CN110738318A - Method, system and device for evaluating network structure running time and generating evaluation model - Google Patents
Method, system and device for evaluating network structure running time and generating evaluation model Download PDFInfo
- Publication number
- CN110738318A CN110738318A CN201910859244.8A CN201910859244A CN110738318A CN 110738318 A CN110738318 A CN 110738318A CN 201910859244 A CN201910859244 A CN 201910859244A CN 110738318 A CN110738318 A CN 110738318A
- Authority
- CN
- China
- Prior art keywords
- neuron
- neurons
- network structure
- evaluation
- operation time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Life Sciences & Earth Sciences (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
- Debugging And Monitoring (AREA)
Abstract
The application discloses a method, a system and a device for evaluating the running time of a network structure and generating an evaluation model, wherein the method for evaluating the running time of the network structure can comprise the following steps: disassembling a network structure to be processed to obtain each neuron forming the network structure; for each neuron, determining the running time of the neuron based on the parameters of the neuron and an evaluation model obtained by pre-training; and determining the running time of the network structure according to the running time of each neuron. By applying the scheme, the accuracy of the evaluation result can be improved.
Description
[ technical field ] A method for producing a semiconductor device
The application relates to the field of deep learning, in particular to a method, a system and a device for evaluating network structure running time and generating an evaluation model.
[ background of the invention ]
Deep learning techniques have been largely successful in many directions. In the deep learning technology, the quality of a network structure (i.e., a Neural network structure) has a very important influence on the effect of a model, and manually designing the network structure requires very rich experience and numerous attempts, and numerous parameters generate explosive combinations and are difficult to implement, so that a Neural network Architecture Search technology (NAS) becomes a research hotspot in recent years.
The NAS is to use an algorithm to replace tedious manual operation to automatically search out an optimal network structure, and the implementation of the NAS mainly comprises key elements such as search space definition, search strategies, search target evaluation and the like.
Currently, the running time of the network structure is generally evaluated in a manner that the computational complexity of the searched network structure is determined, and the running time of the network structure is evaluated according to the computational complexity.
[ summary of the invention ]
In view of the above, the present application provides a method, system and apparatus for network structure runtime evaluation and evaluation model generation.
The specific technical scheme is as follows:
A method for runtime assessment of network structures, comprising:
disassembling a network structure to be processed to obtain each neuron forming the network structure;
for each neuron, determining the operation time of the neuron based on the parameters of the neuron and an evaluation model obtained by pre-training;
and determining the running time of the network structure according to the running time of each neuron.
According to the preferred embodiment, the number of assessment models is or greater than ;
when the number of the evaluation models is greater than , the determining the running time of the neuron based on the parameters of the neuron and the pre-trained evaluation models comprises:
and respectively inputting the parameters of the neurons into each evaluation model to obtain the operation time of the neurons evaluated by each evaluation model, and fusing the obtained operation time to determine the operation time of the neurons.
According to a preferred embodiment of , the determining the neuron's runtime from the fused runtime comprises calculating a mean value of the runtime from the fused runtime, and using the mean value as the neuron's runtime.
According to a preferred embodiment of , the determining the runtime of the network structure based on the runtimes of the neurons comprises summing the runtimes of the neurons, the sum being the runtime of the network structure.
an assessment model generation method, comprising:
selecting a part of neurons from neurons in a preset range;
respectively constructing training data corresponding to each selected neuron, wherein the training data comprise parameters of the neuron and the running time of the neuron;
and training according to the training data to obtain an evaluation model so as to respectively determine the running time of each neuron forming the network structure by using the evaluation model and the parameters of the neuron when the running time of the network structure to be processed needs to be evaluated, and determining the running time of the network structure according to the running time of each neuron.
According to a preferred embodiment of , the selecting the portion of neurons from the predetermined range includes randomly selecting the portion of neurons from all possible neurons in the search space.
According to a preferred embodiment of , the method further comprises sending each selected neuron to a designated device for operation, and obtaining the operation time of the neuron.
According to the preferred embodiment of the present application , the number of evaluation models is ;
or, the number of the evaluation models is greater than , so that when the operation time evaluation needs to be performed on the network structure to be processed, for any neuron, the parameters of the neuron are respectively input into each evaluation model to obtain the operation time of the neuron evaluated by each evaluation model, and the operation time of the neuron is determined by fusing the obtained operation times.
network structure operation time evaluation device comprises a disassembling unit and an evaluation unit;
the disassembling unit is used for disassembling the network structure to be processed to obtain each neuron forming the network structure;
and the evaluation unit is used for determining the operation time of the neurons and determining the operation time of the network structure according to the operation time of each neuron based on the parameters of the neurons and an evaluation model obtained by pre-training aiming at each neuron.
According to the preferred embodiment, the number of assessment models is or greater than ;
when the number of the evaluation models is larger than , the evaluation unit inputs the parameters of the neurons into each evaluation model respectively aiming at each neuron to obtain the operation time of the neurons evaluated by each evaluation model, and the operation time of the neurons is determined by fusing the obtained operation times.
According to the preferred embodiment, the evaluation unit calculates the mean value of the neuron's running time evaluated by each evaluation model for each neuron, and takes the mean value as the neuron's running time.
According to a preferred embodiment of the present application , the evaluation unit adds the running times of the neurons, and takes the sum as the running time of the network structure.
evaluation model generation devices comprise an acquisition unit, a construction unit and a training unit;
the acquisition unit is used for selecting part of neurons from the neurons in a preset range;
the constructing unit is configured to respectively construct training data corresponding to the selected neurons, where the training data includes parameters of the neurons and a running time of the neurons;
the training unit is used for training according to the training data to obtain an evaluation model, so that when the operation time of the network structure to be processed needs to be evaluated, the operation time of each neuron forming the network structure is determined by using the evaluation model and parameters of the neuron, and the operation time of the network structure is determined according to the operation time of each neuron.
According to the preferred embodiment of , the obtaining unit randomly selects a part of neurons from all possible neurons in the search space.
According to a preferred embodiment of , the constructing unit further is configured to, for each selected neuron, send the neuron to a designated device for operation, respectively, so as to obtain a running time of the neuron.
According to the preferred embodiment of the present application , the number of evaluation models is ;
or, the number of the evaluation models is greater than , so that when the operation time evaluation needs to be performed on the network structure to be processed, for any neuron, the parameters of the neuron are respectively input into each evaluation model to obtain the operation time of the neuron evaluated by each evaluation model, and the operation time of the neuron is determined by fusing the obtained operation times.
A network architecture runtime evaluation system, comprising:
a network structure runtime evaluation apparatus as described above, and an evaluation model generation apparatus as described above.
computer apparatus comprising a memory, a processor and a computer program stored on the memory and executable on the processor, the processor when executing the program implementing the method as described above.
computer readable storage medium having stored thereon a computer program which, when executed by a processor, carries out the method as described above.
Based on the introduction, the method and the device for evaluating the network structure have the advantages that the evaluation model can be used for determining the running time of each neuron forming the network structure respectively, and then the running time of the network structure can be determined according to the running time of each neuron, so that the running time of the network structure can be evaluated more accurately, and the accuracy of an evaluation result is improved.
[ description of the drawings ]
Fig. 1 is a flowchart of an embodiment of a method for evaluating a network structure runtime according to the present application.
Fig. 2 is a flowchart of an embodiment of an evaluation model generation method according to the present application.
Fig. 3 is a flowchart of an embodiment of an evaluation model generation and network structure runtime evaluation method according to the present application.
Fig. 4 is a schematic structural diagram of a network runtime evaluation apparatus according to an embodiment of the present application.
Fig. 5 is a schematic structural diagram of an evaluation model generation apparatus according to an embodiment of the present application.
FIG. 6 illustrates a block diagram of an exemplary computer system/server 12 suitable for use in implementing embodiments of the present application.
[ detailed description ] embodiments
In order to make the technical solution of the present application more clear and understandable, the solution of the present application is further illustrated by with reference to the drawings and examples.
All other embodiments, which can be derived by one skilled in the art from the embodiments given herein without making any inventive step, are intended to be encompassed by the present disclosure.
In addition, it should be understood that the term "and/or" herein only describes kinds of association relations describing the association objects, which means that there may be three kinds of relations, for example, A and/or B, which may mean three kinds of relations of A alone, A and B together, and B alone.
Fig. 1 is a flowchart of an embodiment of a method for evaluating a network structure runtime according to the present application. As shown in fig. 1, the following detailed implementation is included.
In 101, the network structure to be processed is decomposed to obtain neurons constituting the network structure.
At 102, for each neuron, the operation time of the neuron is determined based on the parameters of the neuron and the pre-trained evaluation model.
In 103, the operating time of the network structure is determined from the operating times of the neurons.
Each network structure is composed of a plurality of neurons (OP), so that if the running time of each neuron can be acquired respectively, the running time of the whole network structure can be determined according to the running time of each neuron.
Therefore, for the network structure to be processed, the network structure may be disassembled first, so as to obtain each neuron constituting the network structure, and how to disassemble the network structure is the prior art.
And aiming at each neuron obtained by disassembling, the running time of the neuron can be respectively obtained. In this embodiment, for each neuron obtained by disassembly, the running time of the neuron may be determined based on the parameter of the neuron and an evaluation model obtained by pre-training, respectively.
The parameters of each neuron are known, and the evaluation model can be obtained by pre-training. The inputs to the evaluation model may be parameters of the neuron and the outputs may be the runtime of the evaluated neuron.
The number of the evaluation models can be , or more than , that is, a plurality of evaluation models, so as to improve the accuracy of the evaluation result, the number of the evaluation models can be a plurality of evaluation models, so that for each neuron, the parameters of the neuron can be respectively input into each evaluation model, thereby obtaining the running time of the neuron evaluated by each evaluation model, and further, the obtained running times can be fused to finally determine the running time of the neuron.
The specific model for the evaluation model is not limited, and may include a random forest model, a decision tree model, an eXtreme Gradient Boosting (XGBoost) model, and the like.
After the running time of each neuron is obtained, the running time of the network structure can be determined according to the running time of each neuron. Preferably, the run times of the neurons may be summed, and the sum taken as the run time of the network structure.
For example, assuming that the network structure includes 20 neurons in total, for convenience of description, referred to as neurons 1 to 20, the operating times of neurons 1 to 20 can be obtained, respectively, and then the operating times of neurons 1 to 20 can be added, and the sum is used as the operating time of the network structure.
Based on the above description, in the solution of the present application, when obtaining the runtime of each neuron, an evaluation model obtained by pre-training is needed, and the following is a description of the generation manner of the evaluation model.
Fig. 2 is a flowchart of an embodiment of an evaluation model generation method according to the present application. As shown in fig. 2, the following detailed implementation is included.
In 201, a portion of neurons is selected from neurons within a predetermined range.
In 202, for each selected neuron, training data corresponding to the neuron is respectively constructed, and the training data includes parameters of the neuron and a running time of the neuron.
In 203, an evaluation model is obtained by training according to the training data, so that when the operation time of the network structure to be processed needs to be evaluated, the operation time of each neuron forming the network structure is determined by using the evaluation model and the parameters of the neuron, and the operation time of the network structure is determined according to the operation time of each neuron.
In this embodiment, the neurons in the predetermined range may refer to all possible neurons in the search space. All possible neurons can be enumerated first according to the search space, i.e., which neurons, parameters of different neurons, etc. Some of the neurons in the search space may be randomly selected for all possible neurons in the search space, for example, about 20% of the neurons may be selected for constructing the training data.
For each selected neuron, training data corresponding to the neuron can be constructed, and the training data may include parameters of the neuron and a running time of the neuron. And for each selected neuron, respectively sending the neuron to a specified device for running, thereby obtaining the running time of the neuron. For example, the neuron may be sent to a mobile phone or other embedded device for running, so as to obtain the actual running time of the neuron, which is implemented in the prior art.
And (4) training to obtain an evaluation model according to the constructed training data. The specific model for the evaluation model is not limited, and may include a random forest model, a decision tree model, an XGBoost model, and the like.
Therefore, when the running time of the network structure to be processed needs to be evaluated, the evaluation model and the parameters of the neurons can be used for respectively determining the running time of each neuron forming the network structure, and further the running time of the network structure can be determined according to the running time of each neuron.
Thus, when the running time of the network structure to be processed needs to be evaluated, for any neuron, the parameters of the neuron can be respectively input into each evaluation model, so that the running time of the neuron evaluated by each evaluation model is obtained, and the obtained running times can be fused to finally determine the running time of the neuron.
With the above introduction in mind, fig. 3 is a flowchart of an embodiment of a method for generating an evaluation model and evaluating a network structure runtime according to the present application. As shown in fig. 3, the following detailed implementation is included.
In 301, a portion of neurons is randomly selected from all possible neurons in the search space.
Such as randomly selecting about 20% of the neurons.
At 302, for each selected neuron, training data corresponding to the neuron is constructed, and the training data includes parameters of the neuron and a running time of the neuron.
For each selected neuron, the neuron can be sent to a specified device to run, so that the running time of the neuron is obtained.
At 303, three evaluation models are trained according to the constructed training data.
If 1000 pieces of training data are constructed, the 1000 pieces of training data can be utilized to respectively train and obtain a random forest model, a decision tree model and an XGboost model.
At 304, the network structure to be processed is disassembled to obtain the neurons that constitute the network structure.
In 305, for each neuron obtained by the disassembly, the parameters of the neuron are respectively input into three evaluation models to obtain the running time of the neuron evaluated by the three evaluation models, and the three running times obtained by fusion are used for determining the running time of the neuron.
Preferably, the mean value of the running time of the neuron estimated by the random forest model, the decision tree model and the XGBoost model can be calculated, and the mean value is used for determining the running time of the neuron.
At 306, the run times of the neurons are summed, and the sum is taken as the run time of the network structure.
It should be noted that for the sake of simplicity, the foregoing method embodiments are described as series combinations of acts, but those skilled in the art should understand that the present application is not limited by the described order of acts, as some steps may occur in other orders or concurrently depending on the application.
In the above embodiments, the descriptions of the respective embodiments have respective emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to related descriptions of other embodiments.
In short, by adopting the scheme of the embodiment of the method, the operation time of each neuron forming the network structure can be respectively determined by utilizing the evaluation model, and then the operation time of the network structure can be determined by adding the operation time of each neuron, so that the operation time of the network structure can be accurately evaluated, and the accuracy of the evaluation result is improved.
The above is a description of an embodiment of the method, and the following is a description of the scheme of the present application proceeding to step by an embodiment of the apparatus.
Fig. 4 is a schematic structural diagram of a network runtime evaluation apparatus according to an embodiment of the present application. As shown in fig. 4, includes: the method comprises the following steps: a dismantling unit 401 and an evaluation unit 402.
A disassembling unit 401, configured to disassemble the network structure to be processed, so as to obtain each neuron forming the network structure.
An evaluation unit 402, configured to determine, for each neuron, an operating time of the neuron based on the parameter of the neuron and an evaluation model obtained through pre-training, and determine an operating time of the network structure according to the operating time of each neuron.
Each network structure is composed of a plurality of neurons, so that if the running time of each neuron can be acquired respectively, the running time of the whole network structure can be determined according to the running time of each neuron.
For this reason, for the network structure to be processed, the disassembling unit 401 may first perform disassembling processing on the network structure, so as to obtain each neuron constituting the network structure, and how to perform disassembling is the prior art.
And aiming at each neuron obtained by disassembling, the running time of the neuron can be respectively obtained. In this embodiment, for each neuron obtained by disassembly, the evaluation unit 402 may determine the running time of the neuron based on the parameter of the neuron and an evaluation model obtained by pre-training.
The parameters of each neuron are known, and the evaluation model can be obtained by pre-training. The inputs to the evaluation model may be parameters of the neuron and the outputs may be the runtime of the evaluated neuron.
The evaluation unit 402 may input parameters of the neuron into each evaluation model, thereby obtaining the run time of the neuron evaluated by each evaluation model, and further may determine the run time of the neuron by fusing the obtained run times.
After the running time of each neuron is obtained, the running time of the network structure can be determined according to the running time of each neuron. Preferably, the evaluation unit 402 may add the running times of the neurons, and the sum is used as the running time of the network structure.
Fig. 5 is a schematic structural diagram of an evaluation model generation apparatus according to an embodiment of the present application. As shown in fig. 5, includes: an acquisition unit 501, a construction unit 502 and a training unit 503.
An obtaining unit 501 is configured to select a part of neurons from neurons in a predetermined range.
A constructing unit 502, configured to respectively construct, for each selected neuron, training data corresponding to the neuron, where the training data includes a parameter of the neuron and a running time of the neuron;
the training unit 503 is configured to train the evaluation model according to the training data, so that when the operation time of the network structure to be processed needs to be evaluated, the operation time of each neuron constituting the network structure is determined by using the evaluation model and the parameters of the neuron, and the operation time of the network structure is determined according to the operation time of each neuron.
In this embodiment, the neurons in the predetermined range may refer to all possible neurons in the search space. All possible neurons can be enumerated first according to the search space, i.e., which neurons, parameters of different neurons, etc. For all possible neurons in the search space, the obtaining unit 501 may randomly select a part of the neurons, for example, select about 20% of the neurons, for constructing the training data.
The constructing unit 502 may respectively construct training data corresponding to each selected neuron, where the training data may include parameters of the neuron and a running time of the neuron. And for each selected neuron, respectively sending the neuron to a specified device for running, thereby obtaining the running time of the neuron. For example, the neuron may be sent to a cell phone or other embedded device to run to obtain the actual running time of the neuron.
The training unit 503 may train to obtain an evaluation model according to the constructed training data. The specific model for the evaluation model is not limited, and may include a random forest model, a decision tree model, an XGBoost model, and the like.
Therefore, when the running time of the network structure to be processed needs to be evaluated, the evaluation model and the parameters of the neurons can be used for respectively determining the running time of each neuron forming the network structure, and further the running time of the network structure can be determined according to the running time of each neuron.
Thus, when the running time of the network structure to be processed needs to be evaluated, for any neuron, the parameters of the neuron can be respectively input into each evaluation model, so that the running time of the neuron evaluated by each evaluation model is obtained, and the obtained running times can be fused to determine the running time of the neuron.
The application also discloses kinds of network structure running time evaluation systems, which comprise a network structure running time evaluation device shown in fig. 4 and an evaluation model generation device shown in fig. 5.
For the specific work flow of the above device and system embodiments, please refer to the related description in the foregoing method embodiments, and further description is omitted.
FIG. 6 illustrates a block diagram of an exemplary computer system/server 12 suitable for use in implementing embodiments of the present application, the computer system/server 12 shown in FIG. 6 is merely an example and should not impose any limitations on the scope of use or functionality of embodiments of the present application.
As shown in FIG. 6, the computer system/server 12 is in the form of a general purpose computing device, the components of the computer system/server 12 may include, but are not limited to or more processors (processing units) 16, memory 28, and a bus 18 that connects the various system components, including the memory 28 and the processors 16.
Computer system/server 12 typically includes a variety of computer system readable media. Such media may be any available media that is accessible by computer system/server 12 and includes both volatile and nonvolatile media, removable and non-removable media.
Program/utility 40 having sets (at least ) of program modules 42 may be stored, for example, in memory 28, such program modules 42 including, but not limited to, an operating system, or more application programs, other program modules, and program data, each or some combination of these examples possibly including implementation of a network environment.
The computer system/server 12 may also communicate with or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), or more devices that enable a user to interact with the computer system/server 12, and/or any device (e.g., network card, modem, etc.) that enables the computer system/server 12 to communicate with or more other computing devices.this communication may be via an input/output (I/O) interface 22. furthermore, the computer system/server 12 may also communicate with or more networks (e.g., Local Area Network (LAN), domain network (WAN) and/or public network, such as the Internet) via a network adapter 20. As shown in FIG. 6, the network adapter 20 communicates with other modules of the computer system/server 12 via a bus 18. it should be understood that, although not shown, other hardware and/or software modules may be used in conjunction with the computer system/server 12, including, but not limited to, microcode, device drivers, redundant processing units, external drive arrays, disk drive systems, RAID disk storage systems, and data storage systems, etc.
The processor 16 executes various functional applications and data processing by executing programs stored in the memory 28, for example implementing the methods in the embodiments shown in fig. 1, 2 or 3.
The present application also discloses computer-readable storage media having stored thereon a computer program which, when executed by a processor, will implement the method as in the embodiments of fig. 1, 2 or 3.
A more specific example (a non-exhaustive list) of the computer readable storage medium includes an electrical connection having or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
A computer readable signal medium may include a propagated data signal with computer readable program code embodied therein, for example, in baseband or as part of a carrier wave .
Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.
Computer program code for carrying out operations for aspects of the present application may be written in or more programming languages, including an object oriented programming language such as Java, Smalltalk, C + +, or a combination thereof, as well as conventional procedural programming languages, such as the "C" programming language or similar programming languages.
For example, the above-described device embodiments are merely illustrative, and for example, the division of the units into only logical functional divisions, and other divisions may be realized in practice.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, that is, may be located in places, or may also be distributed on multiple network units.
In addition, functional units in the embodiments of the present application may be integrated into processing units, or each unit may exist alone physically, or two or more units are integrated into units.
The software functional unit is stored in storage media and includes several instructions to make computer devices (which may be personal computers, servers, or network devices) or processors (processors) execute part of the steps of the methods described in the embodiments of the present application.
The above description is only exemplary of the present application and should not be taken as limiting the present application, as any modification, equivalent replacement, or improvement made within the spirit and principle of the present application should be included in the scope of protection of the present application.
Claims (19)
- The method for evaluating the running time of the network structures is characterized by comprising the following steps of:disassembling a network structure to be processed to obtain each neuron forming the network structure;for each neuron, determining the operation time of the neuron based on the parameters of the neuron and an evaluation model obtained by pre-training;and determining the running time of the network structure according to the running time of each neuron.
- 2. The method of claim 1,the number of the evaluation models is or more than ;when the number of the evaluation models is greater than , the determining the running time of the neuron based on the parameters of the neuron and the pre-trained evaluation models comprises:and respectively inputting the parameters of the neurons into each evaluation model to obtain the operation time of the neurons evaluated by each evaluation model, and fusing the obtained operation time to determine the operation time of the neurons.
- 3. The method of claim 2,determining the operation time of the neuron by the operation time obtained by the fusion comprises the following steps: and calculating the mean value of the obtained running times, and taking the mean value as the running time of the neuron.
- 4. The method of claim 1,the determining the operation time of the network structure according to the operation time of each neuron includes: adding the running time of each neuron, and taking the added sum as the running time of the network structure.
- 5, evaluation model generation methods, which are characterized by comprising the following steps:selecting a part of neurons from neurons in a preset range;respectively constructing training data corresponding to each selected neuron, wherein the training data comprise parameters of the neuron and the running time of the neuron;and training according to the training data to obtain an evaluation model so as to respectively determine the running time of each neuron forming the network structure by using the evaluation model and the parameters of the neuron when the running time of the network structure to be processed needs to be evaluated, and determining the running time of the network structure according to the running time of each neuron.
- 6. The method of claim 5,the selecting a part of neurons from the neurons in the predetermined range includes: randomly selecting part of neurons from all possible neurons in the search space.
- 7. The method of claim 5,the method further includes sending the selected neurons to a designated device for operation, respectively, to obtain the operation time of the neurons.
- 8. The method of claim 5,the number of the evaluation models is ;or, the number of the evaluation models is greater than , so that when the operation time evaluation needs to be performed on the network structure to be processed, for any neuron, the parameters of the neuron are respectively input into each evaluation model to obtain the operation time of the neuron evaluated by each evaluation model, and the operation time of the neuron is determined by fusing the obtained operation times.
- 9, kinds of network structure operation time assessment device, which is characterized in that it includes a disassembling unit and an assessment unit;the disassembling unit is used for disassembling the network structure to be processed to obtain each neuron forming the network structure;and the evaluation unit is used for determining the operation time of the neurons and determining the operation time of the network structure according to the operation time of each neuron based on the parameters of the neurons and an evaluation model obtained by pre-training aiming at each neuron.
- 10. The apparatus of claim 9,the number of the evaluation models is or more than ;when the number of the evaluation models is larger than , the evaluation unit inputs the parameters of the neurons into each evaluation model respectively aiming at each neuron to obtain the operation time of the neurons evaluated by each evaluation model, and the operation time of the neurons is determined by fusing the obtained operation times.
- 11. The apparatus of claim 10,the evaluation unit calculates the mean value of the neuron operation time evaluated by each evaluation model for each neuron, and takes the mean value as the neuron operation time.
- 12. The apparatus of claim 9,the evaluation unit adds the running times of the neurons, and takes the sum as the running time of the network structure.
- The evaluation model generation device is characterized by comprising an acquisition unit, a construction unit and a training unit;the acquisition unit is used for selecting part of neurons from the neurons in a preset range;the constructing unit is configured to respectively construct training data corresponding to the selected neurons, where the training data includes parameters of the neurons and a running time of the neurons;the training unit is used for training according to the training data to obtain an evaluation model, so that when the operation time of the network structure to be processed needs to be evaluated, the operation time of each neuron forming the network structure is determined by using the evaluation model and parameters of the neuron, and the operation time of the network structure is determined according to the operation time of each neuron.
- 14. The apparatus of claim 13,the acquisition unit randomly selects a part of neurons from all possible neurons in a search space.
- 15. The apparatus of claim 13,the building unit is further configured to, at step , send the selected neuron to a specific device for operation, so as to obtain an operation time of the neuron.
- 16. The apparatus of claim 13,the number of the evaluation models is ;or, the number of the evaluation models is greater than , so that when the operation time evaluation needs to be performed on the network structure to be processed, for any neuron, the parameters of the neuron are respectively input into each evaluation model to obtain the operation time of the neuron evaluated by each evaluation model, and the operation time of the neuron is determined by fusing the obtained operation times.
- The runtime evaluation system for network structures of 17 and , comprising:a network structure runtime assessment apparatus as claimed in any of claims 9-12 and an assessment model generation apparatus as claimed in any of claims 13-16.
- Computer apparatus of comprising a memory, a processor, and a computer program stored on said memory and executable on said processor, wherein said processor when executing said program implements a method of any of claims 1-8 as claimed in .
- 19, computer readable storage medium, on which a computer program is stored, which program, when being executed by a processor, is adapted to carry out the method of any of claims 1-8, .
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910859244.8A CN110738318B (en) | 2019-09-11 | 2019-09-11 | Network structure operation time evaluation and evaluation model generation method, system and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910859244.8A CN110738318B (en) | 2019-09-11 | 2019-09-11 | Network structure operation time evaluation and evaluation model generation method, system and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110738318A true CN110738318A (en) | 2020-01-31 |
| CN110738318B CN110738318B (en) | 2023-05-26 |
Family
ID=69267586
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910859244.8A Active CN110738318B (en) | 2019-09-11 | 2019-09-11 | Network structure operation time evaluation and evaluation model generation method, system and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110738318B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111353601A (en) * | 2020-02-25 | 2020-06-30 | 北京百度网讯科技有限公司 | Method and apparatus for predicting delay of model structure |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106663224A (en) * | 2014-06-30 | 2017-05-10 | 亚马逊科技公司 | Interactive interfaces for machine learning model evaluations |
| US20180253645A1 (en) * | 2017-03-03 | 2018-09-06 | International Business Machines Corporation | Triage of training data for acceleration of large-scale machine learning |
| CN109472361A (en) * | 2018-11-13 | 2019-03-15 | 钟祥博谦信息科技有限公司 | Neural network optimization |
| CN110070117A (en) * | 2019-04-08 | 2019-07-30 | 腾讯科技（深圳）有限公司 | A kind of data processing method and device |
| CN110210558A (en) * | 2019-05-31 | 2019-09-06 | 北京市商汤科技开发有限公司 | Assess the method and device of neural network performance |
-
2019
- 2019-09-11 CN CN201910859244.8A patent/CN110738318B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106663224A (en) * | 2014-06-30 | 2017-05-10 | 亚马逊科技公司 | Interactive interfaces for machine learning model evaluations |
| US20180253645A1 (en) * | 2017-03-03 | 2018-09-06 | International Business Machines Corporation | Triage of training data for acceleration of large-scale machine learning |
| CN109472361A (en) * | 2018-11-13 | 2019-03-15 | 钟祥博谦信息科技有限公司 | Neural network optimization |
| CN110070117A (en) * | 2019-04-08 | 2019-07-30 | 腾讯科技（深圳）有限公司 | A kind of data processing method and device |
| CN110210558A (en) * | 2019-05-31 | 2019-09-06 | 北京市商汤科技开发有限公司 | Assess the method and device of neural network performance |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111353601A (en) * | 2020-02-25 | 2020-06-30 | 北京百度网讯科技有限公司 | Method and apparatus for predicting delay of model structure |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110738318B (en) | 2023-05-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11687811B2 (en) | Predicting user question in question and answer system | |
| JP7316453B2 (en) | Object recommendation method and device, computer equipment and medium | |
| CN114503108A (en) | Adding countermeasure robustness to a trained machine learning model | |
| CN111046176B (en) | Countermeasure sample generation method and device, electronic equipment and storage medium | |
| US20210216904A1 (en) | Knowledge Aided Feature Engineering | |
| CN105701120A (en) | Method and apparatus for determining semantic matching degree | |
| US20230359899A1 (en) | Transfer learning based on cross-domain homophily influences | |
| CN107766577B (en) | Public opinion monitoring method, device, equipment and storage medium | |
| US11164136B2 (en) | Recommending personalized job recommendations from automated review of writing samples and resumes | |
| CN113409898B (en) | Molecular structure acquisition method and device, electronic equipment and storage medium | |
| US11188517B2 (en) | Annotation assessment and ground truth construction | |
| CN110675250A (en) | Credit line management method and device based on user marketing score and electronic equipment | |
| CN117371428A (en) | Text processing method and device based on large language model | |
| CN111160049B (en) | Text translation method, apparatus, machine translation system, and storage medium | |
| CN110674397B (en) | Method, device, equipment and readable medium for training age point prediction model | |
| CN113627513A (en) | Training data generation method and system, electronic device and storage medium | |
| CN110738318A (en) | Method, system and device for evaluating network structure running time and generating evaluation model | |
| CN110728355A (en) | Neural network architecture searching method, device, computer equipment and storage medium | |
| CN111860580A (en) | Recognition model obtaining and category recognition method, device and storage medium | |
| US12033037B2 (en) | Open feature library management | |
| CN112989219B (en) | Point-of-interest recommendation method and device, electronic equipment and storage medium | |
| CN113656070A (en) | Random instruction verification method and device for processor, electronic equipment and storage medium | |
| US20190197564A1 (en) | Product space representation mapping | |
| EP3834094A1 (en) | Computer aided design of custom cellular lattice kernels according to material properties | |
| CN117151247B (en) | Method, apparatus, computer device and storage medium for modeling machine learning task |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |