Data analysis management method and system in multi-server scenario

Abstract

The present application provides a data analysis management method and system in a multi-server scenario, the method comprising the following steps: a first server receives a neural network model to be trained and labeled sample data sent by an external device; the first server extracts the labeled sample data and performs a training operation to obtain a first coefficient vector for this training, updates the weight data of each layer of the neural network model according to the first coefficient vector to obtain an updated neural network model, performs a training operation on the labeled sample again to obtain a second coefficient vector for the next training, and calculates the difference between the second coefficient vector and the first coefficient vector to obtain a difference vector. The technical solution provided by the present application has the advantage of high computational efficiency.

Description

Technical Field

[0001] The present invention relates to the field of electronic equipment, and in particular to a data analysis management method and system in a multi-server scenario. Background Art

[0002] In terms of physical form, cloud data centers are built from servers on many physical hosts. Data analysis in multi-server scenarios has a wide range of applications in current cloud platforms. For example, AI-based data processing is one such approach. Existing AI processing methods are slow, especially for training-type calculations, which are slow due to the large amount of data and the large number of iterations, affecting efficiency. Summary of the invention

[0003] The embodiments of the present invention provide a data analysis management method and system method and system in a multi-server scenario, which can improve data calculation efficiency and enhance user experience.

[0004] In a first aspect, an embodiment of the present invention provides a data analysis management method and system method in a multi-server scenario, the method comprising the following steps:

[0005] The first server receives a neural network model to be trained and labeled sample data sent by an external device;

[0006] The first server extracts the marked sample data and performs a training operation to obtain a first coefficient vector for this training, updates the weight data of each layer of the neural network model according to the first coefficient vector to obtain an updated neural network model, performs a training operation on the marked sample again to obtain a second coefficient vector for the next training, and calculates the difference between the second coefficient vector and the first coefficient vector to obtain a difference vector;

[0007] The first server obtains n servers with computing resources in a multi-server scenario, extracts the number of historical training times x, and allocates x / n training times to each of the n servers. The initial neural network model of the i-th server among the n servers is the neural network model updated according to the i-th coefficient vector, and the i-th coefficient vector = the first coefficient vector + the difference vector*(i-1)*x / n;

[0008] The first server obtains n confidence rates corresponding to n training results reported by n servers, and selects the trained neural network model of the server corresponding to the highest confidence rate from the n confidence rates to determine it as the trained neural network model.

[0009] In a second aspect, a data analysis management system in a multi-server scenario is provided, the system comprising: multiple servers, wherein:

[0010] The first server is used to receive a neural network model to be trained and marked sample data sent by an external device; extract the marked sample data to perform a training operation to obtain a first coefficient vector of this training, update the weight data of each layer of the neural network model according to the first coefficient vector to obtain an updated neural network model, perform a training operation on the marked sample again to obtain a second coefficient vector for the next training, and calculate the difference between the second coefficient vector and the first coefficient vector to obtain a difference vector; the first server obtains n servers with computing resources in a multi-server scenario, extracts the number of historical training times x, and allocates x / n training times to each of the n servers. The initial neural network model of the i-th server among the n servers is the neural network model updated according to the i-th coefficient vector, and the i-th coefficient vector = the first coefficient vector + the difference vector * (i-1) * x / n; obtain n confidence rates corresponding to the n training results reported by the n servers, and select the trained neural network model of the server corresponding to the highest confidence rate from the n confidence rates to determine it as the trained neural network model.

[0011] According to a third aspect, a computer-readable storage medium is provided, which stores a program for electronic data exchange, wherein the program enables a terminal to execute the method provided in the first aspect.

[0012] Implementing the embodiments of the present invention has the following beneficial effects:

[0013] It can be seen that the technical solution of the present application, the technical solution provided by the present application, the first server receives the neural network model to be trained and the labeled sample data sent by the external device. The first server extracts the labeled sample data and performs a training operation to obtain the first coefficient vector of this training, updates the weight data of each layer of the neural network model according to the first coefficient vector to obtain the updated neural network model, performs a training operation on the labeled sample again to obtain the second coefficient vector of the next training, and calculates the difference between the second coefficient vector and the first coefficient vector to obtain the difference vector. The first server obtains n servers with computing resources in a multi-server scenario, extracts the number of historical training times x, and allocates x / n training times to each of the n servers. The initial neural network model of the i-th server among the n servers is the neural network model updated according to the i-th coefficient vector, and the i-th coefficient vector = the first coefficient vector + the difference vector * (i-1) * x / n. The first server obtains the n confidence rates corresponding to the n training results reported by the n servers, and selects the trained neural network model of the server corresponding to the highest confidence rate from the n confidence rates to determine it as the trained neural network model. In this way, when training samples of the neural network model, the training tasks can be divided into multiple segments, and then the corresponding labeled samples can be trained separately through multiple servers, thereby improving the training efficiency and the user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings in the following description are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0015] Figure 1 A schematic diagram of the structure of a computer device

[0016] Figure 2 It is a flowchart of a data analysis management method in a multi-server scenario;

[0017] Figure 3 It is a structural diagram of a data analysis and management system in a multi-server scenario. DETAILED DESCRIPTION

[0018] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] The terms "first", "second", "third" and "fourth" etc. in the specification and claims of the present invention and the drawings are used to distinguish different objects, rather than 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 may optionally include steps or units that are not listed, or may optionally include other steps or units that are inherent to these processes, methods, products or devices.

[0020] Reference to "embodiments" herein means that a particular feature, result, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present invention. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.

[0021] See also Figure 1 , Figure 1A server is provided, which can be a computer device of Windows, Hongmeng and other systems, and of course can also be a computer device of other systems, such as Oracle, etc. This application is not limited to the above specific systems, such as Figure 1 As shown, the above-mentioned computer device may specifically include: a processor, a memory, a display screen, and a communication circuit. The above-mentioned components may be connected via a bus or in other ways. This application does not limit the specific method of the above-mentioned connection.

[0022] There can be multiple servers mentioned above, and multiple servers can form a cloud platform, data center, data platform and other data analysis platforms or systems. Of course, multiple servers can also form an internal data processing system. This application does not limit the form of the specific system composed of multiple servers. It only requires that the multiple servers can communicate with each other and transmit (including sending and receiving) data to each other.

[0023] See also Figure 2 , Figure 2 A data analysis management method and system method in a multi-server scenario are provided. Figure 2 As shown, this method can be used in Figure 1 The multi-server completion shown in the figure may specifically include:

[0024] Step S201: The first server receives a neural network model to be trained and labeled sample data sent by an external device (a device other than the multi-server).

[0025] Illustratively, the labeled sample data is data with a marked recognition result. Taking face recognition as an example, the labeled sample data may be training sample data with Zhang San marked on it (i.e., input data with a known result) and training sample data with data not marked on it.

[0026] The training of neural network models actually involves multiple training sessions to calculate and distinguish labeled sample data. Taking face recognition as an example, there are 100 labeled sample data, of which 60 are sample data of target objects and 40 are sample data of non-target objects. The purpose of training is to distinguish the 100 samples. The calculation result can distinguish them and realize specific scenarios, such as face recognition.

[0027] Step S202: The first server extracts the marked sample data and performs a training operation to obtain the first coefficient vector of this training. The weight data of each layer of the neural network model is updated according to the first coefficient vector to obtain an updated neural network model. The marked sample is trained again to obtain the second coefficient vector of the next training. The difference between the second coefficient vector and the first coefficient vector is calculated to obtain a difference vector.

[0028] Illustratively, the above training operation may include: a forward operation, a reverse operation and an update operation.

[0029] When training a neural network model, first perform a forward operation to obtain the forward operation result, and then perform a reverse operation on the forward operation result. Since the neural network model has multiple layers, each reverse operation of a layer will obtain the adjustment coefficient of this layer, and then multiply the weight of this layer by the adjustment coefficient to update the layer. By traversing all levels, the neural network model can be trained once.

[0030] Step S203: The first server obtains n servers with computing resources in a multi-server scenario, extracts the number of historical training times x, and allocates x / n training times to each of the n servers. The initial neural network model of the i-th server among the n servers is the neural network model updated based on the i-th coefficient vector, and the i-th coefficient vector = the first coefficient vector + difference vector*(i-1)*x / n.

[0031] For example, the above x is a relatively large value, such as 100, 1000, etc., and its specific value can be determined according to the number of historical trainings.

[0032] Step S204: The first server obtains n confidence rates corresponding to n training results reported by n servers, and selects the trained neural network model corresponding to the server with the highest confidence rate from the n confidence rates to determine it as the trained neural network model.

[0033] The technical solution provided by the present application is that the first server receives the neural network model to be trained and the marked sample data sent by the external device. The first server extracts the marked sample data and performs a training operation to obtain the first coefficient vector of this training, updates the weight data of each layer of the neural network model according to the first coefficient vector to obtain the updated neural network model, performs a training operation on the marked sample again to obtain the second coefficient vector of the next training, and calculates the difference between the second coefficient vector and the first coefficient vector to obtain the difference vector. The first server obtains n servers with computing resources in a multi-server scenario, extracts the number of historical training times x, and allocates x / n training times to each of the n servers. The initial neural network model of the i-th server among the n servers is the neural network model updated according to the i-th coefficient vector, and the i-th coefficient vector = the first coefficient vector + the difference vector * (i-1) * x / n. The first server obtains n confidence rates corresponding to the n training results reported by the n servers, and selects the trained neural network model of the server corresponding to the highest confidence rate from the n confidence rates to determine it as the trained neural network model. In this way, when training samples of the neural network model, the training tasks can be divided into multiple segments, and then the corresponding labeled samples can be trained separately through multiple servers, thereby improving the training efficiency and the user experience.

[0034] For example, the first server extracts the labeled sample data and performs a training operation to obtain the first coefficient vector of this training, which may specifically include:

[0035] The first server inputs the marked sample data as input data into the neural network model to be trained to perform a forward operation to obtain a forward operation result, and reversely inputs the forward operation result into the neural network model to perform a reverse operation. Each layer of reverse operation is performed to obtain a coefficient value of one layer, and multiple layers of reverse operation are performed to obtain multiple layers of coefficient values. The coefficient values ​​of multiple layers are then arranged in a column in ascending order according to the number of layers to obtain a first coefficient vector.

[0036] Here we take 5 layers as an example. In actual applications, the number of layers of the neural network model may exceed 5 layers, for example, it may be 30 layers, 48 ​​layers or even more layers. The specific number of layers is not limited here.

[0037] After performing the reverse operation, there are 5 more layers of coefficients, for example, 1.1, 0.95, 0.98, 1.02, and 1.06, which correspond to layers 1 to 5 in order. Then, the above 5 coefficients can be arranged in ascending order from layers 1 to 5 to obtain the first coefficient vector.

[0038] For example, the method of obtaining the above n confidence rates may specifically include:

[0039] Each time the i-th server updates the neural network model, it performs a forward operation on the labeled sample data to obtain a forward operation result, performs an activation operation on the forward operation result to obtain a confidence rate, performs x / n training and activation operations to obtain x / n confidence rates, and selects the highest confidence rate from the x / n confidence rates to determine as the confidence rate reported by the i-th server.

[0040] See also Figure 3 , Figure 3 A schematic diagram of the structure of a data analysis management system in a multi-server scenario is provided, wherein the system comprises: multiple servers, wherein:

[0041] The first server is used to receive a neural network model to be trained and marked sample data sent by an external device; extract the marked sample data to perform a training operation to obtain a first coefficient vector of this training, update the weight data of each layer of the neural network model according to the first coefficient vector to obtain an updated neural network model, perform a training operation on the marked sample again to obtain a second coefficient vector for the next training, and calculate the difference between the second coefficient vector and the first coefficient vector to obtain a difference vector; the first server obtains n servers with computing resources in a multi-server scenario, extracts the number of historical training times x, and allocates x / n training times to each of the n servers. The initial neural network model of the i-th server among the n servers is the neural network model updated according to the i-th coefficient vector, and the i-th coefficient vector = the first coefficient vector + the difference vector * (i-1) * x / n; obtain n confidence rates corresponding to the n training results reported by the n servers, and select the trained neural network model of the server corresponding to the highest confidence rate from the n confidence rates to determine it as the trained neural network model.

[0042] For example,

[0043] The first server is specifically used to input the marked sample data as input data into the neural network model to be trained to perform a forward operation to obtain a forward operation result, and reversely input the forward operation result into the neural network model to perform a reverse operation, each layer of reverse operation is performed to obtain a coefficient value of one layer, and multiple layers of reverse operation are performed to obtain multiple layers of coefficient values, and then the multiple layers of coefficient values ​​are arranged in a column in ascending order according to the number of layers to obtain a first coefficient vector.

[0044] For example,

[0045] The i-th server is used to update the neural network model each time, perform a forward operation on the labeled sample data to obtain a forward operation result, perform an activation operation on the forward operation result to obtain a confidence rate, perform x / n training and activation operations to obtain x / n confidence rates, and select the highest confidence rate from the x / n confidence rates to determine as the confidence rate reported by the i-th server.

[0046] For example, the first server in the embodiment of the present application can also be used to execute the following Figure 2 The detailed solutions, optional solutions, etc. of the illustrated embodiment will not be described in detail here.

[0047] An embodiment of the present invention also provides a computer storage medium, wherein the computer storage medium stores a computer program for electronic data exchange, and the computer program enables a computer to execute part or all of the steps of the data analysis management method in any multi-server scenario recorded in any of the above method embodiments.

[0048] An embodiment of the present invention also provides a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to enable a computer to execute part or all of the steps of any data analysis management method in a multi-server scenario as recorded in the above method embodiments.

[0049] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should know that the present invention is not limited by the described order of actions, because according to the present invention, certain steps can be received in other orders or performed simultaneously. Secondly, those skilled in the art should also know that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0050] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0051] In the several embodiments provided in the present application, it should be understood that the disclosed device can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of the units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, and the indirect coupling or communication connection of the device or unit can be electrical or other forms.

[0052] A person skilled in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing related hardware through a program, and the program can be stored in a computer-readable memory, and the memory can include: a flash drive, a read-only memory (English: Read-Only Memory, abbreviated as: ROM), a random access memory (English: Random Access Memory, abbreviated as: RAM), a magnetic disk or an optical disk, etc.

[0053] The embodiments of the present invention are described in detail above. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A data analysis management method in a multi-server scenario, characterized in that: The method comprises the following steps: The first server receives a neural network model to be trained and labeled sample data sent by an external device; The first server extracts the marked sample data and performs a training operation to obtain a first coefficient vector for this training, updates the weight data of each layer of the neural network model according to the first coefficient vector to obtain an updated neural network model, performs a training operation on the marked sample again to obtain a second coefficient vector for the next training, and calculates the difference between the second coefficient vector and the first coefficient vector to obtain a difference vector; The first server obtains n servers with computing resources in a multi-server scenario, extracts the number of historical training times x, and allocates x / n training times to each of the n servers. The initial neural network model of the i-th server among the n servers is the neural network model updated according to the i-th coefficient vector, and the i-th coefficient vector = the first coefficient vector + difference vector * (i-1) * x / n; The first server obtains n confidence rates corresponding to n training results reported by n servers, and selects the trained neural network model of the server corresponding to the highest confidence rate from the n confidence rates to determine it as the trained neural network model; The first server extracts the labeled sample data and performs a training operation to obtain the first coefficient vector of this training, which specifically includes: The first server inputs the marked sample data as input data into the neural network model to be trained to perform a forward operation to obtain a forward operation result, and reversely inputs the forward operation result into the neural network model to perform a reverse operation, and performs a layer of reverse operation to obtain a layer of coefficient values, and performs a multi-layer reverse operation to obtain a multi-layer coefficient value, and then arranges the multi-layer coefficient values ​​into a column in ascending order according to the number of layers to obtain a first coefficient vector; The methods for obtaining n confidence rates specifically include: Each time the i-th server updates the neural network model, it performs a forward operation on the labeled sample data to obtain a forward operation result, performs an activation operation on the forward operation result to obtain a confidence rate, performs x / n training and activation operations to obtain x / n confidence rates, and selects the highest confidence rate from the x / n confidence rates to determine as the confidence rate reported by the i-th server.

2. The method according to claim 1, characterized in that The labeled sample data is data with labeled recognition results.

3. The method according to claim 1, characterized in that A training operation includes: forward operation, reverse operation and update operation.

4. A data analysis and management system in a multi-server scenario, characterized in that: The system comprises: multiple servers, wherein: The first server is used to receive the neural network model to be trained and the marked sample data sent by the external device; extract the marked sample data to perform a training operation to obtain the first coefficient vector of this training, update the weight data of each layer of the neural network model according to the first coefficient vector to obtain the updated neural network model, perform a training operation on the marked sample again to obtain the second coefficient vector of the next training, and calculate the difference between the second coefficient vector and the first coefficient vector to obtain the difference vector; the first server obtains n servers with computing resources in a multi-server scenario, extracts the number of historical training times x, and allocates x / n training times to each of the n servers. The initial neural network model of the i-th server among the n servers is the neural network model updated according to the i-th coefficient vector, and the i-th coefficient vector = the first coefficient vector + the difference vector * (i-1) * x / n; obtain n confidence rates corresponding to the n training results reported by the n servers, and select the trained neural network model of the server corresponding to the highest confidence rate from the n confidence rates to determine it as the trained neural network model; The first server is specifically used to input the marked sample data as input data into the neural network model to be trained to perform a forward operation to obtain a forward operation result, and to reversely input the forward operation result into the neural network model to perform a reverse operation, and to perform a layer of reverse operation to obtain a layer of coefficient values, and to perform a multi-layer reverse operation to obtain a multi-layer coefficient value, and then to arrange the multi-layer coefficient values ​​into a column in ascending order according to the number of layers to obtain a first coefficient vector; The i-th server is used to update the neural network model each time, perform a forward operation on the labeled sample data to obtain a forward operation result, perform an activation operation on the forward operation result to obtain a confidence rate, perform x / n training and activation operations to obtain x / n confidence rates, and select the highest confidence rate from the x / n confidence rates to determine as the confidence rate reported by the i-th server.

5. The system according to claim 4, characterized in that A training operation includes: forward operation, backward operation and update operation.

6. A computer-readable storage medium storing a program for electronic data exchange, wherein: The program enables the terminal to execute the method provided in any one of claims 1-3.

Images