CN107992299A - Neutral net hyper parameter extraction conversion method, system, device and storage medium - Google Patents
Neutral net hyper parameter extraction conversion method, system, device and storage medium Download PDFInfo
- Publication number
- CN107992299A CN107992299A CN201711207509.3A CN201711207509A CN107992299A CN 107992299 A CN107992299 A CN 107992299A CN 201711207509 A CN201711207509 A CN 201711207509A CN 107992299 A CN107992299 A CN 107992299A
- Authority
- CN
- China
- Prior art keywords
- hyper parameter
- target template
- conversion
- neutral net
- configuration parameters
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F8/00—Arrangements for software engineering
- G06F8/30—Creation or generation of source code
- G06F8/33—Intelligent editors
-
- 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
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Stored Programmes (AREA)
Abstract
This application discloses a kind of neutral net hyper parameter extraction conversion method, system, device and computer-readable recording medium, including：Using the network profile and hyper parameter storage file for changing neutral net in script extraction Caffe frames, network configuration parameters, hyper parameter dimension and positional information are obtained respectively；Using change script by network configuration parameters, hyper parameter dimension and positional information be converted to the corresponding form of target template, and be written in target template；Using target template, object format file is generated；The application extracts network profile and hyper parameter storage file using script is changed, obtain parameters, parameters are written in target template, utilize target template corresponding with the form of object format file, generate object format file, hyper parameter storage file is completed to the conversion of the FPGA object format files that can be run, conversion of the hyper parameter storage file to the FPGA object format files that can be run is automatically performed, improves efficiency.
Description
Technical field
The present invention relates to deep learning isomery to accelerate field, more particularly to a kind of neutral net hyper parameter extraction conversion side
Method, system, device and computer-readable recording medium.
Background technology
Caffe is a clear and efficient deep learning frame, based on C++/CUDA frameworks, support order line,
Python and MATLAB interfaces；A set of basic programming framework can be provided inside it in the direct seamless switchings of CPU and GPU,
To realize the depth convolutional neural networks under GPU parallel architectures, deep learning scheduling algorithm.Based on the network under Caffe frames
Configuration file plays the role of defining its network model, mainly defines all convolutional layers and the number being connected with convolutional layer
According to stream, you can according to the network defined in network profile, complete the training or test of network model.Based on Caffe frames
Lower hyper parameter is all stored under hyper parameter storage file, the network parameter that this document is finished with structured form storage training.Cause
Handling mass data for deep learning needs substantial amounts of computing capability, and the development of itself is also faced with numerous predicaments, such as depth
Learning software autgmentability is not high enough, calculated performance is not high enough, the problems such as energy consumption is big etc. is identified on deep learning line.
The common accelerating engine of deep learning has GPU and FPGA.GPU has more core calculations unit, this allows it to possess
More powerful parallel processing capability, is the current more common acceleration means of deep learning calculating platform.But its high valency
The power consumption of lattice and super large brings problems to large scale deployment.And to give full play to the performance of GPU, it is necessary to wait pending
The data of processing reach certain magnitude, and which increase the time delay of data processing.FPGA is programmable gate array, it can pass through
Reconstruction calculations unit is programmed, comparing GPU has the characteristics of low-power consumption, low delay, high performance-price ratio.
In the prior art, can not be grafted directly to based on the neutral net hyper parameter storage file made under Caffe frames
Operation on FPGA seriously affects efficiency, it is necessary to the manually conversion of progress file.
Therefore, how to research and develop one kind and neutral net hyper parameter storage file be efficiently transplanted to operation method on FPGA,
It is current technological difficulties.
The content of the invention
In view of this, it is an object of the invention to provide a kind of neutral net hyper parameter extraction conversion method, system, device
And computer-readable recording medium, neutral net hyper parameter storage file can be efficiently transplanted on FPGA and run.It has
Body scheme is as follows：
A kind of neutral net hyper parameter extracts conversion method, including：
Using the network profile and hyper parameter storage file for changing neutral net in script extraction Caffe frames, divide
Huo Qu not network configuration parameters, hyper parameter dimension and positional information；
The network configuration parameters, the hyper parameter dimension and the positional information are converted to using the conversion script
With the corresponding form of target template, and it is written in the target template；
Using the target template, object format file is generated；
Wherein, the target template is template corresponding with the form of the object format file.
Optionally, it is described to obtain network using the network profile for changing neutral net in script extraction Caffe frames
The process of parameter is configured, including：
Using the conversion script, travel through and extract the configuration parameter number of each convolutional layer in the network profile
According to obtaining the network configuration parameters.
Optionally, it is described using the hyper parameter storage file for changing neutral net in script extraction Caffe frames, obtain super
The process of parameter dimensions and positional information, including：
Using the conversion script according to the network configuration parameters, obtained from the hyper parameter storage file described super
Parameter dimensions and the positional information.
Optionally, it is described to utilize the conversion script by the network configuration parameters, the hyper parameter dimension and institute's rheme
Confidence breath be converted to the corresponding form of target template, and be written in the target template, generate object format file
Process, including：
The network configuration parameters, the hyper parameter dimension and the positional information are converted to using the conversion script
Storage format corresponding with the target template；
The target template is called, makes the storage format of target template described in the conversion scripts match, by the network
Configuration parameter, the hyper parameter dimension and the positional information are written in the target template, generate object format file.
The invention also discloses a kind of neutral net hyper parameter to extract converting system, including：
Parameter extraction module, for using change script extraction Caffe frames in neutral net network profile and
Hyper parameter storage file, obtains network configuration parameters, hyper parameter dimension and positional information respectively；
Template writing module, for using it is described conversion script by the network configuration parameters, the hyper parameter dimension and
The positional information be converted to the corresponding form of target template, and be written in the target template；
File generating module, for utilizing the target template, generates object format file；
Wherein, the target template is template corresponding with the form of the object format file.
Optionally, the parameter extraction module, including：
Network parameter extraction unit, for utilizing the conversion script, travels through and extracts every in the network profile
The configuration parameter data of a convolutional layer, obtains the network configuration parameters.
Optionally, the parameter extraction module, including：
Hyper parameter extraction unit, for utilizing the conversion script according to the network configuration parameters, from the hyper parameter
The hyper parameter dimension and the positional information are obtained in storage file.
Optionally, the template writing module, including：
Format conversion unit, for using it is described conversion script by the network configuration parameters, the hyper parameter dimension and
The positional information is converted to storage format corresponding with the target template；
Template writing unit, for calling the target template, makes depositing for target template described in the conversion scripts match
Form is stored up, the network configuration parameters, the hyper parameter dimension and the positional information are written in the target template, it is raw
Into object format file.
The invention also discloses a kind of neutral net hyper parameter to extract conversion equipment, including：
Memory, for storing instruction；Wherein, described instruction is using nerve in conversion script extraction Caffe frames
The network profile and hyper parameter storage file of network, obtain network configuration parameters, hyper parameter dimension and positional information respectively；
The network configuration parameters, the hyper parameter dimension and the positional information are converted to and target mould using the conversion script
The corresponding form of plate, and be written in the target template；Using the target template, object format file is generated；Wherein,
The target template is template corresponding with the form of the object format file；
Processor, for performing the instruction in the memory.
The invention also discloses a kind of computer-readable recording medium, god is stored with the computer-readable recording medium
Conversion program is extracted through network hyper parameter, the neutral net hyper parameter extraction conversion program is realized as before when being executed by processor
The step of stating neutral net hyper parameter extraction conversion method.
In the present invention, neutral net hyper parameter extraction conversion method, including：Extracted using script is changed in Caffe frames
The network profile and hyper parameter storage file of neutral net, obtain network configuration parameters, hyper parameter dimension and position respectively
Information；It is corresponding with target template using changing script by network configuration parameters, hyper parameter dimension and positional information and being converted to
Form, and be written in target template；Using target template, object format file is generated；Wherein, target template is and target lattice
The corresponding template of form of formula file.
The present invention is literary using the network profile of neutral net and hyper parameter storage in script extraction Caffe frames is changed
Part, obtains network configuration parameters, hyper parameter dimension and positional information, so that parameters be converted to and target template phase respectively
Corresponding form so that network configuration parameters, hyper parameter dimension and positional information can be written in target template, finally utilized
Target template corresponding with the form of object format file, generates object format file, completes hyper parameter storage file to FPGA
The conversion for the object format file that can be run, using changing script and target template, it is only necessary to as soon as time conversion command manually is assigned,
Conversion of the hyper parameter storage file to the FPGA object format files that can be run can be automatically performed, improve efficiency, reduce because
Wrong possibility caused by human error.
Brief description of the drawings
In order to illustrate more clearly about the embodiment of the present invention or technical scheme of the prior art, below will be to embodiment or existing
There is attached drawing needed in technology description to be briefly described, it should be apparent that, drawings in the following description are only this
The embodiment of invention, for those of ordinary skill in the art, without creative efforts, can also basis
The attached drawing of offer obtains other attached drawings.
Fig. 1 extracts conversion method flow diagram for a kind of neutral net hyper parameter disclosed by the embodiments of the present invention；
Fig. 2 extracts conversion method flow diagram for another neutral net hyper parameter disclosed by the embodiments of the present invention；
Fig. 3 extracts conversion system structure schematic diagram for a kind of neutral net hyper parameter disclosed by the embodiments of the present invention.
Embodiment
Below in conjunction with the attached drawing in the embodiment of the present invention, the technical solution in the embodiment of the present invention is carried out clear, complete
Site preparation describes, it is clear that described embodiment is only part of the embodiment of the present invention, instead of all the embodiments.It is based on
Embodiment in the present invention, those of ordinary skill in the art are obtained every other without making creative work
Embodiment, belongs to the scope of protection of the invention.
It is understood that since the neutral net established under Caffe frames can not directly be run on FPGA, need for this
The neutral net established under Caffe frames is converted to the file format that can be run on FPGA, the present invention is implemented for this
Example discloses a kind of neutral net hyper parameter extraction conversion method, and shown in Figure 1, this method includes：
Step S11：Stored using the network profile of neutral net and hyper parameter in script extraction Caffe frames is changed
File, obtains network configuration parameters, hyper parameter dimension and positional information respectively.
Specifically, after conversion command input by user is received, built using changing script and extracting based on Caffe frames
The network profile and hyper parameter storage file of vertical neutral net, and obtain out network from network profile respectively and match somebody with somebody
Parameter is put, excess of export parameter dimensions and positional information are obtained from hyper parameter storage file.
Step S12：Network configuration parameters, hyper parameter dimension and positional information are converted to and target mould using script is changed
The corresponding form of plate, and be written in target template.
Specifically, script is changed by the file format of the network configuration parameters got, hyper parameter dimension and positional information
Be converted to the corresponding form of target template, and be saved in conversion script in, recycle conversion script by network configuration parameters,
Hyper parameter dimension and positional information are write in target template accordingly.
Step S13：Using target template, object format file is generated.
Specifically, since target template is template corresponding with the form of object format file, i.e. target template is target
Formatted file does not write data unprocessed form, so after target template is filled in, object format is generated using target template
File, completes conversion of the hyper parameter storage file to the FPGA object format files that can be run.
As it can be seen that the network profile for changing neutral net in script extraction Caffe frames is utilized in the embodiment of the present invention
With hyper parameter storage file, network configuration parameters, hyper parameter dimension and positional information are obtained respectively, so that parameters be changed
For with the corresponding form of target template so that network configuration parameters, hyper parameter dimension and positional information can be written to target
In template, target template corresponding with the form of object format file is finally utilized, generates object format file, completes hyper parameter
Storage file to the FPGA object format files that can be run conversion, using changing script and target template, it is only necessary to manually assign
Conversion command, so that it may be automatically performed conversion of the hyper parameter storage file to the FPGA object format files that can be run, improve
Efficiency, reduces wrong possibility caused by human error.
Wherein, transfer of the embodiment of the present invention change of feet originally can be the script based on Python editor, and target template is
Mustache templates, object format file are OpenCL files, and certainly according to the change of practical operation situation, programming language can be with
It is changed according to practical application request, is not limited herein.
The embodiment of the invention discloses a kind of specific neutral net hyper parameter extraction conversion method, implement relative to upper one
Example, the present embodiment have made further instruction and optimization to technical solution.It is shown in Figure 2, specifically：
Step S21：Using script is changed, travel through and extract the configuration parameter number of each convolutional layer in network profile
According to obtaining network configuration parameters.
Specifically, using the configuration parameter data for changing each convolutional layer in script traverses network configuration file, while often
After the configuration parameter data for traversing a convolutional layer, extracted, obtain the configuration parameter data collection of all convolutional layers, match somebody with somebody
Supplemental characteristic defecate collection is put as the network configuration parameters in network profile.
For example, when conversion script is the script based on Python editor, conversion script is chosen neural in Caffe frames
The network profile of network, and network profile is appointed as Net parameters, then travel through each convolutional layer in Net parameters
Configuration parameter data, conversion script extract the configuration parameter data of each convolutional layer, obtain network configuration parameters at the same time.
Step S22：Using script is changed according to network configuration parameters, hyper parameter dimension is obtained from hyper parameter storage file
And positional information.
Step S23：Network configuration parameters, hyper parameter dimension and positional information are converted to and target mould using script is changed
The corresponding storage format of plate.
Specifically, conversion script by the array of network configuration parameters, hyper parameter dimension and positional information with target template
In corresponding storage format deposit conversion script.
For example, during conversion script and mustache templates based on Python editor, the dictionary of script itself is changed
It is corresponding with the dictionary of mustache templates, therefore, script is changed by the number of network configuration parameters, hyper parameter dimension and positional information
Group is converted to storage format corresponding with the dictionary of target template and is stored in the dictionary of conversion script itself.
Step S24：Invocation target template, makes the storage format of conversion scripts match target template, by network configuration parameters,
Hyper parameter dimension and positional information are written in target template, generate object format file.
For example, during conversion script and mustache templates based on Python editor, invocation target template, makes conversion
The dictionary format of scripts match target template, by change script dictionary include network configuration parameters, hyper parameter dimension and
In the array write-in target template of positional information, so as to be deposited using the hyper parameter that target template generation OpenCL codes can be run
Store up bin file.
The present invention also further discloses a kind of neutral net hyper parameter extraction converting system, and shown in Figure 3, this is
System includes：
Parameter extraction module 11, for utilizing the network profile for changing neutral net in script extraction Caffe frames
With hyper parameter storage file, network configuration parameters, hyper parameter dimension and positional information are obtained respectively；
Template writing module 12, script is changed by network configuration parameters, hyper parameter dimension and positional information turn for utilizing
Be changed to the corresponding form of target template, and be written in target template；
File generating module 13, for using target template, generating object format file；
Wherein, target template is template corresponding with the form of object format file.
As it can be seen that the network profile for changing neutral net in script extraction Caffe frames is utilized in the embodiment of the present invention
With hyper parameter storage file, network configuration parameters, hyper parameter dimension and positional information are obtained respectively, so that parameters be changed
For with the corresponding form of target template so that network configuration parameters, hyper parameter dimension and positional information can be written to target
In template, target template corresponding with the form of object format file is finally utilized, generates object format file, completes hyper parameter
Storage file to the FPGA object format files that can be run conversion, using changing script and target template, it is only necessary to manually assign
Conversion command, so that it may be automatically performed conversion of the hyper parameter storage file to the FPGA object format files that can be run, improve
Efficiency, reduces wrong possibility caused by human error.
In the embodiment of the present invention, above-mentioned parameter extraction module 11, can specifically include network parameter extraction unit and super ginseng
Number extraction unit；Wherein,
Network parameter extraction unit, for using script is changed, traveling through and extracting each convolutional layer in network profile
Configuration parameter data, obtain network configuration parameters.
Hyper parameter extraction unit, for using script is changed according to network configuration parameters, being obtained from hyper parameter storage file
Take hyper parameter dimension and positional information.
Above-mentioned template writing module 12, can specifically include format conversion unit and template writing unit；Wherein,
Format conversion unit, script is changed by the conversion of network configuration parameters, hyper parameter dimension and positional information for utilizing
For storage format corresponding with target template；
Template writing unit, for invocation target template, makes the storage format of conversion scripts match target template, by network
Configuration parameter, hyper parameter dimension and positional information are written in target template, generate object format file.
In addition, the embodiment of the invention also discloses a kind of neutral net hyper parameter to extract conversion equipment, including：
Memory, for storing instruction；Wherein, instruct using neutral net in conversion script extraction Caffe frames
Network profile and hyper parameter storage file, obtain network configuration parameters, hyper parameter dimension and positional information respectively；Utilize
Conversion script by network configuration parameters, hyper parameter dimension and positional information be converted to the corresponding form of target template, and write
Enter into target template；Using target template, object format file is generated；Wherein, target template is and object format file
The corresponding template of form；
Processor, for performing the instruction in memory.
More specifical instruction on being stored in above-mentioned memory may be referred to corresponding interior disclosed in previous embodiment
Hold, no longer repeated herein.
In addition, the embodiment of the invention also discloses a kind of computer-readable recording medium, on computer-readable recording medium
Neutral net hyper parameter extraction conversion program is stored with, neutral net hyper parameter extraction conversion program is realized when being executed by processor
Such as the step of neutral net hyper parameter extracts conversion method in previous embodiment.
Finally, it is to be noted that, herein, relational terms such as first and second and the like be used merely to by
One entity or operation are distinguished with another entity or operation, without necessarily requiring or implying these entities or operation
Between there are any actual relationship or order.Moreover, term " comprising ", "comprising" or its any other variant meaning
Covering non-exclusive inclusion, so that process, method, article or equipment including a series of elements not only include that
A little key elements, but also including other elements that are not explicitly listed, or further include for this process, method, article or
The intrinsic key element of equipment.In the absence of more restrictions, the key element limited by sentence "including a ...", is not arranged
Except also there are other identical element in the process, method, article or apparatus that includes the element.
Professional further appreciates that, with reference to each exemplary unit of the embodiments described herein description
And algorithm steps, can be realized with electronic hardware, computer software or the combination of the two, in order to clearly demonstrate hardware and
The interchangeability of software, generally describes each exemplary composition and step according to function in the above description.These
Function is performed with hardware or software mode actually, application-specific and design constraint depending on technical solution.Specialty
Technical staff can realize described function to each specific application using distinct methods, but this realization should not
Think beyond the scope of this invention.
Above to a kind of neutral net hyper parameter extraction conversion method, system, device and computer provided by the present invention
Readable storage medium storing program for executing is described in detail, and specific case used herein carries out the principle of the present invention and embodiment
Illustrate, the explanation of above example is only intended to help to understand the method and its core concept of the present invention；Meanwhile for ability
The those skilled in the art in domain, according to the thought of the present invention, there will be changes, comprehensive in specific embodiments and applications
Upper described, this specification content should not be construed as limiting the invention.
Claims (10)
1. a kind of neutral net hyper parameter extracts conversion method, it is characterised in that including：
Using the network profile and hyper parameter storage file for changing neutral net in script extraction Caffe frames, obtain respectively
Take network configuration parameters, hyper parameter dimension and positional information；
Using the conversion, script is by the network configuration parameters, the hyper parameter dimension and the positional information is converted to and mesh
The corresponding form of template is marked, and is written in the target template；
Using the target template, object format file is generated；
Wherein, the target template is template corresponding with the form of the object format file.
2. neutral net hyper parameter according to claim 1 extracts conversion method, it is characterised in that described utilize changes foot
The network profile of neutral net obtains the process of network configuration parameters in this extraction Caffe frames, including：
Using the conversion script, travel through and extract the configuration parameter data of each convolutional layer in the network profile, obtain
To the network configuration parameters.
3. neutral net hyper parameter according to claim 2 extracts conversion method, it is characterised in that described utilize changes foot
The hyper parameter storage file of neutral net in this extraction Caffe frames, obtains the process of hyper parameter dimension and positional information, bag
Include：
Using the conversion script according to the network configuration parameters, the hyper parameter is obtained from the hyper parameter storage file
Dimension and the positional information.
4. neutral net hyper parameter according to any one of claims 1 to 3 extracts conversion method, it is characterised in that described
The network configuration parameters, the hyper parameter dimension and the positional information are converted to and target mould using the conversion script
The corresponding form of plate, and be written in the target template, the process of object format file is generated, including：
Using the conversion, script is by the network configuration parameters, the hyper parameter dimension and the positional information is converted to and institute
State the corresponding storage format of target template；
The target template is called, makes the storage format of target template described in the conversion scripts match, by the network configuration
Parameter, the hyper parameter dimension and the positional information are written in the target template, generate object format file.
5. a kind of neutral net hyper parameter extracts converting system, it is characterised in that including：
Parameter extraction module, for being joined using the network profile for changing neutral net in script extraction Caffe frames with super
Number storage file, obtains network configuration parameters, hyper parameter dimension and positional information respectively；
Template writing module, for using the conversion script by the network configuration parameters, the hyper parameter dimension and described
Positional information be converted to the corresponding form of target template, and be written in the target template；
File generating module, for utilizing the target template, generates object format file；
Wherein, the target template is template corresponding with the form of the object format file.
6. neutral net hyper parameter according to claim 5 extracts converting system, it is characterised in that the parameter extraction mould
Block, including：
Network parameter extraction unit, for utilizing the conversion script, travels through and extracts and each rolled up in the network profile
The configuration parameter data of lamination, obtains the network configuration parameters.
7. neutral net hyper parameter according to claim 6 extracts converting system, it is characterised in that the parameter extraction mould
Block, including：
Hyper parameter extraction unit, for, according to the network configuration parameters, being stored using the conversion script from the hyper parameter
The hyper parameter dimension and the positional information are obtained in file.
8. converting system is extracted according to claim 5 to 7 any one of them neutral net hyper parameter, it is characterised in that described
Template writing module, including：
Format conversion unit, for using the conversion script by the network configuration parameters, the hyper parameter dimension and described
Positional information is converted to storage format corresponding with the target template；
Template writing unit, for calling the target template, makes the storage lattice of target template described in the conversion scripts match
Formula, the network configuration parameters, the hyper parameter dimension and the positional information are written in the target template, generate mesh
Mark formatted file.
9. a kind of neutral net hyper parameter extracts conversion equipment, it is characterised in that including：
Memory, for storing instruction；Wherein, described instruction is using neutral net in conversion script extraction Caffe frames
Network profile and hyper parameter storage file, obtain network configuration parameters, hyper parameter dimension and positional information respectively；Utilize
The conversion script is converted to the network configuration parameters, the hyper parameter dimension and the positional information and target template phase
Corresponding form, and be written in the target template；Using the target template, object format file is generated；Wherein, it is described
Target template is template corresponding with the form of the object format file；
Processor, for performing the instruction in the memory.
10. a kind of computer-readable recording medium, it is characterised in that be stored with nerve net on the computer-readable recording medium
Network hyper parameter extracts conversion program, and the neutral net hyper parameter extraction conversion program realizes that right such as will when being executed by processor
The step of seeking any one of 1 to the 4 neutral net hyper parameter extraction conversion method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711207509.3A CN107992299B (en) | 2017-11-27 | 2017-11-27 | Neural network hyper-parameter extraction and conversion method, system, device and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711207509.3A CN107992299B (en) | 2017-11-27 | 2017-11-27 | Neural network hyper-parameter extraction and conversion method, system, device and storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107992299A true CN107992299A (en) | 2018-05-04 |
| CN107992299B CN107992299B (en) | 2021-08-10 |
Family
ID=62032328
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711207509.3A Active CN107992299B (en) | 2017-11-27 | 2017-11-27 | Neural network hyper-parameter extraction and conversion method, system, device and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107992299B (en) |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108920177A (en) * | 2018-06-28 | 2018-11-30 | 郑州云海信息技术有限公司 | Mapping method of the deep learning model configuration file to FPGA configuration file |
| CN108985448A (en) * | 2018-06-06 | 2018-12-11 | 北京大学 | Neural Networks Representation standard card cage structure |
| CN109242105A (en) * | 2018-08-17 | 2019-01-18 | 第四范式（北京）技术有限公司 | Tuning method, apparatus, equipment and the medium of hyper parameter in machine learning model |
| CN109858610A (en) * | 2019-01-08 | 2019-06-07 | 广东浪潮大数据研究有限公司 | A kind of accelerated method of convolutional neural networks, device, equipment and storage medium |
| CN110689138A (en) * | 2018-12-29 | 2020-01-14 | 北京中科寒武纪科技有限公司 | Operation method, device and related product |
| CN110764744A (en) * | 2018-07-25 | 2020-02-07 | 赛灵思公司 | Intermediate representation generation method and device for neural network computation |
| CN111181758A (en) * | 2019-08-01 | 2020-05-19 | 腾讯科技（深圳）有限公司 | Configuration file generation method and device |
| WO2021031137A1 (en) * | 2019-08-21 | 2021-02-25 | 深圳鲲云信息科技有限公司 | Artificial intelligence application development system, computer device and storage medium |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160342888A1 (en) * | 2015-05-20 | 2016-11-24 | Nec Laboratories America, Inc. | Memory efficiency for convolutional neural networks operating on graphics processing units |
| CN106886487A (en) * | 2015-12-15 | 2017-06-23 | 北京京航计算通讯研究所 | Method for evaluating FPGA software reliabilities |
| CN107229969A (en) * | 2017-06-21 | 2017-10-03 | 郑州云海信息技术有限公司 | A kind of convolutional neural networks implementation method and device based on FPGA |
| CN107316013A (en) * | 2017-06-14 | 2017-11-03 | 西安电子科技大学 | Hyperspectral image classification method with DCNN is converted based on NSCT |
| CN107341545A (en) * | 2017-07-25 | 2017-11-10 | 郑州云海信息技术有限公司 | A kind of deep neural network arithmetic system and method |
| WO2017193769A1 (en) * | 2016-05-11 | 2017-11-16 | 星环信息科技（上海）有限公司 | Method and apparatus for conversion between machine learning models |
-
2017
- 2017-11-27 CN CN201711207509.3A patent/CN107992299B/en active Active
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20160342888A1 (en) * | 2015-05-20 | 2016-11-24 | Nec Laboratories America, Inc. | Memory efficiency for convolutional neural networks operating on graphics processing units |
| CN106886487A (en) * | 2015-12-15 | 2017-06-23 | 北京京航计算通讯研究所 | Method for evaluating FPGA software reliabilities |
| WO2017193769A1 (en) * | 2016-05-11 | 2017-11-16 | 星环信息科技（上海）有限公司 | Method and apparatus for conversion between machine learning models |
| CN107316013A (en) * | 2017-06-14 | 2017-11-03 | 西安电子科技大学 | Hyperspectral image classification method with DCNN is converted based on NSCT |
| CN107229969A (en) * | 2017-06-21 | 2017-10-03 | 郑州云海信息技术有限公司 | A kind of convolutional neural networks implementation method and device based on FPGA |
| CN107341545A (en) * | 2017-07-25 | 2017-11-10 | 郑州云海信息技术有限公司 | A kind of deep neural network arithmetic system and method |
Non-Patent Citations (1)
| Title |
|---|
| YIJIN GUAN 等: "FP-DNN: An Automated Framework for Mapping Deep Neural Networks onto FPGAs with RTL-HLS Hybrid Templates", 《IN PROCEEDINGS OF THE IEEE INTERNATIONAL SYMPOSIUM ON FIELD-PROGRAMMABLE CUSTOM COMPUTING MACHINES (FCCM 2017)》 * |
Cited By (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108985448A (en) * | 2018-06-06 | 2018-12-11 | 北京大学 | Neural Networks Representation standard card cage structure |
| CN108985448B (en) * | 2018-06-06 | 2020-11-17 | 北京大学 | Neural network representation standard framework structure |
| CN108920177A (en) * | 2018-06-28 | 2018-11-30 | 郑州云海信息技术有限公司 | Mapping method of the deep learning model configuration file to FPGA configuration file |
| CN110764744A (en) * | 2018-07-25 | 2020-02-07 | 赛灵思公司 | Intermediate representation generation method and device for neural network computation |
| CN110764744B (en) * | 2018-07-25 | 2023-12-08 | 赛灵思公司 | Intermediate representation generation method and device for neural network calculation |
| CN109242105A (en) * | 2018-08-17 | 2019-01-18 | 第四范式（北京）技术有限公司 | Tuning method, apparatus, equipment and the medium of hyper parameter in machine learning model |
| CN109242105B (en) * | 2018-08-17 | 2024-03-15 | 第四范式（北京）技术有限公司 | Code optimization method, device, equipment and medium |
| CN110689138A (en) * | 2018-12-29 | 2020-01-14 | 北京中科寒武纪科技有限公司 | Operation method, device and related product |
| US11893414B2 (en) | 2018-12-29 | 2024-02-06 | Cambricon Technologies Corporation Limited | Operation method, device and related products |
| CN109858610A (en) * | 2019-01-08 | 2019-06-07 | 广东浪潮大数据研究有限公司 | A kind of accelerated method of convolutional neural networks, device, equipment and storage medium |
| CN111181758A (en) * | 2019-08-01 | 2020-05-19 | 腾讯科技（深圳）有限公司 | Configuration file generation method and device |
| WO2021031137A1 (en) * | 2019-08-21 | 2021-02-25 | 深圳鲲云信息科技有限公司 | Artificial intelligence application development system, computer device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107992299B (en) | 2021-08-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107992299A (en) | Neutral net hyper parameter extraction conversion method, system, device and storage medium | |
| CN103135976B (en) | code automatic generation method and device | |
| CN101763064B (en) | Numerical control machining process design system and method of aircraft complex components facing process object | |
| CN104573122B (en) | It is a kind of from AIX platforms to the oracle database Migration tools of K UX platform migrations | |
| CN108269063A (en) | The online synergic editing method of word document and system | |
| CN104978411B (en) | A kind of automobile development method and apparatus of bullet train | |
| JP2019526143A (en) | Method and apparatus for performing transform-based learning on medical images | |
| CN103049383A (en) | Development and testing cloud system | |
| CN101464799A (en) | MPI parallel programming system based on visual modeling and automatic skeleton code generation method | |
| CN107480115B (en) | Method and system for format conversion of caffe frame residual error network configuration file | |
| CN110516817A (en) | A kind of model training data load method and device | |
| CN102830973A (en) | Method and layered structure for double-layer MVC (model, view and controller) developed by Web application with mass data | |
| CN109710357A (en) | A kind of method and system based on Unity3D engine implementation server operation | |
| CN113779663B (en) | BIM-based subway station enclosure structure three-dimensional modeling method, system and medium | |
| Biscarat et al. | Towards a realistic track reconstruction algorithm based on graph neural networks for the HL-LHC | |
| CN110263860A (en) | A kind of freeway traffic flow prediction technique and device | |
| CN103400050A (en) | Multiple-user cooperative nuclear reactor risk determining method and system | |
| CN102375897A (en) | System for generating dendrogram of welding and assembly structure of vehicle body | |
| CN105488052B (en) | Structural data is switched to the method and system of format formfile | |
| KR101087197B1 (en) | The method of linking computer aided design system with product data management system in real time | |
| CN104112052B (en) | Design method and device for side wall framework structure of electric locomotive | |
| CN102880749A (en) | 2D-3D (two-dimensional-three-dimensional) union layout method of communication satellite | |
| CN104599318B (en) | A kind of method and system of the seamless fusion of plant three-dimensional model gridding | |
| CN101216769A (en) | Hiberarchy system description language SmartC to C code automatic conversion method | |
| CN113806923A (en) | Pharmacokinetic-pharmacodynamic model hyper-parameter automatic learning method and device based on nlmixr package |
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 |