CN107844833A - A kind of data processing method of convolutional neural networks, device and medium - Google Patents
A kind of data processing method of convolutional neural networks, device and medium Download PDFInfo
- Publication number
- CN107844833A CN107844833A CN201711215424.XA CN201711215424A CN107844833A CN 107844833 A CN107844833 A CN 107844833A CN 201711215424 A CN201711215424 A CN 201711215424A CN 107844833 A CN107844833 A CN 107844833A
- Authority
- CN
- China
- Prior art keywords
- data
- convolutional neural
- neural networks
- processing
- level
- 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.)
- Pending
Links
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
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F7/00—Methods or arrangements for processing data by operating upon the order or content of the data handled
- G06F7/38—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation
- G06F7/48—Methods or arrangements for performing computations using exclusively denominational number representation, e.g. using binary, ternary, decimal representation using non-contact-making devices, e.g. tube, solid state device; using unspecified devices
- G06F7/57—Arithmetic logic units [ALU], i.e. arrangements or devices for performing two or more of the operations covered by groups G06F7/483 – G06F7/556 or for performing logical operations
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- Neurology (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- General Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Compression Or Coding Systems Of Tv Signals (AREA)
- Mobile Radio Communication Systems (AREA)
Abstract
Include the invention discloses a kind of data processing method of convolutional neural networks, device and medium, the step of this method:Default configuration file is obtained, and initial configuration is carried out to establish convolutional neural networks to Caffe frameworks by configuration file;Initial data is obtained by FPGA, and each initial data is handled to obtain final result data according to the logical order between treated layers level in convolutional neural networks;Wherein, the result data of this level is transmitted to next processing level after the operation of the processing to currently incoming data is terminated and receives the new incoming data of this level to continue processing operation by treated layers level.This method realizes parallel and streamlined treatment effect, improve resource utilization, while operational performance same as the prior art is reached, the overall power consumption of relative reduction.In addition, the present invention also provides the data processing equipment and medium of a kind of convolutional neural networks, beneficial effect is as described above.
Description
Technical field
The present invention relates to data processing field, more particularly to a kind of data processing method of convolutional neural networks, device
And medium.
Background technology
Convolutional neural networks are to develop and cause a kind of efficient identification method paid attention to extensively in recent years, have been increasingly becoming
Current speech is analyzed and the study hotspot of field of image recognition, and its weights share network structure and are allowed to be more closely similar to give birth to
Thing neutral net, the complexity of network model is reduced, reduce the quantity of weights.The advantages of convolutional neural networks, is at image
It is particularly evident during reason, image is avoided feature extraction sum complicated in tional identification algorithm directly as input data
According to the process of reconstruction.
Caffe is one of efficient framework for handling convolutional neural networks, because the features such as its speed is fast, versatile is wide
General application.Current convolutional neural networks are generally to support the convolutional neural networks based on Caffe framework establishments to enter by GPU
The calculation process of row data, and then convolutional neural networks are realized to voice or the discriminance analysis of image, although GPU is with more strong
Big operational performance, but be relatively large power consumption as caused by powerful operational performance, therefore improve use relatively
The holistic cost of convolutional neural networks processing data.
As can be seen here, there is provided a kind of data processing method of convolutional neural networks, ensureing with powerful operational performance
Meanwhile overall power consumption is reduced, it is those skilled in the art's urgent problem to be solved.
The content of the invention
It is an object of the invention to provide a kind of data processing method of convolutional neural networks, device and medium, realize simultaneously
The treatment effect of row and streamlined, it can ensure that the available resources in FPGA without leaving unused, improve resource utilization, reached
While operational performance same as the prior art, the overall power consumption of relative reduction.
In order to solve the above technical problems, the present invention provides a kind of data processing method of convolutional neural networks, including:
Default configuration file is obtained, and initial configuration is carried out to establish convolution to Caffe frameworks by configuration file
Neutral net;
Initial data is obtained by FPGA, and it is each according to the logical order processing between treated layers level in convolutional neural networks
Initial data is to obtain final result data;
Wherein, treated layers level passes the result data of this level after the operation of the processing to currently incoming data is terminated
Transport to next processing level and receive the new incoming data of this level to continue processing operation;Processing operation includes being based on
The convolution conversion operation of Winograd algorithms.
Preferably, this method further comprises:
It is shown result data as classification results.
Preferably, configuration file specifically includes:
Read in network profile and weight file.
Preferably, processing level specifically includes:
Convolution levels, ReLU levels, norm levels and MaxPool levels.
Preferably, obtaining initial data by FPGA is specially:
Initial data is obtained in DDR memory by FPGA.
Preferably, initial data is specially view data.
In addition, the present invention also provides a kind of data processing equipment of convolutional neural networks, including:
Configuration module, for obtaining default configuration file, and initialization is carried out to Caffe frameworks by configuration file and matched somebody with somebody
Put to establish convolutional neural networks;
Processing module, for obtaining initial data by FPGA, and according between treated layers level in convolutional neural networks
Logical order handles each initial data to obtain final result data.
Preferably, the device further comprises:
Display module, for being shown result data as classification results.
In addition, the present invention also provides a kind of data processing equipment of convolutional neural networks, including:
Memory, for storing computer program;
Processor, the step of the data processing method of convolutional neural networks described above is realized during for performing computer program
Suddenly.
In addition, the present invention also provides a kind of computer-readable recording medium, meter is stored with computer-readable recording medium
Calculation machine program, the step of the data processing method of convolutional neural networks described above is realized when computer program is executed by processor
Suddenly.
The data processing method of convolutional neural networks provided by the present invention, by FPGA acquisition initial data, and according to
Initial data is handled according to the logical order of the processing level in convolutional neural networks, because FPGA essence is that one kind can
With the integrated circuit of customization, it is embodied in for the different disposal logic of data in different circuit functions, therefore in convolution god
Through can mark off multiple processing units with identical data processing logic, i.e. phase in each processing level under network
With functional circuit, and then each processing level can in the multiple data of synchronization parallel processing, and due to function electricity
Road can continue the data newly to be arrived to handle that are powered after having handled the electric signal for characterizing current data.Therefore treated layers level can
To carry out data processing simultaneously, result data is transmitted to next processing level or receives a upper processing in data processing
The incoming data of level, and then realize that multiple processing levels are handled the streamlined of data.Drawn further, since FPGA itself has
It is divided into the characteristic of multiple processing units, therefore when performing the convolution conversion operation of convolutional neural networks, Winograd should be used
Algorithm, the dot matrix that convolution is converted into adaptation multiplied unit multiplies, to ensure overall treatment efficiency.It can be seen that using FPGA to volume
Data under product neutral net environment are handled, and realize parallel and streamlined treatment effect, and this method can ensure
Available resources in FPGA improve resource utilization, therefore reaching operational performance same as the prior art without leaving unused
Meanwhile the power consumption that relative reduction is overall.In addition, the present invention also provides data processing equipment and Jie of a kind of convolutional neural networks
Matter, beneficial effect are as described above.
Brief description of the drawings
In order to illustrate the embodiments of the present invention more clearly, the required accompanying drawing used in embodiment will be done simply below
Introduce, it should be apparent that, drawings in the following description are only some embodiments of the present invention, for ordinary skill people
For member, on the premise of not paying creative work, other accompanying drawings can also be obtained according to these accompanying drawings.
Fig. 1 is a kind of flow chart of the data processing method of convolutional neural networks provided in an embodiment of the present invention;
Fig. 2 is a kind of data processing equipment structure chart of convolutional neural networks provided in an embodiment of the present invention.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme 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, rather than whole embodiments.Based on this
Embodiment in invention, for those of ordinary skill in the art under the premise of creative work is not made, what is obtained is every other
Embodiment, belong to the scope of the present invention.
The core of the present invention is to provide a kind of data processing method of convolutional neural networks, realizes parallel and streamlined
Treatment effect, it can ensure that the available resources in FPGA without leaving unused, improve resource utilization, reached and prior art phase
While with operational performance, the overall power consumption of relative reduction.Another core of the present invention is to provide a kind of convolutional neural networks
Data processing equipment and medium.
In order that those skilled in the art more fully understand the present invention program, with reference to the accompanying drawings and detailed description
The present invention is described in further detail.
Embodiment one
Fig. 1 is a kind of flow chart of the data processing method of convolutional neural networks provided in an embodiment of the present invention.It refer to
Fig. 1, the specific steps of the data processing method of convolutional neural networks include:
Step S10:Obtain default configuration file, and by configuration file Caffe frameworks are carried out initial configuration with
Establish convolutional neural networks.
It should be noted that the purpose of this step is to build convolutional neural networks by Caffe frameworks, in the form of framework
Realize that convolutional neural networks can farthest save the development time.Caffe is the efficient framework for handling convolutional neural networks
One of, because it has the characteristics that speed is fast, versatile and more practicality.It should be noted that the species of configuration file
With content should according to reality build demand depending on, be not specifically limited herein.
Step S11:Initial data is obtained by FPGA, and it is suitable according to the logic between treated layers level in convolutional neural networks
Sequence handles each initial data to obtain final result data.
Wherein, treated layers level passes the result data of this level after the operation of the processing to currently incoming data is terminated
Transport to next processing level and receive the new incoming data of this level to continue processing operation;Processing operation includes being based on
The convolution conversion operation of Winograd algorithms.
It should be noted that because this method is handled data by FPGA, it is therefore desirable to first pass through FPGA acquisitions
Initial data, the mode of acquisition can be the initial data that user is passed in real time, or prestore in memory
Initial data, it is not specifically limited herein.Because the convolutional Neural networking built based on Caffe frameworks is in processing initial data
When, it is necessary to which the order according to processing level is handled data step by step, therefore each processing level receives the biography of this layer
Enter data to be handled, and the data handled are passed to next layer.Due to FPGA have can streamlined calculate advantage, because
, when data volume is larger, each level can work independently from each other for this, that is, handle the data in level after processing is output,
The data newly to arrive can be immediately subjected to, compared with the single process of CPU or GPU unlatchings successively processing data, relative reduction
Overall resource consumption.Further, since FPGA often has multiple Logical processing units, therefore also contain in each processing level
There are multiple Logical processing units, when data volume is larger, single treatment level can rely on the numerous logical process lists included
First multiple data of parallel processing, the whole utilization of resource is improved, reduce power consumption.In addition, the essence using Winograd algorithms
It is that convolution is converted into dot matrix to multiply, being more applicable for FPGA has the characteristic of multiple Logical processing units.
The data processing method of convolutional neural networks provided by the present invention, by FPGA acquisition initial data, and according to
Initial data is handled according to the logical order of the processing level in convolutional neural networks, because FPGA essence is that one kind can
With the integrated circuit of customization, it is embodied in for the different disposal logic of data in different circuit functions, therefore in convolution god
Through can mark off multiple processing units with identical data processing logic, i.e. phase in each processing level under network
With functional circuit, and then each processing level can in the multiple data of synchronization parallel processing, and due to function electricity
Road can continue the data newly to be arrived to handle that are powered after having handled the electric signal for characterizing current data.Therefore treated layers level can
To carry out data processing simultaneously, result data is transmitted to next processing level or receives a upper processing in data processing
The incoming data of level, and then realize that multiple processing levels are handled the streamlined of data.Drawn further, since FPGA itself has
It is divided into the characteristic of multiple processing units, therefore when performing the convolution conversion operation of convolutional neural networks, Winograd should be used
Algorithm, the dot matrix that convolution is converted into adaptation multiplied unit multiplies, to ensure overall treatment efficiency.It can be seen that using FPGA to volume
Data under product neutral net environment are handled, and realize parallel and streamlined treatment effect, and this method can ensure
Available resources in FPGA improve resource utilization, therefore reaching operational performance same as the prior art without leaving unused
Meanwhile the power consumption that relative reduction is overall.
Embodiment two
On the basis of above-described embodiment, as a preferred embodiment, this method further comprises:
It is shown result data as classification results.
It is understood that the use of the purpose of convolutional neural networks is that the data such as image or voice are analyzed and divided
Class, therefore the result data obtained is for image or the classification results of voice.By the displaying to classification results, help
Understand the implementation status of convolutional neural networks in user, improve Consumer's Experience.
In addition, as a preferred embodiment, configuration file specifically includes:
Read in network profile and weight file.
It should be noted that it is to ensure that Caffe frameworks normally build convolution to read in network profile and weight file
Neutral net, and ensure the configuration file of the convolutional neural networks most basic function.User can also increase new match somebody with somebody as needed
File is put, is not specifically limited herein.
In addition, as a preferred embodiment, processing level specifically includes:
Convolution levels, ReLU levels, norm levels and MaxPool levels.
It should be noted that realized in FPGA convolutional neural networks data processing mainly pass through by
Convolution levels, ReLU levels, norm levels and MaxPool levels are migrated to realization in FPGA, wherein
The operand of convolution levels accounts for the overwhelming majority of whole network operand, and exists between level in data processing
Logical order, initial data is processed by the order and then obtains result data.It should be noted that user can basis
The specific needs of convolutional neural networks computing are realized in FPGA increases new level, is not specifically limited herein.
In addition, as a preferred embodiment, it is specially by FPGA acquisitions initial data:
Initial data is obtained in DDR memory by FPGA.
Because DDR memory has a higher clock frequency, therefore to the read or write speeds of data faster, and then by original evidence
Number is buffered in DDR memory, and initial data is obtained in DDR memory by FPGA, can farthest be ensured
PFGA obtains the efficiency of initial data, and then ensures the overall execution efficiency that convolutional neural networks calculate.
In addition, as a preferred embodiment, initial data is specially view data.
Because convolutional neural networks are when to image procossing, advantage is particularly evident, allows image directly as input number
According to avoiding the process of feature extraction complicated in tional identification algorithm and data reconstruction.Image is handled by convolutional neural networks
The efficiency of data is relatively higher, and integrated operation complexity is lower.
Embodiment three
Hereinbefore it is described in detail for a kind of embodiment of the data processing method of convolutional neural networks, this
Invention also provides a kind of data processing equipment of convolutional neural networks, embodiment and the implementation of method part due to device part
Example is mutually corresponding, therefore the embodiment of device part refers to the description of the embodiment of method part, wouldn't repeat here.
Fig. 2 is a kind of data processing equipment structure chart of convolutional neural networks provided in an embodiment of the present invention.Such as Fig. 2 institutes
Show, a kind of data processing equipment of convolutional neural networks provided in an embodiment of the present invention, including:
Configuration module 10, Caffe frameworks are initialized for obtaining default configuration file, and by configuration file
Configure to establish convolutional neural networks.
Processing module 11, for obtaining initial data by FPGA, and according between treated layers level in convolutional neural networks
Logical order handle each initial data to obtain final result data.
The data processing equipment of convolutional neural networks provided by the present invention, by FPGA acquisition initial data, and according to
Initial data is handled according to the logical order of the processing level in convolutional neural networks, because FPGA essence is that one kind can
With the integrated circuit of customization, it is embodied in for the different disposal logic of data in different circuit functions, therefore in convolution god
Through can mark off multiple processing units with identical data processing logic, i.e. phase in each processing level under network
With functional circuit, and then each processing level can in the multiple data of synchronization parallel processing, and due to function electricity
Road can continue the data newly to be arrived to handle that are powered after having handled the electric signal for characterizing current data.Therefore treated layers level can
To carry out data processing simultaneously, result data is transmitted to next processing level or receives a upper processing in data processing
The incoming data of level, and then realize that multiple processing levels are handled the streamlined of data.Drawn further, since FPGA itself has
It is divided into the characteristic of multiple processing units, therefore when performing the convolution conversion operation of convolutional neural networks, Winograd should be used
Algorithm, the dot matrix that convolution is converted into adaptation multiplied unit multiplies, to ensure overall treatment efficiency.It can be seen that using FPGA to volume
Data under product neutral net environment are handled, and realize parallel and streamlined treatment effect, and the present apparatus can ensure
Available resources in FPGA improve resource utilization, therefore reaching operational performance same as the prior art without leaving unused
Meanwhile the power consumption that relative reduction is overall.
On the basis of embodiment three, the device also includes:
Display module, for being shown result data as classification results.
Example IV
The present invention also provides a kind of data processing equipment of convolutional neural networks, including:
Memory, for storing computer program;
Processor, the step of the data processing method of convolutional neural networks described above is realized during for performing computer program
Suddenly.
The data processing equipment of convolutional neural networks provided by the present invention, by FPGA acquisition initial data, and according to
Initial data is handled according to the logical order of the processing level in convolutional neural networks, because FPGA essence is that one kind can
With the integrated circuit of customization, it is embodied in for the different disposal logic of data in different circuit functions, therefore in convolution god
Through can mark off multiple processing units with identical data processing logic, i.e. phase in each processing level under network
With functional circuit, and then each processing level can in the multiple data of synchronization parallel processing, and due to function electricity
Road can continue the data newly to be arrived to handle that are powered after having handled the electric signal for characterizing current data.Therefore treated layers level can
To carry out data processing simultaneously, result data is transmitted to next processing level or receives a upper processing in data processing
The incoming data of level, and then realize that multiple processing levels are handled the streamlined of data.Drawn further, since FPGA itself has
It is divided into the characteristic of multiple processing units, therefore when performing the convolution conversion operation of convolutional neural networks, Winograd should be used
Algorithm, the dot matrix that convolution is converted into adaptation multiplied unit multiplies, to ensure overall treatment efficiency.It can be seen that using FPGA to volume
Data under product neutral net environment are handled, and realize parallel and streamlined treatment effect, and the present apparatus can ensure
Available resources in FPGA improve resource utilization, therefore reaching operational performance same as the prior art without leaving unused
Meanwhile the power consumption that relative reduction is overall.
The present invention also provides a kind of computer-readable recording medium, and computer journey is stored with computer-readable recording medium
Sequence, the step of data processing method of convolutional neural networks described above is realized when computer program is executed by processor.
The computer-readable recording medium of the data processing of convolutional neural networks provided by the present invention, is obtained by FPGA
Initial data, and initial data is handled according to the logical order of the processing level in convolutional neural networks, due to
FPGA essence is a kind of customizable integrated circuit, and different circuit work(is embodied in for the different disposal logic of data
On energy, therefore it can mark off multiple handled with identical data in each processing level under convolutional neural networks and patrol
The processing unit collected, i.e. identical functional circuit, and then each processing level can be in the more numbers of synchronization parallel processing
According to, and can continue the number that is newly arrived to handle of being powered after having handled the electric signal for characterizing current data due to functional circuit
According to.Therefore treated layers level can carry out data processing simultaneously, in data processing transmit result data to next place
Manage level or receive the incoming data of upper processing level, and then realize that multiple processing levels are handled the streamlined of data.This
Outside, because FPGA itself has the characteristic for being divided into multiple processing units, therefore the convolution conversion of convolutional neural networks is being performed
During operation, Winograd algorithms should be used, the dot matrix that convolution is converted into adaptation multiplied unit multiplies, to ensure that disposed of in its entirety is imitated
Rate.It can be seen that being handled using FPGA the data under convolutional neural networks environment, parallel and streamlined processing effect is realized
Fruit, this computer-readable recording medium can ensure that available resources in FPGA without idle, improve resource utilization, therefore
Reach operational performance same as the prior art while, the overall power consumption of relative reduction.
A kind of data processing method of convolutional neural networks, device and medium provided by the present invention have been carried out in detail above
It is thin to introduce.Each embodiment is described by the way of progressive in specification, and what each embodiment stressed is and other realities
Apply the difference of example, between each embodiment identical similar portion mutually referring to.For device disclosed in embodiment
Speech, because it is corresponded to the method disclosed in Example, so description is fairly simple, related part is referring to method part illustration
.It should be pointed out that for those skilled in the art, under the premise without departing from the principles of the invention, also
Some improvement and modification can be carried out to the present invention, these are improved and modification also falls into the protection domain of the claims in the present invention
It is interior.
It should also be noted that, in this manual, such as first and second or the like relational terms be used merely to by
One entity or operation make a distinction with another entity or operation, and not necessarily require or imply these entities or operation
Between any this actual relation or order be present.Moreover, term " comprising ", "comprising" or its any other variant meaning
Covering including for nonexcludability, so that process, method, article or equipment including a series of elements not only include that
A little key elements, but also the other element including being not expressly set out, or also 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 other identical element in the process including the key element, method, article or equipment being also present.
Claims (10)
- A kind of 1. data processing method of convolutional neural networks, it is characterised in that including:Default configuration file is obtained, and initial configuration is carried out to establish convolution to Caffe frameworks by the configuration file Neutral net;Initial data is obtained by FPGA, and it is each according to the logical order processing between treated layers level in the convolutional neural networks The initial data is to obtain final result data;Wherein, each processing level passes the result data of this level after the operation of the processing to currently incoming data is terminated Transport to next processing level and receive the new incoming data of this level and operated with continuing the processing;In the processing operation Including the convolution conversion operation based on Winograd algorithms.
- 2. according to the method for claim 1, it is characterised in that this method further comprises:The result data is shown as classification results.
- 3. according to the method for claim 1, it is characterised in that the configuration file specifically includes:Read in network profile and weight file.
- 4. according to the method for claim 1, it is characterised in that the processing level specifically includes:Convolution levels, ReLU levels, norm levels and MaxPool levels.
- 5. according to the method for claim 1, it is characterised in that described to be specially by FPGA acquisitions initial data:The initial data is obtained in DDR memory by the FPGA.
- 6. according to the method described in claim 1-5 any one, it is characterised in that the initial data is specially picture number According to.
- A kind of 7. data processing equipment of convolutional neural networks, it is characterised in that including:Configuration module, for obtaining default configuration file, and initialization is carried out to Caffe frameworks by the configuration file and matched somebody with somebody Put to establish convolutional neural networks;Processing module, for obtaining initial data by FPGA, and according between treated layers level in the convolutional neural networks Logical order handles each initial data to obtain final result data.
- 8. device according to claim 7, it is characterised in that the device further comprises:Display module, for the result data to be shown as classification results.
- A kind of 9. data processing equipment of convolutional neural networks, it is characterised in that including:Memory, for storing computer program;Processor, the convolutional neural networks as described in any one of claim 1 to 6 are realized during for performing the computer program Data processing method the step of.
- 10. a kind of computer-readable recording medium, it is characterised in that be stored with computer on the computer-readable recording medium Program, the convolutional neural networks as described in any one of claim 1 to 6 are realized when the computer program is executed by processor The step of data processing method.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711215424.XA CN107844833A (en) | 2017-11-28 | 2017-11-28 | A kind of data processing method of convolutional neural networks, device and medium |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201711215424.XA CN107844833A (en) | 2017-11-28 | 2017-11-28 | A kind of data processing method of convolutional neural networks, device and medium |
Publications (1)
Publication Number | Publication Date |
---|---|
CN107844833A true CN107844833A (en) | 2018-03-27 |
Family
ID=61680322
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201711215424.XA Pending CN107844833A (en) | 2017-11-28 | 2017-11-28 | A kind of data processing method of convolutional neural networks, device and medium |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107844833A (en) |
Cited By (15)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108564168A (en) * | 2018-04-03 | 2018-09-21 | 中国科学院计算技术研究所 | A kind of design method to supporting more precision convolutional neural networks processors |
CN108961246A (en) * | 2018-07-10 | 2018-12-07 | 吉林大学 | A kind of scanning electron microscope image hole recognition methods based on artificial intelligence |
CN109359729A (en) * | 2018-09-13 | 2019-02-19 | 深思考人工智能机器人科技(北京)有限公司 | It is a kind of to realize data cached system and method on FPGA |
CN109359730A (en) * | 2018-09-26 | 2019-02-19 | 中国科学院计算技术研究所 | Neural network processor towards fixed output normal form Winograd convolution |
CN109558329A (en) * | 2018-12-10 | 2019-04-02 | 广东浪潮大数据研究有限公司 | A kind of program detecting method, device, equipment and readable storage medium storing program for executing |
CN109685210A (en) * | 2018-12-29 | 2019-04-26 | 百度在线网络技术(北京)有限公司 | Convolutional neural networks processing method, convolutional neural networks device |
CN109740747A (en) * | 2018-12-29 | 2019-05-10 | 北京中科寒武纪科技有限公司 | Operation method, device and Related product |
CN110097172A (en) * | 2019-03-18 | 2019-08-06 | 中国科学院计算技术研究所 | A kind of convolutional neural networks data processing method and device based on winograd convolution algorithm |
CN110163337A (en) * | 2018-11-12 | 2019-08-23 | 腾讯科技(深圳)有限公司 | Data processing method, device, equipment and storage medium neural network based |
CN110222760A (en) * | 2019-06-04 | 2019-09-10 | 东南大学 | A kind of fast image processing method based on winograd algorithm |
CN110377472A (en) * | 2019-07-25 | 2019-10-25 | 北京中星微电子有限公司 | The method and device of positioning chip run-time error |
US10635951B1 (en) | 2018-10-24 | 2020-04-28 | Alibaba Group Holding Limited | Fast computation of a convolutional neural network |
CN111142925A (en) * | 2019-12-23 | 2020-05-12 | 山东浪潮通软信息科技有限公司 | Pipeline type data processing method, equipment and storage medium |
CN111416743A (en) * | 2020-03-19 | 2020-07-14 | 华中科技大学 | Convolutional network accelerator, configuration method and computer readable storage medium |
WO2022246639A1 (en) * | 2021-05-25 | 2022-12-01 | Nvidia Corporation | Hardware circuit for deep learning task scheduling |
Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101246508A (en) * | 2008-02-26 | 2008-08-20 | 江苏大学 | Neural network missing data estimation machine and evaluation method based on FPGA |
CN105956660A (en) * | 2016-05-16 | 2016-09-21 | 浪潮集团有限公司 | Neural network chip realization method used for real-time image identification |
CN106228240A (en) * | 2016-07-30 | 2016-12-14 | 复旦大学 | Degree of depth convolutional neural networks implementation method based on FPGA |
-
2017
- 2017-11-28 CN CN201711215424.XA patent/CN107844833A/en active Pending
Patent Citations (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101246508A (en) * | 2008-02-26 | 2008-08-20 | 江苏大学 | Neural network missing data estimation machine and evaluation method based on FPGA |
CN105956660A (en) * | 2016-05-16 | 2016-09-21 | 浪潮集团有限公司 | Neural network chip realization method used for real-time image identification |
CN106228240A (en) * | 2016-07-30 | 2016-12-14 | 复旦大学 | Degree of depth convolutional neural networks implementation method based on FPGA |
Cited By (25)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN108564168A (en) * | 2018-04-03 | 2018-09-21 | 中国科学院计算技术研究所 | A kind of design method to supporting more precision convolutional neural networks processors |
CN108564168B (en) * | 2018-04-03 | 2021-03-09 | 中国科学院计算技术研究所 | Design method for neural network processor supporting multi-precision convolution |
CN108961246A (en) * | 2018-07-10 | 2018-12-07 | 吉林大学 | A kind of scanning electron microscope image hole recognition methods based on artificial intelligence |
CN108961246B (en) * | 2018-07-10 | 2021-05-07 | 吉林大学 | Scanning electron microscope image pore identification method based on artificial intelligence |
CN109359729A (en) * | 2018-09-13 | 2019-02-19 | 深思考人工智能机器人科技(北京)有限公司 | It is a kind of to realize data cached system and method on FPGA |
CN109359729B (en) * | 2018-09-13 | 2022-02-22 | 深思考人工智能机器人科技(北京)有限公司 | System and method for realizing data caching on FPGA |
CN109359730A (en) * | 2018-09-26 | 2019-02-19 | 中国科学院计算技术研究所 | Neural network processor towards fixed output normal form Winograd convolution |
US10635951B1 (en) | 2018-10-24 | 2020-04-28 | Alibaba Group Holding Limited | Fast computation of a convolutional neural network |
RU2722473C1 (en) * | 2018-10-24 | 2020-06-01 | Алибаба Груп Холдинг Лимитед | Fast calculation of convolutional neural network |
CN110163337A (en) * | 2018-11-12 | 2019-08-23 | 腾讯科技(深圳)有限公司 | Data processing method, device, equipment and storage medium neural network based |
CN109558329A (en) * | 2018-12-10 | 2019-04-02 | 广东浪潮大数据研究有限公司 | A kind of program detecting method, device, equipment and readable storage medium storing program for executing |
CN109740747A (en) * | 2018-12-29 | 2019-05-10 | 北京中科寒武纪科技有限公司 | Operation method, device and Related product |
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 |
US11893414B2 (en) | 2018-12-29 | 2024-02-06 | Cambricon Technologies Corporation Limited | Operation method, device and related products |
CN109685210A (en) * | 2018-12-29 | 2019-04-26 | 百度在线网络技术(北京)有限公司 | Convolutional neural networks processing method, convolutional neural networks device |
CN110097172A (en) * | 2019-03-18 | 2019-08-06 | 中国科学院计算技术研究所 | A kind of convolutional neural networks data processing method and device based on winograd convolution algorithm |
CN110222760B (en) * | 2019-06-04 | 2023-05-23 | 东南大学 | Quick image processing method based on winograd algorithm |
CN110222760A (en) * | 2019-06-04 | 2019-09-10 | 东南大学 | A kind of fast image processing method based on winograd algorithm |
CN110377472A (en) * | 2019-07-25 | 2019-10-25 | 北京中星微电子有限公司 | The method and device of positioning chip run-time error |
CN110377472B (en) * | 2019-07-25 | 2021-05-18 | 重庆中星微人工智能芯片技术有限公司 | Method and device for positioning operation error of chip |
CN111142925A (en) * | 2019-12-23 | 2020-05-12 | 山东浪潮通软信息科技有限公司 | Pipeline type data processing method, equipment and storage medium |
CN111416743A (en) * | 2020-03-19 | 2020-07-14 | 华中科技大学 | Convolutional network accelerator, configuration method and computer readable storage medium |
CN111416743B (en) * | 2020-03-19 | 2021-09-03 | 华中科技大学 | Convolutional network accelerator, configuration method and computer readable storage medium |
WO2022246639A1 (en) * | 2021-05-25 | 2022-12-01 | Nvidia Corporation | Hardware circuit for deep learning task scheduling |
US11983566B2 (en) | 2021-05-25 | 2024-05-14 | Nvidia Corporation | Hardware circuit for deep learning task scheduling |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107844833A (en) | A kind of data processing method of convolutional neural networks, device and medium | |
CN109740747B (en) | Operation method, device and Related product | |
CN106529670B (en) | It is a kind of based on weight compression neural network processor, design method, chip | |
CN104915322B (en) | A kind of hardware-accelerated method of convolutional neural networks | |
JP2019036298A (en) | Intelligent high bandwidth memory system and logic dies therefor | |
CN106503791A (en) | System and method for the deployment of effective neutral net | |
CN107480789A (en) | The efficient conversion method and device of a kind of deep learning model | |
CN108549583A (en) | Big data processing method, device, server and readable storage medium storing program for executing | |
CN106951926A (en) | The deep learning systems approach and device of a kind of mixed architecture | |
CN109754068A (en) | Transfer learning method and terminal device based on deep learning pre-training model | |
CN109086134A (en) | A kind of operation method and device of deep learning operation | |
CN107491811A (en) | Method and system and neural network processor for accelerans network processing unit | |
CN110163362A (en) | A kind of computing device and method | |
CN110276447A (en) | A kind of computing device and method | |
CN104536832A (en) | Virtual machine deployment method | |
Jiang et al. | Federated learning algorithm based on knowledge distillation | |
CN110351145A (en) | A kind of radio network functions method of combination of the virtualization based on economic benefit | |
CN108446758B (en) | Artificial intelligence calculation-oriented neural network data serial flow processing method | |
CN109086871A (en) | Training method, device, electronic equipment and the computer-readable medium of neural network | |
CN109146765A (en) | A kind of image processing method, primary processor, coprocessor and electronic equipment | |
CN108494705A (en) | A kind of network message high_speed stamping die and method | |
CN107346444A (en) | Intelligent chip based on high in the clouds big data deep learning | |
CN104468379B (en) | Virtual Hadoop clustered nodes system of selection and device based on most short logical reach | |
CN102646262A (en) | Reconfigurable visual preprocessor and visual processing system | |
CN109711367A (en) | Operation method, device and Related product |
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 | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180327 |
|
RJ01 | Rejection of invention patent application after publication |