

[0039] Figure 6 A schematic diagram of a flow chart of a fourth embodiment of a method for searching a neural network structure according to the present invention;

[0040] Figure 7 A schematic diagram of a flow chart of a fifth embodiment of a method for searching a neural network structure according to the present invention;

[0041] Figure 8 A schematic diagram of a flow chart of a sixth embodiment of a method for searching a neural network structure according to the present invention;

[0042] Fig. 9 A flowchart of a seventh embodiment of a method for searching a neural network structure according to the present invention;

[0043] Fig.10 This is a timing diagram of the interaction between the terminal and the server involved in the eighth embodiment of the method for searching the neural network structure of the present invention. DETAILED DESCRIPTION

[0044] In order to better understand the above technical solution, exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments described herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0045] The main solution of the present invention is: the terminal receives the training instruction sent by the server; obtains the first parameter corresponding to the neural network structure according to the training instruction; searches for the neural network structure according to the first parameter to obtain the target neural network structure corresponding to the first parameter; and trains the target neural network structure using training data.

[0046] Since the terminal carries the work of searching the neural network that should be performed on the server, the server only needs to send the first parameter corresponding to the target neural network structure to realize the search of the neural network structure, so that the terminal resources that should be idle are reasonably utilized. Therefore, the present invention improves the resource utilization rate of the neural network structure search method.

[0047] As an implementation method, the terminal can Figure 1 .

[0048] The embodiment of the present invention relates to a terminal, which includes: a processor 101, such as a CPU, a memory 102, and a communication bus 103. The communication bus 103 is used to realize connection and communication between these components.

[0049] The memory 102 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. Figure 1 , the memory 103 as a computer storage medium may include a detection program; and the processor 101 may be used to call the detection program stored in the memory 102 and perform the following operations:

[0050] Receiving a training instruction sent by the server;

[0051] Acquire a first parameter corresponding to the neural network structure according to the training instruction;

[0052] Searching for a neural network structure according to the first parameter to obtain a target neural network structure corresponding to the first parameter;

[0053] The target neural network structure is trained using training data.

[0054] In one embodiment, the processor 101 may be used to call the detection program stored in the memory 102 and perform the following operations:

[0055] Parsing the training instruction to obtain data information;

[0056] If the data information includes a first parameter corresponding to the neural network structure, executing the step of acquiring the first parameter corresponding to the neural network structure according to the training instruction;

[0057] If the data information includes a preset neural network structure, the preset neural network structure is trained using training data.

[0058] In one embodiment, the processor 101 may be used to call the detection program stored in the memory 102 and perform the following operations:

[0059] Obtaining a second parameter corresponding to the trained target neural network structure;

[0060] The second parameter is sent to the server, so that the server generates a training instruction based on the second parameter.

[0061] In one embodiment, the processor 101 may be used to call the detection program stored in the memory 102 and perform the following operations:

[0062] The trained target neural network structure is obtained and sent to the server and the shared storage queue, wherein the terminal searches the shared storage queue for the target neural network corresponding to the first parameter.

[0063] In the technical solution of this embodiment, the terminal receives the training instruction sent by the server; obtains the first parameter corresponding to the neural network structure according to the training instruction; searches for the neural network structure according to the first parameter to obtain the target neural network structure corresponding to the first parameter; and trains the target neural network structure using training data. Since the terminal carries the work of searching the neural network that should be performed on the server, the server only needs to send the first parameter corresponding to the target neural network structure to realize the search of the neural network structure, so that the terminal resources that should be idle are reasonably utilized. Therefore, the present invention improves the resource utilization rate of the method for searching the neural network structure.

[0064] As an implementation method, the terminal can Figure 2 .

[0065] The embodiment of the present invention relates to a server, which includes: a processor 201, such as a CPU, a memory 202, and a communication bus 203. The communication bus 203 is used to realize connection and communication between these components.

[0066] The memory 202 may be a high-speed RAM memory or a stable memory (non-volatile memory), such as a disk memory. Figure 1 , the memory 203 as a computer storage medium may include a detection program; and the processor 201 may be used to call the detection program stored in the memory 202 and perform the following operations:

[0067] Obtaining a first parameter corresponding to a preset neural network structure;

[0068] generating a training instruction according to the first parameter;

[0069] The training instruction is sent to each terminal, wherein the terminal obtains the first parameter according to the training instruction, searches for the neural network structure according to the first parameter to obtain the target neural network structure corresponding to the first parameter, and uses the training data to train the target neural network structure.

[0070] In one embodiment, the processor 201 may be used to call the detection program stored in the memory 202 and perform the following operations:

[0071] Determining whether to store a first parameter corresponding to a preset neural network structure;

[0072] If it is determined that the first parameter corresponding to the neural network structure is not stored, generating a training instruction according to the preset neural network structure, and sending the training instruction to the terminal, so that the terminal uses the training data to train the preset neural network structure;

[0073] If it is determined that the first parameter corresponding to the neural network structure is not stored, the step of obtaining the first parameter corresponding to the neural network structure is performed.

[0074] In one embodiment, the processor 201 may be used to call the detection program stored in the memory 202 and perform the following operations:

[0075] Receiving the trained target neural network structure and a second parameter corresponding to the trained target neural network structure sent by the terminal;

[0076] The preset neural network and the first parameters are updated according to the trained target neural network structure and the second parameters.

[0077] In the technical solution of this embodiment, the server determines whether the first parameter corresponding to the neural network structure is stored; if it is determined that the first parameter is stored, a training instruction carrying the first parameter is sent to the terminal, so that the terminal searches for the target neural network structure according to the first parameter and trains the target neural network structure using training data. Since the server transfers the work of searching the neural network that should be performed on the server to the terminal, the server only needs to send the first parameter corresponding to the target neural network structure to realize the search of the neural network structure, so that the terminal resources that should be idle are reasonably utilized. Therefore, the present invention improves the resource utilization rate of the search method for the neural network structure.

[0078] In order to better understand the above technical solution, the above technical solution will be described in detail below in conjunction with the accompanying drawings and specific implementation methods.

[0079] Reference Figure 3 , Figure 3 The first embodiment of the method for searching a neural network structure of the present invention comprises the following steps:

[0080] Step S10, receiving a training instruction sent by the server.

[0081] In this embodiment, the terminal receives training instructions from the server, and the terminal includes electronic devices such as computers, which are not limited here. The training instructions are used to instruct the terminal device to obtain a neural network structure and perform training. Representative network structures include BP networks, RBF networks, Hopfield networks, self-organizing feature mapping networks, etc., which are not limited here.

[0082] Optionally, after obtaining the training command sent by the server, the terminal initializes the distributed framework to ensure that the network communication between the server and the terminal is normal. After the initialization operation, the server is responsible for task distribution to distribute the tasks to each terminal in the distributed architecture, and is responsible for the control of the terminals in the entire cluster by starting the socket background service, and then initiates training and search instructions to each terminal through SHH, where SSH is the abbreviation of Secure Shell, which is a security protocol based on the application layer and the transport layer. SSH is relatively reliable and is a protocol designed to provide security for remote login sessions and other network services. The use of the SSH protocol can effectively prevent information leakage during remote management. SSH was originally a program on the UNIX system and later quickly expanded to other operating platforms. SSH can fill in the gaps in the network when used correctly. SSH clients are available for multiple platforms. Almost all UNIX platforms - including HP-UX, Linux, AIX, Solaris, Digital UNIX, Irix, and other platforms - can run SSH.

[0083] Optionally, after receiving the SHH command, the terminal starts the training program and initializes it. The terminal binds the socket server according to the server's IP, and sends a request signal to the server through the socket to start training. After receiving the request signal, the server sends a training instruction. Specifically, the server determines whether there is a first parameter corresponding to the neural network structure. If not, it loads the prepared pre-trained network structure file as the preset neural network structure and adds it to the socket information. The information consists of a dictionary, which contains the network structure and the keys corresponding to the first parameter of the neural network structure. The server then sends the socket information to the terminal; if so, the first parameter corresponding to the neural network structure is sent to the terminal.

[0084] Step S20, obtaining a first parameter corresponding to the neural network structure according to the training instruction.

[0085] In this embodiment, the above-mentioned first parameter is a parameter corresponding to the neural network structure, which can be used to search for the neural network structure in the shared storage queue. Optionally, the neural network structure can be acquired or generated according to the first parameter.

[0086] Optionally, the terminal determines whether the socket information contains a preset neural network structure or a first parameter. If it is a preset neural network structure, the preset neural network structure is directly used for training.

[0087] Step S30, searching for a neural network structure according to the first parameter to obtain a target neural network structure corresponding to the first parameter.

[0088] In this embodiment, if it is the first parameter, the network search process is started to search the network structure according to the first parameter, and then the target neural network structure is read from the shared storage queue for training.

[0089] Step S40: training the target neural network structure using training data.

[0090] In this embodiment, the training data is a training set for training the target neural network structure. In the fields of machine learning, it is generally necessary to divide the samples into three independent parts: a training set, a validation set, and a test set. The training set is used to estimate the model, and the validation set is used to determine the network structure or control the parameters of the model complexity.

[0091] In the technical solution of this embodiment, the server determines whether the first parameter corresponding to the neural network structure is stored; if it is determined that the first parameter is stored, a training instruction carrying the first parameter is sent to the terminal, so that the terminal searches for the target neural network structure according to the first parameter and trains the target neural network structure using training data. Since the server transfers the work of searching the neural network that should be performed on the server to the terminal, the server only needs to send the first parameter corresponding to the target neural network structure to realize the search of the neural network structure, so that the terminal resources that should be idle are reasonably utilized. Therefore, the present invention improves the resource utilization rate of the search method for the neural network structure.

[0092] Reference Figure 4 , Figure 4 This is a second embodiment of the method for searching a neural network structure of the present invention, based on the first embodiment, and after step S10, further comprising:

[0093] Step S50, parsing the training instruction to obtain data information.

[0094] In this embodiment, the training instruction is parsed to obtain the data information it carries. Optionally, the data information is the above-mentioned socket information.

[0095] Step S60, if the data information includes a first parameter corresponding to the neural network structure, execute the step of obtaining the first parameter corresponding to the neural network structure according to the training instruction.

[0096] Step S70: If the data information includes a preset neural network structure, the preset neural network structure is trained using training data.

[0097] In this embodiment, the preset neural network structure is pre-set, and optionally, the preset neural network structure is used for the first round of neural network structure training. Specifically, the server will determine whether there is a first parameter corresponding to the neural network structure. If not, the prepared pre-trained network structure file is loaded as the preset neural network structure and added to the socket information, which is composed of a dictionary, including the network structure and the key corresponding to the first parameter corresponding to the neural network structure, and then the server sends the socket information to the terminal; if yes, the first parameter corresponding to the neural network structure is sent to the terminal.

[0098] In the technical solution of this embodiment, after setting the preset neural network structure, without storing the first parameter, the technical solution of this embodiment can form a closed loop that can automatically search and train the neural network structure, thereby improving the flexibility of the implementation of this embodiment.

[0099] Reference Figure 5 , Figure 5 The third embodiment of the method for searching a neural network structure of the present invention is based on the first or second embodiment, and after step S70, further includes:

[0100] Step S80, obtaining a second parameter corresponding to the trained target neural network structure.

[0101] In this embodiment, after the terminal uses the training data to train the preset neural network structure or the target neural network structure, the parameters of the training target will change. In theory, it becomes a new neural network structure.

[0102] Step S90: sending the second parameter to the server, so that the server generates a training instruction based on the second parameter.

[0103] In this embodiment, the second parameter may be stored, and in the next round of training, it may be used as the first parameter to search for a corresponding neural network in the shared storage queue for training.

[0104] In the technical solution of this embodiment, the parameters obtained by training are used for the next round of training, so that the technical solution of this embodiment can form a closed loop that can automatically search and train the neural network structure, thereby improving the flexibility of the implementation of this embodiment.

[0105] Reference Figure 6 , Figure 6 The fourth embodiment of the method for searching a neural network structure of the present invention is based on any one of the first to third embodiments, and after step S70, further includes:

[0106] Step S100, obtaining the trained target neural network structure and sending it to the server and the shared storage queue, wherein the terminal searches the shared storage queue for the target neural network corresponding to the first parameter.

[0107] In this embodiment, after completing training and verification with the specified data set on the terminal, the training and verification results including accuracy, loss, etc. are sent to the server; the socket service process receives and packages them into the first parameter on the terminal, which is used as the parameter source for the next round of neural network structure search; after the search is completed, the searched neural network structure is added to the shared storage queue for reading in the next round of training; after the current search and training are completed, the server initiates the search and training task again according to the end information returned by the terminal. Start a new round of training and search. Optionally, when the number of repeated training reaches the set number of iterations or running time, the main process of the server will wait for the latest round of training and search to end and then terminate the entire program.

[0108] In the technical solution of this embodiment, the target neural network structure is sent to the shared storage queue and the server for storage at the same time. When the terminal receives the first parameter of the target neural network model sent by the server, it can directly search and obtain the corresponding target neural network structure in the shared storage queue according to the first parameter without regenerating the target neural network structure. Therefore, the implementation efficiency of this embodiment is improved.

[0109] Reference Figure 7 , Figure 7 The fifth embodiment of the method for searching a neural network structure of the present invention is based on any one of the first to fourth embodiments, and includes:

[0110] Step S110, obtaining a first parameter corresponding to a preset neural network structure.

[0111] Step S120: generating a training instruction according to the first parameter.

[0112] Step S130, sending the training instruction to each terminal, wherein the terminal obtains the first parameter according to the training instruction, searches for the neural network structure according to the first parameter to obtain the target neural network structure corresponding to the first parameter, and uses the training data to train the target neural network structure.

[0113] In this embodiment, the server side queries whether the first parameter has been stored in the memory.

[0114] Optionally, if it is determined that the first parameter is stored, a training instruction carrying the first parameter is sent to the terminal, so that the terminal searches for a target neural network structure according to the first parameter and trains the target neural network structure using training data.

[0115] In this embodiment, the server determines whether there is a first parameter corresponding to the neural network structure. If not, it loads the prepared pre-trained network structure file as the preset neural network structure and adds it to the socket information, which consists of a dictionary containing the network structure and the keys corresponding to the first parameter corresponding to the neural network structure. The server then sends the socket information to the terminal; if yes, it sends the first parameter corresponding to the neural network structure to the terminal.

[0116] In the technical solution of this embodiment, the server determines whether the first parameter corresponding to the neural network structure is stored; if it is determined that the first parameter is stored, a training instruction carrying the first parameter is sent to the terminal, so that the terminal searches for the target neural network structure according to the first parameter and trains the target neural network structure using training data. Since the server assigns the work of neural network search to the terminal, the server only needs to send the first parameter corresponding to the target neural network structure to realize the search of the neural network structure, so that the terminal resources that should have been idle are reasonably utilized. Therefore, the present invention improves the resource utilization rate of the neural network structure search method.

[0117] Reference Figure 8 , Figure 8 The sixth embodiment of the method for searching a neural network structure of the present invention is based on any one of the first to fifth embodiments, and after step S110, further includes:

[0118] Step S140, determining whether a first parameter corresponding to a preset neural network structure is stored.

[0119] Step S150, if it is determined that the first parameter corresponding to the neural network structure is not stored, a training instruction is generated according to the preset neural network structure, and the training instruction is sent to the terminal, so that the terminal uses the training data to train the preset neural network structure.

[0120] Step S160: If it is determined that the first parameter corresponding to the neural network structure is not stored, the step of obtaining the first parameter corresponding to the neural network structure is performed.

[0121] In this embodiment, in this embodiment, the server determines whether there is a first parameter corresponding to the neural network structure. If not, it loads the prepared pre-trained network structure file as the preset neural network structure and adds it to the socket information, which consists of a dictionary containing the network structure and the keys corresponding to the first parameter corresponding to the neural network structure. The server then sends the socket information to the terminal; if yes, it sends the first parameter corresponding to the neural network structure to the terminal.

[0122] In the technical solution of this embodiment, after setting the preset neural network structure, without storing the first parameter, the technical solution of this embodiment can form a closed loop that can automatically search and train the neural network structure, thereby improving the flexibility of the implementation of this embodiment.

[0123] Reference Fig. 9 , Fig. 9 The seventh embodiment of the method for searching a neural network structure of the present invention is based on any one of the first to sixth embodiments, and after step S130, further includes:

[0124] Step S170, receiving the trained target neural network structure and the second parameter corresponding to the trained target neural network structure sent by the terminal.

[0125] Step S180, updating the preset neural network and the first parameters according to the trained target neural network structure and the second parameters.

[0126] In the technical solution of this embodiment, the server can extract parameters of the target neural network structure sent by the terminal and use them as the first parameters for the next round of training. Therefore, the technical solution of this embodiment can form a closed loop that can automatically search and train the neural network structure, thereby improving the flexibility of the implementation of this embodiment.

[0127] Please refer to Fig.10 , Fig.10 This is a timing diagram of interaction among a mobile terminal, a server and a lock body according to the eighth embodiment of the present invention.

[0128] In the eighth embodiment, the server sends a training instruction to the terminal, the terminal sends a training request to the server, the server determines whether the first parameter is stored, the server sends the first parameter to the terminal, the terminal trains the target neural network model according to the first parameter, the terminal stores and sends the trained target neural network model, the server stores the target neural network model, and the server initiates the next round of training instructions. Since the server has given the work of searching the neural network that should have been done on the server to the terminal, the server only needs to send the first parameter corresponding to the target neural network structure to realize the search of the neural network structure, so that the terminal resources that should have been idle are reasonably utilized. Therefore, the present invention improves the resource utilization rate of the search method for the neural network structure.

[0129] To achieve the above-mentioned purpose, an embodiment of the present invention further provides a terminal, which includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, each step of the method for searching the neural network structure as described above is implemented.

[0130] To achieve the above objectives, an embodiment of the present invention further provides a server, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the various steps of the method for searching the neural network structure as described above when executing the computer program.

[0131] To achieve the above objectives, an embodiment of the present invention further provides a readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the various steps of the method for searching the neural network structure as described above are implemented.

[0132] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0133] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0134] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0135] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0136] It should be noted that in the claims, any reference signs placed between brackets shall not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claim. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention may be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In a unit claim enumerating several means, several of these means may be embodied by the same item of hardware. The use of the words first, second, and third etc. does not indicate any order. These words may be interpreted as names.

[0137] Although the preferred embodiments of the present invention have been described, those skilled in the art may make other changes and modifications to these embodiments once they have learned the basic creative concept. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention.

[0138] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for searching a neural network structure, characterized in that: The neural network structure search method is applied to a terminal, the terminal and the server are in the same distributed architecture, and the neural network structure search method includes: Receiving a training instruction sent by the server; Acquire a first parameter corresponding to the neural network structure according to the training instruction; According to the first parameter, searching for a neural network structure in a shared storage queue to obtain a target neural network structure corresponding to the first parameter; The target neural network structure is trained using training data, and the server performs the neural network search, so that the server can search the neural network structure by sending the first parameter corresponding to the target neural network structure, thereby improving the terminal resource utilization rate; Acquire a trained target neural network structure, and send the target neural network structure to the server and the shared storage queue, wherein the terminal searches the shared storage queue for a target neural network corresponding to the updated first parameter; Obtaining a second parameter corresponding to the trained target neural network structure; The second parameter is sent to the server, so that the server updates the first parameter based on the second parameter and generates a next round of training instructions.

2. The method for searching a neural network structure according to claim 1, wherein: After the step of receiving the training instruction sent by the server, the method further includes: Parsing the training instruction to obtain data information; If the data information includes a first parameter corresponding to the neural network structure, executing the step of acquiring the first parameter corresponding to the neural network structure according to the training instruction; If the data information includes a preset neural network structure, the preset neural network structure is trained using training data.

3. A method for searching a neural network structure, characterized in that: The neural network structure search method is applied to a server, the server and the terminal are in the same distributed architecture, and the neural network structure search method includes: Obtaining a first parameter corresponding to a preset neural network structure; generating a training instruction according to the first parameter; The training instruction is sent to each terminal, wherein the terminal obtains the first parameter according to the training instruction, searches for the neural network structure according to the first parameter to obtain the target neural network structure corresponding to the first parameter, and trains the target neural network structure using training data, so that the terminal undertakes the neural network search work of the server, thereby improving the terminal resource utilization rate; Receiving the trained target neural network structure sent by the terminal, and a second parameter corresponding to the trained target neural network structure; According to the trained target neural network structure, the preset neural network is updated; And, updating the first parameter according to the second parameter.

4. The method for searching a neural network structure according to claim 3, wherein: After the step of obtaining the first parameter corresponding to the preset neural network structure, the method further includes: Determining whether to store a first parameter corresponding to a preset neural network structure; If it is determined that the first parameter corresponding to the neural network structure is not stored, generating a training instruction according to the preset neural network structure, and sending the training instruction to the terminal, so that the terminal uses the training data to train the preset neural network structure; If it is determined that the first parameter corresponding to the neural network structure is not stored, the step of obtaining the first parameter corresponding to the neural network structure is performed.

5. A terminal, characterized in that: The terminal includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the steps of the method for searching a neural network structure as described in any one of claims 1 to 2 are implemented.

6. A server, characterized in that: The server includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, and when the processor executes the computer program, the steps of the method for searching a neural network structure as described in any one of claims 3 to 4 are implemented.

7. A readable storage medium, characterized in that: The readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for searching a neural network structure according to any one of claims 1 to 4.

Images