Low-power multi-agent cooperative enhancement method

Abstract

The application discloses a low-power-consumption multi-agent cooperative enhancement method. The method comprises the following steps: obtaining an initial task and an initial model for processing the initial task; dividing the initial task to obtain a plurality of divided sub-initial tasks; dividing the initial model based on the plurality of sub-initial tasks to obtain a plurality of sub-initial models corresponding to the plurality of sub-initial tasks, wherein each sub-initial model comprises a plurality of computing layers. The application solves the technical problem that the parameter quantity is large, the battery consumption is large, and the battery use efficiency is low due to the back-and-forth movement of data between the memory and the computing core in the calculation process in the prior art, thereby shortening the computable life cycle of the low-power-consumption agent.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of computer data optimization, and in particular, to a low-power multi-agent collaborative enhancement method. BACKGROUND

[0002] Low-power agents are an important sensing unit that plays an important role in current sensing information content-rich, sensing duration-growth, and sensing environment-demanding diversified complex sensing environments. Low-power agents are usually powered by batteries, including low-power processing chips and peripheral circuits designed specifically for energy consumption optimization. They can carry some intelligent AI applications for intelligent computing to provide intelligent services and solutions for users with small energy consumption in long-term continuous sensing application scenarios or in energy-constrained use scenarios.

[0003] Embedded low-power agents usually use low-power processors and components to reduce system running energy consumption, but also limit the computing power of the device, facing the problem of insufficient resources: (1) Limited processor performance: embedded devices usually use low-power, low-cost processors, which have relatively low performance and may not be able to handle complex computing tasks. (2) Limited memory: embedded devices usually have limited memory capacity, which may limit the number of applications they can run simultaneously or their ability to run large applications. (3) Environmental restrictions: embedded devices are often used in various environments, such as industrial control, automotive, medical devices, etc., which may have special requirements for device performance and power consumption, further limiting the availability of computing resources.

[0004] The basic idea to solve the current problem is to optimize computing and resource management, choose appropriate hardware and software components, and balance computing needs, power consumption, and cost in design. At the same time, the introduction of new technologies and solutions (such as deep learning accelerators, more efficient algorithms, etc.) may also help improve the computing performance of embedded devices. The core of the problem is that although the energy consumption required by the computation itself can be gradually reduced by new computing methods, the mode determined by the centralized processing information method of the von Neumann architecture - moving data from memory to processor and back, the energy consumption generated by moving sensing data in the system is the main reason.

[0005] Therefore, we propose an invention design, named a low-power multi-agent collaborative enhancement method, in which we use a group / multiple groups of low-power intelligent hardware to form a swarm agent cluster, reduce the consumption of the battery by individual agents during the calculation process through swarm agent task planning and memory reuse calculation optimization algorithm, and improve the battery usage efficiency, so that the low-power agent can have a longer computing life cycle. SUMMARY

[0006] The embodiment of the present application provides a low-power-consumption multi-agent cooperative enhancement method, so as to at least solve the technical problems of large parameter quantity, large battery consumption, low battery use efficiency and short low-power-consumption agent computable life cycle caused by data moving back and forth between memory and computing cores in the prior art.

[0007] According to an aspect of the embodiment of the present application, a low-power-consumption multi-agent cooperative enhancement method is provided. The method can include: obtaining an initial task and an initial model for processing the initial task;

[0008] dividing the initial task to obtain a plurality of divided sub-initial tasks; dividing the initial model based on the plurality of sub-initial tasks to obtain a plurality of sub-initial models corresponding to the plurality of sub-initial tasks, wherein each sub-initial model comprises a plurality of computing layers; determining a plurality of feature maps output by each computing layer in the process of computing the plurality of sub-initial tasks in each computing layer of the plurality of sub-initial models, wherein each feature map output by each computing layer comprises a plurality of sub-feature maps, and each feature map output by each computing layer comprises an abnormal situation part; after obtaining each feature map output by each computing layer, determining a set of each feature map corresponding to each computing layer based on each feature map output by each computing layer; calculating position information of the plurality of sub-feature maps in each feature map output by each computing layer in each feature map based on the size of each feature map output by each computing layer; determining the size of the abnormal situation part in each feature map based on the position information of the plurality of sub-feature maps in each feature map output by each computing layer; and determining the size of the normal situation part of all feature maps corresponding to each computing layer based on the set of each feature map corresponding to each computing layer and the size of the abnormal situation part in each feature map.

[0009] Optionally, the method further includes: a general expression of the process of the initial task model processing the initial task is:

[0010]

[0011] 0≤c<N,0≤z<M,0≤x,y<E,E=(H-R+U) / U

[0012] wherein, represents the output of the current computing layer when the initial task model processes the initial task, represents the input of the current computing layer when the initial task model processes the initial task, x, y, F and B represent the input, output, filter and bias parameter matrix of the data of the current computing layer when the initial task model processes the initial task, k, i, j represent the number of layers of the current computing layer when the initial task model processes the initial task; the value range of k, i, j is 0~C-1, 0~R-1, 0~R-1 respectively, c is the number of channels of the output data of the current computing layer when the initial task model processes the initial task; M and N are the filter and output data of the current computing layer when the initial task model processes the initial task; H, E and R represent the size of the input, output data and filter of the data of the current computing layer when the initial task model processes the initial task, U represents the given step of the current computing layer when the initial task model processes the initial task, and z is the value in the matrix of the bias parameter.

[0013] Optionally, based on the plurality of sub-initial tasks, the initial model is divided to obtain a plurality of sub-initial models corresponding to the plurality of sub-initial tasks, including: based on the channel number of the output feature map of the plurality of sub-initial tasks at any layer or the layer number of the plurality of sub-initial tasks at any layer, the initial model is divided to obtain a plurality of sub-initial models corresponding to the plurality of sub-initial tasks.

[0014] Optionally, based on the channel number of the output feature map of the plurality of sub-initial tasks at any layer or the layer number of the plurality of sub-initial tasks at any layer, the initial model is divided to obtain a plurality of sub-initial models corresponding to the plurality of sub-initial tasks.

[0015]

[0016] wherein, and represents the input data and the output data feature map; represents the jth th channel; represents the ith th channel; W i,j is a weight matrix; the symbol * represents a 2D convolution operation, B i represents the ith bias parameter, and the value range of j is 1~e.

[0017] Optionally, in the process of calculating the plurality of sub-initial tasks by each layer of the plurality of sub-initial models, the expression of each feature map of the plurality of feature maps output by each layer of the plurality of sub-initial models is determined.

[0018]

[0019] wherein, is the width and height of each feature map processed by each sub-initial model of each calculation layer for each initial task, n represents the number of the sub-initial model, and n ranges from 1 to l.

[0020] Optionally, after obtaining each feature map output by each calculation layer, an expression of a corresponding set of each feature map of each calculation layer is determined based on each feature map output by each calculation layer, and the expression comprises:

[0021]

[0022] wherein τ total represents the corresponding set of each feature map of each calculation layer.

[0023] Optionally, the size of the abnormal situation part in each feature map is determined based on the position information of the plurality of sub-feature maps in each feature map output by each calculation layer, and the size of the abnormal situation part in each feature map comprises:

[0024]

[0025] wherein min and max are functions of taking minimum value and maximum value, respectively, represents the length of the abnormal situation part of the previous calculation layer, represents the width of the abnormal situation part of the current calculation layer, v represents the length of the abnormal situation part of each feature map, and h represents the width of the abnormal situation part of each feature map, represents the total size of the abnormal part of each feature map.

[0026] Optionally, an expression of the size of the normal situation part of all feature maps corresponding to each calculation layer is determined based on the corresponding set of each feature map of each calculation layer and the size of the abnormal situation part in each feature map, and the expression comprises:

[0027]

[0028] wherein τ actual represents the size of the normal situation part of all feature maps corresponding to each calculation layer, represents the size of the abnormal situation part in each feature map sequentially removed from the corresponding set of each feature map of each calculation layer.

[0029] The present application has the following beneficial effects:

[0030] The application provides a low-power multi-agent cooperative enhancement method, which uses multiple intelligent devices to perform energy consumption optimization for the same task. The energy consumption optimization process can be divided into two stages: (1) complex task splitting, so that the task can be adapted to multiple agents for calculation; (2) timing optimization of the task loaded into the memory, low-power swarm intelligence agent joint enhancement, through multi-task decomposition, fragmentation execution-merging, the AI model with large size that cannot be executed originally can be calculated, and the conclusion is derived, solving the technical problems of large parameter quantity, large battery consumption and low battery use efficiency caused by the movement of data between memory and calculation core in the calculation process in the prior art, so that the low-power intelligent agent can have a longer computable life cycle. BRIEF DESCRIPTION OF DRAWINGS

[0031] The accompanying drawings, which are included to provide a further understanding of the application, form a part of the specification and are included to illustrate the illustrative embodiments of the application and to explain the principles on which the application is based. In the drawings:

[0032] Figure 1 FIG. 1 is a flowchart of a low-power multi-agent cooperative enhancement method according to an embodiment of the application. DETAILED DESCRIPTION

[0033] In order to enable those skilled in the art to better understand the application scheme, the technical solutions in the embodiments of the application will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the application. Obviously, the described embodiments are only a part of the embodiments of the application, not all the embodiments of the application. Based on the embodiments in the application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the application.

[0034] It should be noted that the terms "first", "second", and the like in the description and claims of the application and the above drawings are used to distinguish similar objects, and do not necessarily indicate a specific order or sequence. It should be understood that the terms used in this way can be interchanged, as appropriate, so that the embodiments of the application described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusion, for example, a process, method, system, product or device that includes a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but can include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0035] Embodiment 1

[0036] According to an embodiment of the application, a low-power multi-agent cooperative enhancement method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system comprising at least one set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described herein can be executed in an order different from that shown herein.

[0037] Figure 1 is a flowchart of a low-power multi-agent cooperative enhancement method according to an embodiment of the application, as Figure 1 shown, the method can include the following steps:

[0038] Step S101, obtaining an initial task and an initial model for processing the initial task.

[0039] In the technical solution provided in the above step S101 of the application, for example, the initial task can be an image classification task or an image segmentation task, etc., and the initial model for processing the initial task can be a neural network model for processing the initial task.

[0040] Step S102, dividing the initial task to obtain a plurality of divided sub-initial tasks.

[0041] In the technical solution provided in the above step S102 of the application, when the initial task is to classify an image, for example, an image is an animal, the head and body of the animal are separated, the head is a sub-initial task, and the body is a sub-initial task.

[0042] Step S103, based on the plurality of sub-initial tasks, dividing the initial model to obtain a plurality of sub-initial models corresponding to the plurality of sub-initial tasks, wherein each sub-initial model comprises a plurality of computing layers.

[0043] In the technical solution provided in the step S103 of the present application, for example, the head of the animal corresponds to one sub-initial model, and the body of the animal corresponds to one sub-initial model, and each sub-initial model comprises a plurality of calculation layers.

[0044] In step S104, in the process of calculating a plurality of sub-initial tasks by each calculation layer in a plurality of sub-initial models, a plurality of feature maps output by each calculation layer are determined, wherein each feature map output by each calculation layer comprises a plurality of sub-feature maps, and each feature map output by each calculation layer comprises an abnormal situation part.

[0045] In the technical solution provided in the step S104 of the present application, in the process of calculating a plurality of sub-initial tasks by each calculation layer in a plurality of sub-initial models, a plurality of feature maps output by each calculation layer are determined, wherein each feature map output by each calculation layer comprises a plurality of sub-feature maps, and each feature map output by each calculation layer comprises an abnormal situation part.

[0046] In step S105, after obtaining each feature map output by each calculation layer, a set of each feature map corresponding to each calculation layer is determined based on each feature map output by each calculation layer.

[0047] In the technical solution provided in the step S105 of the present application, after obtaining each feature map output by each calculation layer, the set of each feature map corresponding to each calculation layer is obtained by putting each feature map output by each calculation layer together.

[0048] In step S106, based on the size of a plurality of sub-feature maps in each feature map output by each calculation layer in each calculation layer, the position information of the plurality of sub-feature maps in each feature map output by each calculation layer is calculated.

[0049] In the technical solution provided in the step S106 of the present application, based on the size of a plurality of sub-feature maps in each feature map output by each calculation layer in each calculation layer, the position information of the plurality of sub-feature maps in each feature map output by each calculation layer is calculated, and the expression is as follows:

[0050]

[0051] wherein, represents the position information of any one sub-feature map in each feature map output by each calculation layer, represents the horizontal coordinate of the position of any one sub-feature map in each feature map output by each calculation layer, represents the vertical coordinate of the position of any one sub-feature map in each feature map output by each calculation layer.

[0052] In step S107, the size of the abnormal situation part in each feature map is determined based on the position information of each feature map output by each layer of the calculation layer.

[0053] In the technical solution provided in step S107, the size of the abnormal situation part in each feature map is determined based on the position information of each feature map output by each layer of the calculation layer.

[0054] In step S108, the size of the normal situation part of all the feature maps corresponding to each layer of the calculation layer is determined based on the set of each feature map corresponding to each layer and the size of the abnormal situation part in each feature map.

[0055] In the technical solution provided in step S108, the size of the normal situation part of all the feature maps corresponding to each layer of the calculation layer is determined based on the set of each feature map corresponding to each layer and the size of the abnormal situation part in each feature map.

[0056] The above method of the embodiment is further described below.

[0057] As an optional embodiment, the method further includes that the general expression of the process of processing the initial task by the initial task model is:

[0058]

[0059] 0≤c<N, 0≤z<M, 0≤x,y<E, E=(H-R+U) / U

[0060] wherein, represents the output of the current calculation layer when the initial task model processes the initial task, represents the input of the current calculation layer when the initial task model processes the initial task, x, y, F and B respectively represent the input, output, filter and bias parameter matrix of the data of the current calculation layer when the initial task model processes the initial task, k, i, j represent the number of layers of the current calculation layer when the initial task model processes the initial task; the value ranges of k, i, j are respectively 0~C-1, 0~R-1, 0~R-1, the parameter c is the number of channels of the output data of the current calculation layer when the initial task model processes the initial task; M and N are the filter and output data of the current calculation layer when the initial task model processes the initial task; H, E and R respectively represent the size of the input, output data and filter of the data of the current calculation layer when the initial task model processes the initial task, U is a given step of the current calculation layer when the initial task model processes the initial task, and z is a value in the bias parameter matrix.

[0061] In this embodiment, the initial task is processed by the above formula to obtain the output of the initial task.

[0062] As an optional embodiment, in step S103, the initial model is divided based on the plurality of sub-initial tasks to obtain a plurality of sub-initial models corresponding to the plurality of sub-initial tasks, including: dividing the initial model based on the number of channels of the output feature map of the plurality of sub-initial tasks at any layer or the number of layers of the plurality of sub-initial tasks at any layer to obtain the plurality of sub-initial models corresponding to the plurality of sub-initial tasks.

[0063] In this embodiment, the initial model can be divided based on the number of channels of the output feature map of the plurality of sub-initial tasks at any layer or the number of layers of the plurality of sub-initial tasks at any layer to obtain the plurality of sub-initial models corresponding to the plurality of sub-initial tasks.

[0064] As an optional embodiment, the initial model is divided based on the number of channels of the output feature map of the plurality of sub-initial tasks at any layer or the number of layers of the plurality of sub-initial tasks at any layer to obtain the expression of the plurality of sub-initial models corresponding to the plurality of sub-initial tasks.

[0065]

[0066] wherein, and denote the feature maps of the input data and the output data; denotes the jth channel thereof; denotes the ith channel; W i,j is a weight matrix; the symbol * represents a 2D convolution operation, B i denotes the ith bias parameter, and j takes a value in the range of 1 to e.

[0067] In this embodiment, the initial model is divided by the above expression to obtain the plurality of sub-initial models corresponding to the plurality of sub-initial tasks, for example, a neural network model is divided into 5 sub-initial models, which are only used for illustration and are not limited.

[0068] As an optional embodiment, in step S104, in the process of calculating the plurality of sub-initial tasks by each layer of the plurality of sub-initial models, the expression of each feature map of the plurality of feature maps output by each layer of the calculation layer is determined.

[0069]

[0070] wherein, is the width and height of each feature map processed by each sub-initial model of each computing layer for each initial task, n represents the nth sub-initial model, and n ranges from 1 to l.

[0071] In this embodiment, the width and height of each feature map processed by each sub-initial model of each computing layer for each initial task are calculated by the above expression. It should be noted that k is a label of each feature map of the multiple feature maps output by each computing layer, k corresponds to l, for example, the first computing layer corresponds to three initial sub-models, each sub-model corresponds to a feature map output by each sub-model, which is only used for illustration and is not limited in particular.

[0072] As an optional embodiment, after obtaining each feature map output by each computing layer, step S105, based on each feature map output by each computing layer, the expression of the corresponding set of each feature map of each layer is determined as follows:

[0073]

[0074] wherein τ total represents the corresponding set of each feature map of each layer.

[0075] In this embodiment, the corresponding set of each feature map of each layer is calculated by the above formula.

[0076] As an optional embodiment, step S107, based on the position information of the multiple sub-feature maps in each feature map in each feature map output by each computing layer, the size of the abnormal situation part in each feature map is determined, including:

[0077]

[0078] wherein min and max are functions of taking minimum and maximum respectively, represents the length of the abnormal situation part of the previous computing layer, represents the width of the abnormal situation part of the current computing layer, v represents the length of the abnormal situation part of each feature map, and h represents the width of the abnormal situation part of each feature map, represents the total size of the abnormal part of each feature map.

[0079] In this embodiment, the size of the abnormal situation part in each feature map is obtained by calculating the position information of the multiple sub-feature maps in each feature map in each feature map output by each computing layer. It should be noted that if the first sub-feature map is stored in the second sub-feature map, that is, its coordinates satisfy such a range. As shown in the following formula:

[0080]

[0081] And in the process of convolution calculation, the value of the convolution kernel horizontal coordinate and vertical coordinate needs to meet Then it is proved that the coordinate of the current sub-feature map belongs to the abnormal case, which should be removed from the whole feature map, It is expressed as the convolution kernel needs to meet the x-coordinate of the convolution kernel Must be contained in the x-coordinate of the feature map Range, but the y-coordinate of the convolution kernel Does not belong to the Then the intersection, that is, the convolution across two feature maps, is the abnormal case part.

[0082] As an optional embodiment, step S108, based on the corresponding set of each feature map of each layer and the size of the abnormal case part in each feature map, the expression of the size of the normal case part of all feature maps corresponding to each calculation layer of each layer is determined as:

[0083]

[0084] Where, τ actua L represents the size of the normal case part of all feature maps corresponding to each calculation layer of each layer, It is expressed as the size of the abnormal case part in each feature map is sequentially removed from the corresponding set of each feature map of each layer.

[0085] In this embodiment, the set of each feature map corresponding to each layer removes the size of the abnormal case part in each feature map once, to obtain the size of the normal case part of all feature maps corresponding to each calculation layer of each layer.

[0086] In the embodiment of the present application, the initial task and the initial model for processing the initial task are obtained; the initial task is divided to obtain a plurality of divided sub-initial tasks; based on the plurality of sub-initial tasks, the initial model is divided to obtain a plurality of sub-initial models corresponding to the plurality of sub-initial tasks, wherein each sub-initial model comprises a plurality of computing layers; in the process of calculating the plurality of sub-initial tasks by each computing layer in the plurality of sub-initial models, a plurality of feature maps output by each computing layer are determined, wherein each feature map output by each computing layer comprises a plurality of sub-feature maps, and each feature map output by each computing layer comprises an abnormal situation part; after each feature map output by each computing layer is obtained, based on each feature map output by each computing layer, a set of each feature map corresponding to each computing layer is determined; based on the size of the plurality of sub-feature maps in each feature map output by each computing layer in each feature map output by each computing layer, the position information of the plurality of sub-feature maps in each feature map output by each computing layer in each feature map is calculated; based on the position information of the plurality of sub-feature maps in each feature map output by each computing layer in each feature map, the size of the abnormal situation part in each feature map is determined; based on the set of each feature map corresponding to each computing layer and the size of the abnormal situation part in each feature map, the size of the normal situation part of all feature maps corresponding to each computing layer is determined, thereby solving the technical problem that the large amount of parameters generated by the data moving back and forth between the memory and the computing core in the calculation process in the prior art, the large consumption of the battery, and the low battery use efficiency, so that the low-power intelligent agent has a short computable life cycle, and the technical effect that the data in the calculation process is formed into a group intelligent agent cluster by using one / multiple low-power intelligent hardware, the group intelligent agent task is planned and coordinated, and the memory reuse calculation optimization algorithm reduces the consumption of the battery by a single intelligent agent in the calculation process, improves the battery use efficiency, and thus makes the low-power intelligent agent have a longer computable life cycle.

[0087] The above-mentioned serial numbers of the embodiments of the present application are only for description, and do not represent the advantages and disadvantages of the embodiments.

[0088] In the above-mentioned embodiments of the present application, the description of each embodiment has its own emphasis, and the parts not described in detail in a certain embodiment can be referred to the related description of other embodiments.

[0089] In the several embodiments provided by the present application, it should be understood that the disclosed technology can be implemented in other ways. Of course, the unit division in the above-mentioned device embodiment is only a logic function division, and there can be another division manner during actual implementation; for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the displayed or discussed mutual couplings or direct couplings or communication connections between the units, can be indirect couplings or communication connections through some interfaces, and can be electrical or other forms.

[0090] The units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, i.e. may be located in one place, or may be distributed to multiple units. Part or all of the units may be selected according to actual needs to achieve the purpose of the embodiment.

[0091] In addition, each functional unit in each embodiment of the present application can be integrated in a first processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The integrated unit can be realized in the form of hardware or in the form of a software functional unit.

[0092] The above is only the preferred embodiment of the present application, and it should be noted that for those skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application.

Claims

1. A low-power multi-agent cooperative enhancement method, characterized in that, include: Obtain the initial task and the initial model for processing the initial task; The initial task is divided into multiple sub-initial tasks. Based on multiple sub-initial tasks, the initial model is divided to obtain multiple sub-initial models corresponding to the multiple sub-initial tasks. Each sub-initial model contains multiple computational layers. During the computation of multiple sub-initial tasks in each computation layer of multiple sub-initial models, multiple feature maps output by each computation layer are determined. Each feature map output by each computation layer includes multiple sub-feature maps and includes anomaly cases. After obtaining each feature map output by each computational layer, the set of each feature map corresponding to each layer is determined based on each feature map output by each computational layer. Based on the size of each feature map output by each computational layer and the size of each feature map output by each computational layer, the position information of each feature map output by each computational layer is calculated. Based on the position information of each feature map in the output of each computation layer, the size of the abnormal part in each feature map is determined. Based on the set of each feature map corresponding to each layer and the size of the abnormal part in each feature map, the size of the normal part of all feature maps corresponding to each computation layer is determined.

2. The method according to claim 1, characterized in that, The method further includes: the general expression for the process of the initial task model processing the initial task is: 0≤c <N,0≤z<M,0≤x,y<E,E=(H-R+U) / U in, This represents the output of the current computational layer when the initial task model processes the initial task. This represents the input of the current computational layer when the initial task model processes the initial task. x, y, F, and B represent the matrices of the input, output, filter, and bias parameters of the current computational layer data when the initial task model processes the initial task, respectively. k, i, and j represent the number of layers in the current computational layer when the initial task model processes the initial task; the values ​​of k, i, and j range from 0 to C⁻¹, 0 to R⁻¹, and 0 to R⁻¹, respectively. Parameter c is the number of channels for the output data of the current computational layer when the initial task model processes the initial task. M and N are the filters and output data of the current computational layer when the initial task model processes the initial task. H, E, and R represent the sizes of the input, output data, and filters of the current computational layer data when the initial task model processes the initial task, respectively. U represents the given step size of the current computational layer when the initial task model processes the initial task. z is the value in the bias parameter matrix.

3. The method according to claim 1, characterized in that, Based on multiple sub-initial tasks, the initial model is divided to obtain multiple sub-initial models corresponding to the multiple sub-initial tasks, including: Based on the number of channels in the output feature maps of multiple sub-initial tasks at any layer or the number of layers of multiple sub-initial tasks at any layer, the initial model is divided to obtain multiple sub-initial models corresponding to multiple sub-initial tasks.

4. The method according to claim 3, characterized in that, Based on the number of channels in the output feature maps of multiple sub-initial tasks at any layer, or the number of layers in any layer for the multiple sub-initial tasks, the initial model is divided, resulting in the following expressions for multiple sub-initial models corresponding to the multiple sub-initial tasks: in, and Feature maps representing input and output data; Indicates its j-th th One channel; Indicates the i-th th One channel; W i,j It is the weight matrix; the symbol * represents a 2D convolution operation, B i Let j represent the i-th bias parameter, where j ranges from 1 to e.

5. The method according to claim 1, characterized in that, During the computation of multiple sub-initial tasks in each computational layer of multiple sub-initial models, the expression for each feature map of the multiple feature maps output by each computational layer is determined: in, It represents the width and height of each feature map after each sub-initial model of each computation layer processes each initial task, where n represents the nth sub-initial model, and the value of n ranges from 1 to 1.

6. The method according to claim 1, characterized in that, After obtaining each feature map output from each computational layer, the expression for determining the set of corresponding feature maps for each layer, based on each feature map output from each computational layer, is as follows: including: Where, τ total This represents the set of each feature map corresponding to each layer.

7. The method according to claim 1, characterized in that, Based on the position information of each feature map in the output of each computation layer from multiple sub-feature maps in each feature map, the size of the outlier portion in each feature map is determined, including: Where min and max are functions that take the minimum and maximum values, respectively. This indicates the length of the exception handling portion of the previous computational layer. represents the width of the outlier portion of the current computation layer, v represents the length of the outlier portion of each feature map, and h represents the width of the outlier portion of each feature map. This represents the total size of the abnormal portion of each feature map.

8. The method according to claim 1, characterized in that, Based on the set of each feature map corresponding to each layer and the size of the outlier portion of each feature map, the expression for determining the size of the normal portion of all feature maps corresponding to each computational layer is as follows: Where, τ actual This represents the size of the normal portion of the entire feature map corresponding to each computational layer. This indicates the size of the outlier portion removed sequentially from the set of feature maps corresponding to each layer.

9. A processor, characterized in that, The processor is used to run a program, wherein the program, when running, performs the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the computer-readable storage medium is located to perform the method according to any one of claims 1 to 8.

Images