CN107992299B - Neural network hyper-parameter extraction and conversion method, system, device and storage medium - Google Patents
Neural network hyper-parameter extraction and conversion method, system, device and storage medium Download PDFInfo
- Publication number
- CN107992299B CN107992299B CN201711207509.3A CN201711207509A CN107992299B CN 107992299 B CN107992299 B CN 107992299B CN 201711207509 A CN201711207509 A CN 201711207509A CN 107992299 B CN107992299 B CN 107992299B
- Authority
- CN
- China
- Prior art keywords
- hyper
- parameter
- file
- format
- network configuration
- 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.)
- Active
Links
Images
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
The application discloses a neural network hyper-parameter extraction and conversion method, a system, a device and a computer readable storage medium, comprising the following steps: extracting a network configuration file and a hyper-parameter storage file of a neural network in a Caffe framework by using a conversion script, and respectively acquiring network configuration parameters, hyper-parameter dimensions and position information; converting the network configuration parameters, the hyper-parameter dimensions and the position information into a format corresponding to the target template by using a conversion script, and writing the format into the target template; generating a target format file by using a target template; according to the method and the device, the network configuration file and the hyper-parameter storage file are extracted by using the conversion script, various parameters are obtained, the parameters are written into the target template, the target format file is generated by using the target template corresponding to the format of the target format file, the conversion from the hyper-parameter storage file to the FPGA operable target format file is completed, the conversion from the hyper-parameter storage file to the FPGA operable target format file is automatically completed, and the efficiency is improved.
Description
Technical Field
The invention relates to the field of deep learning heterogeneous acceleration, in particular to a neural network hyper-parameter extraction and conversion method, a system, a device and a computer readable storage medium.
Background
Caffe is a clear and efficient deep learning framework, and supports command lines, Python and MATLAB interfaces based on a C + +/CUDA architecture; the CPU and the GPU can be directly and seamlessly switched, and a set of basic programming framework is provided in the CPU and the GPU, so that algorithms such as deep convolution neural network and deep learning under a GPU parallel framework can be realized. The network configuration file based on the Caffe framework has the function of defining the network model of the network configuration file, and mainly defines all the convolutional layers and data streams connected with the convolutional layers, so that training or testing of the network model can be completed according to the network defined in the network configuration file. And storing the hyper-parameters under the Caffe framework into a hyper-parameter storage file, wherein the file stores the trained network parameters in a structured form. Because a large amount of computing power is needed for processing a large amount of data in deep learning, the development of the deep learning method faces a lot of difficulties, such as insufficient expansibility of deep learning software, insufficient computing performance, large energy consumption for recognition on a deep learning line, and the like.
Common acceleration engines for deep learning are GPUs and FPGAs. The GPU has more core computing units, so that the GPU has stronger parallel processing capability and is a common acceleration means for more deep learning computing platforms at present. However, the high price and the huge power consumption of the system cause a plurality of problems for large-scale deployment. And if the performance of the GPU is to be fully exerted, the data to be subjected to batch processing needs to reach a certain magnitude, so that the time delay of data processing is increased. The FPGA is a programmable gate array, can reconstruct a computing unit through programming, and has the characteristics of low power consumption, low time delay and high cost performance compared with a GPU.
In the prior art, a neural network hyper-parameter storage file manufactured under a Caffe framework cannot be directly transplanted to an FPGA to operate, file conversion needs to be carried out manually, and efficiency is seriously affected.
Therefore, how to develop an efficient method for transplanting the neural network hyper-parameter storage file to the FPGA for operation is a current technical difficulty.
Disclosure of Invention
In view of this, the present invention provides a method, a system, a device and a computer readable storage medium for extracting and converting a neural network hyper-parameter, which can efficiently transfer a neural network hyper-parameter storage file to an FPGA for operation. The specific scheme is as follows:
a neural network hyper-parameter extraction and conversion method comprises the following steps:
extracting a network configuration file and a hyper-parameter storage file of a neural network in a Caffe framework by using a conversion script, and respectively acquiring network configuration parameters, hyper-parameter dimensions and position information;
converting the network configuration parameters, the hyper-parameter dimensions and the position information into a format corresponding to a target template by using the conversion script, and writing the format into the target template;
generating a target format file by using the target template;
and the target template is a template corresponding to the format of the target format file.
Optionally, the process of extracting a network configuration file of a neural network in a Caffe framework by using a conversion script to obtain a network configuration parameter includes:
and traversing and extracting the configuration parameter data of each convolution layer in the network configuration file by using the conversion script to obtain the network configuration parameters.
Optionally, the process of extracting the hyper-parameter storage file of the neural network in the Caffe framework by using the conversion script to obtain the hyper-parameter dimension and the position information includes:
and acquiring the hyperparameter dimension and the position information from the hyperparameter storage file by using the conversion script according to the network configuration parameters.
Optionally, the converting script converts the network configuration parameters, the hyper-parameter dimensions, and the location information into a format corresponding to a target template, and writes the format into the target template to generate a target format file, where the process includes:
converting the network configuration parameters, the hyper-parameter dimensions and the position information into a storage format corresponding to the target template by using the conversion script;
and calling the target template to enable the conversion script to be matched with the storage format of the target template, writing the network configuration parameters, the hyper-parameter dimensions and the position information into the target template, and generating a target format file.
The invention also discloses a neural network hyper-parameter extraction and conversion system, which comprises:
the parameter extraction module is used for extracting a network configuration file and a hyper-parameter storage file of the neural network in the Caffe framework by using the conversion script and respectively acquiring network configuration parameters, hyper-parameter dimensions and position information;
the template writing module is used for converting the network configuration parameters, the hyper-parameter dimensions and the position information into a format corresponding to a target template by using the conversion script and writing the format into the target template;
the file generating module is used for generating a target format file by utilizing the target template;
and the target template is a template corresponding to the format of the target format file.
Optionally, the parameter extracting module includes:
and the network parameter extraction unit is used for traversing and extracting the configuration parameter data of each convolution layer in the network configuration file by using the conversion script to obtain the network configuration parameters.
Optionally, the parameter extracting module includes:
and the hyper-parameter extraction unit is used for acquiring the hyper-parameter dimension and the position information from the hyper-parameter storage file by using the conversion script according to the network configuration parameters.
Optionally, the template writing module includes:
the format conversion unit is used for converting the network configuration parameters, the hyper-parameter dimensions and the position information into a storage format corresponding to the target template by using the conversion script;
and the template writing unit is used for calling the target template, enabling the conversion script to be matched with the storage format of the target template, writing the network configuration parameters, the hyper-parameter dimensions and the position information into the target template, and generating a target format file.
The invention also discloses a neural network hyper-parameter extraction and conversion device, which comprises:
a memory to store instructions; extracting a network configuration file and a hyper-parameter storage file of a neural network in a Caffe framework by using a conversion script, and respectively acquiring network configuration parameters, hyper-parameter dimensions and position information; converting the network configuration parameters, the hyper-parameter dimensions and the position information into a format corresponding to a target template by using the conversion script, and writing the format into the target template; generating a target format file by using the target template; the target template is a template corresponding to the format of the target format file;
a processor to execute the instructions in the memory.
The invention also discloses a computer readable storage medium, wherein the computer readable storage medium is stored with a neural network hyper-parameter extraction and conversion program, and the neural network hyper-parameter extraction and conversion program realizes the steps of the neural network hyper-parameter extraction and conversion method when being executed by a processor.
The invention discloses a neural network hyper-parameter extraction and conversion method, which comprises the following steps: extracting a network configuration file and a hyper-parameter storage file of a neural network in a Caffe framework by using a conversion script, and respectively acquiring network configuration parameters, hyper-parameter dimensions and position information; converting the network configuration parameters, the hyper-parameter dimensions and the position information into a format corresponding to the target template by using a conversion script, and writing the format into the target template; generating a target format file by using a target template; the target template is a template corresponding to the format of the target format file.
The method extracts the network configuration file and the hyper-parameter storage file of the neural network in the Caffe framework by using the conversion script, respectively obtains the network configuration parameters, the hyper-parameter dimensions and the position information, thereby converting each parameter into a format corresponding to the target template, enabling the network configuration parameters, the hyper-parameter dimensions and the position information to be written into the target template, finally generating the target format file by using the target template corresponding to the format of the target format file, completing the conversion from the hyper-parameter storage file to the FPGA operable target format file, and automatically completing the conversion from the hyper-parameter storage file to the FPGA operable target format file by using the conversion script and the target template only by manually issuing a conversion command once, thereby improving the efficiency and reducing the possibility of errors caused by human errors.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly described below, it is obvious that the drawings in the following description are only embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to the provided drawings without creative efforts.
FIG. 1 is a schematic flow chart of a neural network hyper-parameter extraction and transformation method disclosed in the embodiment of the present invention;
FIG. 2 is a schematic flow chart of another neural network hyper-parameter extraction and transformation method disclosed in the embodiment of the present invention;
fig. 3 is a schematic structural diagram of a neural network hyper-parameter extraction and conversion system disclosed in the embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
It can be understood that, because the neural network established under the Caffe framework cannot be directly operated on the FPGA, the neural network established under the Caffe framework needs to be converted into a file format capable of being operated on the FPGA, and therefore, the embodiment of the present invention discloses a method for extracting and converting hyper-parameters of a neural network, which is shown in fig. 1 and includes:
step S11: and extracting a network configuration file and a hyper-parameter storage file of the neural network in the Caffe framework by using the conversion script, and respectively acquiring network configuration parameters, hyper-parameter dimensions and position information.
Specifically, after a conversion command input by a user is received, a network configuration file and a hyper-parameter storage file of the neural network established based on the Caffe framework are extracted by using a conversion script, network configuration parameters are respectively obtained from the network configuration file, and dimension and position information of the hyper-parameter are obtained from the hyper-parameter storage file.
Step S12: and converting the network configuration parameters, the hyper-parameter dimensions and the position information into a format corresponding to the target template by using the conversion script, and writing the format into the target template.
Specifically, the conversion script converts the file format of the acquired network configuration parameters, the hyper-parameter dimensions and the position information into a format corresponding to the target template, stores the format into the conversion script, and writes the network configuration parameters, the hyper-parameter dimensions and the position information into the target template by using the conversion script.
Step S13: and generating a target format file by using the target template.
Specifically, the target template is a template corresponding to the format of the target format file, that is, the target template is in the original format of the target format file which is not written with data, so that after the target template is completely filled, the target format file is generated by using the target template, and the conversion from the hyper-parameter storage file to the target format file which can be run by the FPGA is completed.
Therefore, in the embodiment of the invention, the network configuration file and the hyper-parameter storage file of the neural network in the Caffe frame are extracted by using the conversion script, the network configuration parameters, the hyper-parameter dimensions and the position information are respectively obtained, so that each parameter is converted into the format corresponding to the target template, the network configuration parameters, the hyper-parameter dimensions and the position information can be written into the target template, finally, the target template corresponding to the format of the target format file is used for generating the target format file, the conversion from the hyper-parameter storage file to the FPGA operable target format file is completed, and the conversion from the hyper-parameter storage file to the FPGA operable target format file can be automatically completed by manually issuing a conversion command by using the conversion script and the target template, so that the efficiency is improved, and the possibility of errors caused by human errors is reduced.
In the embodiment of the present invention, the conversion script may be a script edited based on Python language, the target template is a mustache template, and the target format file is an OpenCL file, and certainly, according to a change of an actual operation condition, the programming language may be changed according to an actual application requirement, which is not limited herein.
The embodiment of the invention discloses a specific neural network hyper-parameter extraction and conversion method, and compared with the previous embodiment, the embodiment further explains and optimizes the technical scheme. Referring to fig. 2, specifically:
step S21: and traversing and extracting the configuration parameter data of each convolution layer in the network configuration file by using the conversion script to obtain the network configuration parameters.
Specifically, the configuration parameter data of each convolution layer in the network configuration file is traversed by using the conversion script, and the configuration parameter data of each convolution layer is extracted after the configuration parameter data of each convolution layer is traversed to obtain configuration parameter data sets of all the convolution layers, wherein the configuration parameter data sets are network configuration parameters in the network configuration file.
For example, when the conversion script is edited based on the Python language, the conversion script selects a network configuration file of a neural network in a Caffe framework, designates the network configuration file as a Net parameter, traverses configuration parameter data of each convolutional layer in the Net parameter, and extracts the configuration parameter data of each convolutional layer at the same time to obtain the network configuration parameter.
Step S22: and acquiring the dimensional and position information of the hyper-parameters from the hyper-parameter storage file by using the conversion script according to the network configuration parameters.
Step S23: and converting the network configuration parameters, the hyper-parameter dimensions and the position information into a storage format corresponding to the target template by using the conversion script.
Specifically, the conversion script stores an array of network configuration parameters, hyper-parameter dimensions, and location information in a storage format corresponding to the target template into the conversion script.
For example, when the conversion script and the mustache template are edited based on the Python language, the dictionary of the conversion script corresponds to the dictionary of the mustache template, so that the conversion script converts the array of the network configuration parameters, the hyper-parameter dimensions and the position information into a storage format corresponding to the dictionary of the target template and stores the storage format in the dictionary of the conversion script.
Step S24: and calling the target template to enable the conversion script to be matched with the storage format of the target template, writing the network configuration parameters, the hyper-parameter dimensions and the position information into the target template, and generating a target format file.
For example, when a conversion script and a mustache template are edited based on a Python language, a target template is called, the conversion script is matched with the dictionary format of the target template, and an array of network configuration parameters, hyper-parameter dimensions and position information included in the dictionary of the conversion script is written into the target template, so that a hyper-parameter storage bin file which can be operated by an OpenCL code is generated by using the target template.
The invention further discloses a neural network hyper-parameter extraction and conversion system, as shown in fig. 3, the system comprises:
the parameter extraction module 11 is configured to extract a network configuration file and a hyper-parameter storage file of the neural network in the Caffe framework by using the conversion script, and respectively acquire a network configuration parameter, a hyper-parameter dimension and position information;
the template writing module 12 is used for converting the network configuration parameters, the hyper-parameter dimensions and the position information into a format corresponding to the target template by using the conversion script and writing the format into the target template;
a file generating module 13, configured to generate a target format file by using a target template;
the target template is a template corresponding to the format of the target format file.
Therefore, in the embodiment of the invention, the network configuration file and the hyper-parameter storage file of the neural network in the Caffe frame are extracted by using the conversion script, the network configuration parameters, the hyper-parameter dimensions and the position information are respectively obtained, so that each parameter is converted into the format corresponding to the target template, the network configuration parameters, the hyper-parameter dimensions and the position information can be written into the target template, finally, the target template corresponding to the format of the target format file is used for generating the target format file, the conversion from the hyper-parameter storage file to the FPGA operable target format file is completed, and the conversion from the hyper-parameter storage file to the FPGA operable target format file can be automatically completed by manually issuing a conversion command by using the conversion script and the target template, so that the efficiency is improved, and the possibility of errors caused by human errors is reduced.
In an embodiment of the present invention, the parameter extraction module 11 may specifically include a network parameter extraction unit and a hyper-parameter extraction unit; wherein the content of the first and second substances,
and the network parameter extraction unit is used for traversing and extracting the configuration parameter data of each convolution layer in the network configuration file by using the conversion script to obtain the network configuration parameters.
And the hyper-parameter extraction unit is used for acquiring the hyper-parameter dimension and the position information from the hyper-parameter storage file by using the conversion script according to the network configuration parameters.
The template writing module 12 may specifically include a format conversion unit and a template writing unit; wherein the content of the first and second substances,
the format conversion unit is used for converting the network configuration parameters, the hyper-parameter dimensions and the position information into a storage format corresponding to the target template by using the conversion script;
and the template writing unit is used for calling the target template, enabling the conversion script to be matched with the storage format of the target template, writing the network configuration parameters, the hyper-parameter dimensions and the position information into the target template, and generating a target format file.
In addition, the embodiment of the invention also discloses a neural network hyper-parameter extraction and conversion device, which comprises:
a memory to store instructions; the instruction comprises a network configuration file and a hyper-parameter storage file of a neural network in a Caffe framework, which are extracted by using a conversion script, and network configuration parameters, hyper-parameter dimensions and position information are respectively obtained; converting the network configuration parameters, the hyper-parameter dimensions and the position information into a format corresponding to the target template by using a conversion script, and writing the format into the target template; generating a target format file by using a target template; the target template is a template corresponding to the format of the target format file;
a processor to execute the instructions in the memory.
For more specific instructions stored in the memory, reference may be made to corresponding contents disclosed in the foregoing embodiments, and details are not repeated here.
In addition, the embodiment of the invention also discloses a computer readable storage medium, wherein a neural network hyper-parameter extraction and conversion program is stored on the computer readable storage medium, and when being executed by a processor, the neural network hyper-parameter extraction and conversion program realizes the steps of the neural network hyper-parameter extraction and conversion method in the foregoing embodiment.
Finally, it should also be noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other identical elements in a process, method, article, or apparatus that comprises the element.
Those of skill would further appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or combinations of both, and that the various illustrative components and steps have been described above generally in terms of their functionality in order to clearly illustrate this interchangeability of hardware and software. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
The detailed description is given, and the principle and the implementation mode of the invention are explained by applying specific examples in the text, and the description of the above embodiments is only used for helping to understand the method and the core idea of the invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (6)
1. A neural network hyper-parameter extraction and conversion method is characterized by comprising the following steps:
extracting a network configuration file and a hyper-parameter storage file of a neural network in a Caffe framework by using a conversion script, and respectively acquiring network configuration parameters, hyper-parameter dimensions and position information;
converting the network configuration parameters, the hyper-parameter dimensions and the position information into a format corresponding to a target template by using the conversion script, and writing the format into the target template;
generating a target format file by using the target template;
the target template is a template corresponding to the format of the target format file;
the process of extracting the network configuration file of the neural network in the Caffe framework by using the conversion script to obtain the network configuration parameters comprises the following steps:
traversing and extracting configuration parameter data of each convolution layer in the network configuration file by using the conversion script to obtain the network configuration parameters;
the process of extracting the hyper-parameter storage file of the neural network in the Caffe framework by using the conversion script and acquiring the hyper-parameter dimension and the position information comprises the following steps:
and acquiring the hyperparameter dimension and the position information from the hyperparameter storage file by using the conversion script according to the network configuration parameters.
2. The neural network hyper-parameter extraction and conversion method according to claim 1, wherein the process of converting the network configuration parameters, the hyper-parameter dimensions, and the location information into a format corresponding to a target template by using the conversion script, and writing the format into the target template to generate a target format file comprises:
converting the network configuration parameters, the hyper-parameter dimensions and the position information into a storage format corresponding to the target template by using the conversion script;
and calling the target template to enable the conversion script to be matched with the storage format of the target template, writing the network configuration parameters, the hyper-parameter dimensions and the position information into the target template, and generating a target format file.
3. A neural network hyper-parameter extraction and conversion system is characterized by comprising:
the parameter extraction module is used for extracting a network configuration file and a hyper-parameter storage file of the neural network in the Caffe framework by using the conversion script and respectively acquiring network configuration parameters, hyper-parameter dimensions and position information;
the template writing module is used for converting the network configuration parameters, the hyper-parameter dimensions and the position information into a format corresponding to a target template by using the conversion script and writing the format into the target template;
the file generating module is used for generating a target format file by utilizing the target template;
the target template is a template corresponding to the format of the target format file;
wherein, the parameter extraction module comprises:
a network parameter extraction unit, configured to traverse and extract configuration parameter data of each convolutional layer in the network configuration file by using the conversion script to obtain the network configuration parameter;
wherein, the parameter extraction module comprises:
and the hyper-parameter extraction unit is used for acquiring the hyper-parameter dimension and the position information from the hyper-parameter storage file by using the conversion script according to the network configuration parameters.
4. The neural network hyper-parameter extraction transformation system of claim 3, wherein the template writing module comprises:
the format conversion unit is used for converting the network configuration parameters, the hyper-parameter dimensions and the position information into a storage format corresponding to the target template by using the conversion script;
and the template writing unit is used for calling the target template, enabling the conversion script to be matched with the storage format of the target template, writing the network configuration parameters, the hyper-parameter dimensions and the position information into the target template, and generating a target format file.
5. A neural network hyper-parameter extraction and conversion device is characterized by comprising:
a memory to store instructions; extracting a network configuration file and a hyper-parameter storage file of a neural network in a Caffe framework by using a conversion script, and respectively acquiring network configuration parameters, hyper-parameter dimensions and position information; converting the network configuration parameters, the hyper-parameter dimensions and the position information into a format corresponding to a target template by using the conversion script, and writing the format into the target template; generating a target format file by using the target template; the target template is a template corresponding to the format of the target format file; the process of extracting the network configuration file of the neural network in the Caffe framework by using the conversion script to obtain the network configuration parameters comprises the following steps: traversing and extracting configuration parameter data of each convolution layer in the network configuration file by using the conversion script to obtain the network configuration parameters; the process of extracting the hyper-parameter storage file of the neural network in the Caffe framework by using the conversion script and acquiring the hyper-parameter dimension and the position information comprises the following steps: acquiring the hyperparameter dimension and the position information from the hyperparameter storage file by using the conversion script according to the network configuration parameters;
a processor to execute the instructions in the memory.
6. A computer-readable storage medium, wherein the computer-readable storage medium has stored thereon a neural network hyper-parameter extraction transformation program, which when executed by a processor implements the steps of the neural network hyper-parameter extraction transformation method according to claim 1 or 2.
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 CN107992299A (en) | 2018-05-04 |
CN107992299B true 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) |
Families Citing this family (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
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 |
CN110764744B (en) * | 2018-07-25 | 2023-12-08 | 赛灵思公司 | Intermediate representation generation method and device for neural network calculation |
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 |
CN109740747B (en) | 2018-12-29 | 2019-11-12 | 北京中科寒武纪科技有限公司 | Operation method, device and Related product |
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 |
CN111181758B (en) * | 2019-08-01 | 2022-02-25 | 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 |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US10489703B2 (en) * | 2015-05-20 | 2019-11-26 | Nec Corporation | 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 |
CN106022483B (en) * | 2016-05-11 | 2019-06-14 | 星环信息科技(上海)有限公司 | The method and apparatus converted between machine learning model |
CN107316013B (en) * | 2017-06-14 | 2020-04-07 | 西安电子科技大学 | Hyperspectral image classification method based on NSCT (non-subsampled Contourlet transform) and DCNN (data-to-neural network) |
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 |
-
2017
- 2017-11-27 CN CN201711207509.3A patent/CN107992299B/en active Active
Also Published As
Publication number | Publication date |
---|---|
CN107992299A (en) | 2018-05-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107992299B (en) | Neural network hyper-parameter extraction and conversion method, system, device and storage medium | |
US11061650B2 (en) | Methods and apparatus to automatically generate code for graphical user interfaces | |
US11593586B2 (en) | Object recognition with reduced neural network weight precision | |
KR102486348B1 (en) | Attention-based sequence transduction neural networks | |
CN107885756B (en) | Deep learning-based dialogue method, device and equipment | |
KR102392094B1 (en) | Sequence processing using convolutional neural networks | |
CN107273936B (en) | GAN image processing method and system | |
Berger | The improved iterative scaling algorithm: A gentle introduction | |
CN111553246B (en) | Chinese character style migration method and system based on multi-task antagonistic learning network | |
KR20190113928A (en) | Device placement optimization through reinforcement learning | |
CN111523324B (en) | Named entity recognition model training method and device | |
CN104077120A (en) | Image and text processing method, system and server | |
CN109543165A (en) | Document creation method and device based on cyclic convolution attention model | |
CN108446758B (en) | Artificial intelligence calculation-oriented neural network data serial flow processing method | |
JP2018536947A5 (en) | ||
CN108563488A (en) | Model training method and device, the method and device for building container mirror image | |
CN110472246A (en) | Work order classification method, device and storage medium | |
CN114881169A (en) | Self-supervised contrast learning using random feature corruption | |
JP2019525233A (en) | Speech recognition method and apparatus | |
CN116484836B (en) | Questionnaire generation system and method based on NLP model, electronic equipment and medium | |
CN104090932A (en) | Content recommending method and device | |
JP6633556B2 (en) | Acoustic model learning device, speech recognition device, acoustic model learning method, speech recognition method, and program | |
CN104916174A (en) | Intelligent touch and talk pen system | |
CN114299204B (en) | Three-dimensional cartoon character model generation method and device | |
CN114676835A (en) | Pattern-based neural network pruning |
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 |