CN107832262A - Convolution algorithm method and device - Google Patents
Convolution algorithm method and device Download PDFInfo
- Publication number
- CN107832262A CN107832262A CN201710983505.8A CN201710983505A CN107832262A CN 107832262 A CN107832262 A CN 107832262A CN 201710983505 A CN201710983505 A CN 201710983505A CN 107832262 A CN107832262 A CN 107832262A
- Authority
- CN
- China
- Prior art keywords
- convolution algorithm
- data
- epicycle
- module
- convolution
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/15—Correlation function computation including computation of convolution operations
- G06F17/153—Multidimensional correlation or convolution
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Artificial Intelligence (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Evolutionary Computation (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Algebra (AREA)
- Databases & Information Systems (AREA)
- Neurology (AREA)
- Complex Calculations (AREA)
Abstract
The invention discloses a kind of convolution algorithm method and device.Wherein, this method includes：During epicycle convolution algorithm is carried out, untreated first data of last round of convolution algorithm are obtained from pending data, and the second data are obtained from convolution algorithm module；Epicycle convolution algorithm is carried out according to above-mentioned first data and above-mentioned second data；The partial data of epicycle convolution algorithm is stored in above-mentioned convolution algorithm module, second data of the above-mentioned partial data as next round convolution algorithm.The present invention solve due to convolution algorithm power consumption it is larger caused by the heavier technical problem of processor load.
Description
Technical field
The present invention relates to image processing field, in particular to a kind of convolution algorithm method and device.
Background technology
In the application of convolutional neural networks, convolution algorithm occupies most of resource of processor.Particularly carrying out
During extensive convolution algorithm, the performance requirement of processor is also greatly increased, the problem of causing processor load heavier.Existing skill
Art is to carry out convolution algorithm using performance more preferably processor mostly, however, this can undoubtedly cause into when solving the problem
This increase.
For it is above-mentioned the problem of, not yet propose effective solution at present.
The content of the invention
The embodiments of the invention provide a kind of convolution algorithm method and device, with least solve due to convolution algorithm power consumption compared with
The heavier technical problem of processor load caused by big.
One side according to embodiments of the present invention, there is provided a kind of convolution algorithm method, including：Carrying out epicycle convolution
During computing, untreated first data of last round of convolution algorithm are obtained from pending data, and from convolution algorithm
The second data are obtained in module；Epicycle convolution algorithm is carried out according to above-mentioned first data and above-mentioned second data；By epicycle convolution
The partial data of computing is stored in above-mentioned convolution algorithm module, second number of the above-mentioned partial data as next round convolution algorithm
According to.
Alternatively, above-mentioned untreated first data of last round of convolution algorithm that obtained from pending data include：Pass through
Mobile convolution algorithm window, extracts above-mentioned first data, wherein, the stride of above-mentioned convolution algorithm window movement is less than above-mentioned convolution algorithm
The width of window.
Alternatively, after epicycle convolution algorithm is carried out according to above-mentioned first data and above-mentioned second data, the above method
Also include：The operation result obtained to epicycle convolution algorithm carries out pond computing.
Alternatively, above-mentioned pond computing includes at least one of：Maximum pond computing, average pond computing.
Alternatively, above-mentioned pending data includes streaming media data.
Another aspect according to embodiments of the present invention, a kind of convolution algorithm device is additionally provided, including：Acquisition module, use
Counted in during epicycle convolution algorithm is carried out, obtaining last round of convolution algorithm untreated first from pending data
According to, and the second data are obtained from convolution algorithm module；Convolution algorithm module, for according to above-mentioned first data and above-mentioned
Two data carry out epicycle convolution algorithm；And the partial data of storage epicycle convolution algorithm, above-mentioned partial data is as next round
Second data of convolution algorithm.
Alternatively, above-mentioned acquisition module is used to perform following steps：Last round of convolution algorithm is obtained from pending data
Untreated first data：By mobile convolution algorithm window, above-mentioned first data are extracted, wherein, above-mentioned convolution algorithm window movement
Stride be less than above-mentioned convolution algorithm window width.
Alternatively, said apparatus also includes：Pond module, the operation result for being obtained to epicycle convolution algorithm carry out pond
Change computing.
Another aspect according to embodiments of the present invention, a kind of terminal is additionally provided, including：Mobile convolution algorithm window；Processing
Device, above-mentioned processor operation program, wherein, for being held from the data of above-mentioned mobile convolution algorithm window output when said procedure is run
The following processing step of row：During epicycle convolution algorithm is carried out, last round of convolution algorithm is obtained from pending data not
First data of processing, and the second data are obtained from convolution algorithm module；According to above-mentioned first data and above-mentioned second number
According to progress epicycle convolution algorithm；The partial data of epicycle convolution algorithm is stored in above-mentioned convolution algorithm module, above-mentioned part
Second data of the data as next round convolution algorithm.
Another aspect according to embodiments of the present invention, additionally provides a kind of storage medium, and above-mentioned storage medium includes storage
Program, wherein, said procedure perform with above-mentioned arbitrary characteristics convolution algorithm method.
In embodiments of the present invention, using during epicycle convolution algorithm is carried out, obtained from pending data
One wheel untreated first data of convolution algorithm, and the second data are obtained from convolution algorithm module；According to the first data and
Second data carry out epicycle convolution algorithm；The partial data of epicycle convolution algorithm is stored in above-mentioned convolution algorithm module, portion
Mode of the divided data as the second data of next round convolution algorithm, by according to last round of convolution algorithm in convolution algorithm module
Last round of untreated first data of convolution algorithm carry out epicycle convolution algorithm in the second data and pending data that retain, should
First data refuse the second Data duplication, have reached lifting convolution algorithm treatment effeciency, have reduced the purpose of convolution algorithm power consumption, from
And the technique effect for mitigating processor load is realized, and then solve processor caused by because convolution algorithm power consumption is larger and bear
The heavier technical problem of lotus.
Brief description of the drawings
Accompanying drawing described herein is used for providing a further understanding of the present invention, forms the part of the application, this hair
Bright schematic description and description is used to explain the present invention, does not form inappropriate limitation of the present invention.In the accompanying drawings：
Fig. 1 is a kind of schematic flow sheet of optional convolution algorithm method according to embodiments of the present invention；
Fig. 2 is a kind of structural representation of optional convolution algorithm device according to embodiments of the present invention.
Embodiment
In order that those skilled in the art more fully understand the present invention program, below in conjunction with the embodiment of the present invention
Accompanying drawing, the technical scheme in the embodiment of the present invention is clearly and completely described, it is clear that described embodiment is only
The embodiment of a part of the invention, rather than whole embodiments.Based on the embodiment in the present invention, ordinary skill people
The every other embodiment that member is obtained under the premise of creative work is not made, it should all belong to the model that the present invention protects
Enclose.
It should be noted that term " first " in description and claims of this specification and above-mentioned accompanying drawing, "
Two " etc. be for distinguishing similar object, without for describing specific order or precedence.It should be appreciated that so use
Data can exchange in the appropriate case, so as to embodiments of the invention described herein can with except illustrating herein or
Order beyond those of description is implemented.In addition, term " comprising " and " having " and their any deformation, it is intended that cover
Cover it is non-exclusive include, be not necessarily limited to for example, containing the process of series of steps or unit, method, system, product or equipment
Those steps or unit clearly listed, but may include not list clearly or for these processes, method, product
Or the intrinsic other steps of equipment or unit.
Embodiment 1
According to embodiments of the present invention, there is provided a kind of embodiment of the method for convolution algorithm method is, it is necessary to illustrate, attached
The step of flow of figure illustrates can perform in the computer system of such as one group computer executable instructions, though also,
So logical order is shown in flow charts, but in some cases, can be with different from shown by order execution herein
Or the step of description.
Fig. 1 is convolution algorithm method according to embodiments of the present invention, as shown in figure 1, this method comprises the following steps：
Step S102, during epicycle convolution algorithm is carried out, last round of convolution algorithm is obtained from pending data
Untreated first data, and the second data are obtained from convolution algorithm module.
Convolution algorithm is a kind of computing for expending very much efficiency, in the application of convolutional neural networks, particularly convolution algorithm
And continuous parallel calculation, the efficiency of most of processor will be occupied.On the other hand, the multimedia application of internet such as crossfire matchmaker
Body is also more and more extensive, therefore, it is necessary to propose a kind of convolution algorithm device, when for convolution algorithm and continuously parallel calculation
Processing data when, can have excellent operational performance and low-power consumption to show, and can processing data crossfire, actually current weight
One of problem wanted.
In the application above-mentioned steps S102, untreated first data of last round of convolution algorithm are obtained from pending data
Including：By mobile convolution algorithm window, the first data are extracted, wherein, the stride of convolution algorithm window movement is less than convolution algorithm window
Width.
Wherein, pending data includes streaming media data.
Step S104, epicycle convolution algorithm is carried out according to the first data and the second data.
In the application above-mentioned steps S104, the first data do not repeat with the second data, are counted according to the first data and second
During epicycle convolution algorithm is carried out, the filter coefficient based on convolution algorithm window is carried out to the first data and the second data
Epicycle convolution algorithm.
Step S106, the partial data of epicycle convolution algorithm is stored in above-mentioned convolution algorithm module, partial data is made
For the second data of next round convolution algorithm.
In the application above-mentioned steps S106, convolutional neural networks can include multiple operation layers, such as convolutional layer, pond window
Deng the number of plies of convolutional layer and pond layer can also be multilayer, and the output of each layer can be with so-called next layer of input, for example, n-th layer
The output of convolutional layer can be the input of n-th layer pond layer, and the output of n-th layer pond layer can be the defeated of N+1 layer convolutional layers
Enter, the output of n-th layer operation layer can be the input of N+1 layer operation layers.In order to improve operation efficiency, n-th layer fortune is being carried out
When calculating the cloud calculation of layer, the partial arithmetic of N+i layer operation layers can be carried out, effectively reduces N+i layers simultaneously using calculation resources
The operand of operation layer, wherein, N, i are positive integer.
Alternatively, after epicycle convolution algorithm is carried out according to the first data and the second data, method also includes：To epicycle
The operation result that convolution algorithm obtains carries out pond computing.Wherein, pond computing includes at least one of：Maximum pondization fortune
Calculate, average pond computing.
By above-mentioned steps, by according to the second data that last round of convolution algorithm retains in convolution algorithm module and waiting to locate
Manage last round of untreated first data of convolution algorithm in data and carry out epicycle convolution algorithm, first data refuse the second data
Repeat, reached lifting convolution algorithm treatment effeciency, reduced the purpose of convolution algorithm power consumption, born it is achieved thereby that mitigating processor
The technique effect of lotus, so solve due to convolution algorithm power consumption it is larger caused by the heavier technical problem of processor load.
Embodiment 2
The embodiment of the present invention additionally provides a kind of convolution algorithm device.It should be noted that the convolution algorithm of the embodiment
Device can be used for the convolution algorithm method for performing the embodiment of the present invention.
Fig. 2 is a kind of schematic diagram of convolution algorithm device according to embodiments of the present invention.As shown in Fig. 2 the convolution algorithm
Including：Acquisition module 20 and convolution algorithm module 22.
Acquisition module 20, for during epicycle convolution algorithm is carried out, last round of volume to be obtained from pending data
Untreated first data of product computing, and the second data are obtained from convolution algorithm module；
Convolution algorithm module 22, for carrying out epicycle convolution algorithm according to first data and second data；With
And the partial data of storage epicycle convolution algorithm, second data of the partial data as next round convolution algorithm.
Alternatively, the acquisition module 20 is used to perform following steps：Last round of convolution fortune is obtained from pending data
Calculate untreated first data：By mobile convolution algorithm window, first data are extracted, wherein, the convolution algorithm window moves
Dynamic stride is less than the width of the convolution algorithm window.
Alternatively, the convolution algorithm device of the present embodiment also includes：Pond module, for what is obtained to epicycle convolution algorithm
Operation result carries out pond computing.
Alternatively, the pond computing includes at least one of：Maximum pond computing, average pond computing.
Alternatively, the pending data includes streaming media data.
In embodiments of the present invention, using during epicycle convolution algorithm is carried out, obtained from pending data
One wheel untreated first data of convolution algorithm, and the second data are obtained from convolution algorithm module；According to the first data and
Second data carry out epicycle convolution algorithm；The partial data of epicycle convolution algorithm is stored in above-mentioned convolution algorithm module, portion
Mode of the divided data as the second data of next round convolution algorithm, by according to last round of convolution algorithm in convolution algorithm module
Last round of untreated first data of convolution algorithm carry out epicycle convolution algorithm in the second data and pending data that retain, should
First data refuse the second Data duplication, have reached lifting convolution algorithm treatment effeciency, have reduced the purpose of convolution algorithm power consumption, from
And the technique effect for mitigating processor load is realized, and then solve processor caused by because convolution algorithm power consumption is larger and bear
The heavier technical problem of lotus.
Embodiment 3
The embodiment of the present invention additionally provides a kind of terminal.It should be noted that the convolution algorithm device of the embodiment can be with
For performing the convolution algorithm method of the embodiment of the present invention.
The terminal includes：Mobile convolution algorithm window；Processor, the processor operation program, wherein, described program operation
When for performing following processing step from the data of the mobile convolution algorithm window output：Carrying out the process of epicycle convolution algorithm
In, untreated first data of last round of convolution algorithm are obtained from pending data, and obtained from convolution algorithm module
Second data；Epicycle convolution algorithm is carried out according to first data and second data；By the part of epicycle convolution algorithm
Data storage is in above-mentioned convolution algorithm module, second data of the partial data as next round convolution algorithm.
The embodiment of the present invention additionally provides a kind of storage medium, and the storage medium includes the program of storage, wherein, the journey
Sequence performs the convolution algorithm method with above-mentioned arbitrary characteristics.
The embodiments of the present invention are for illustration only, do not represent the quality of embodiment.
In the above embodiment of the present invention, the description to each embodiment all emphasizes particularly on different fields, and does not have in some embodiment
The part of detailed description, it may refer to the associated description of other embodiment.
In several embodiments provided herein, it should be understood that disclosed technology contents, others can be passed through
Mode is realized.Wherein, device embodiment described above is only schematical, such as the division of the unit, Ke Yiwei
A kind of division of logic function, can there is an other dividing mode when actually realizing, for example, multiple units or component can combine or
Person is desirably integrated into another system, or some features can be ignored, or does not perform.Another, shown or discussed is mutual
Between coupling or direct-coupling or communication connection can be INDIRECT COUPLING or communication link by some interfaces, unit or module
Connect, can be electrical or other forms.
The unit illustrated as separating component can be or may not be physically separate, show as unit
The part shown can be or may not be physical location, you can with positioned at a place, or can also be distributed to multiple
On unit.Some or all of unit therein can be selected to realize the purpose of this embodiment scheme according to the actual needs.
In addition, each functional unit in each embodiment of the present invention can be integrated in a processing unit, can also
That unit is individually physically present, can also two or more units it is integrated in a unit.Above-mentioned integrated list
Member can both be realized in the form of hardware, can also be realized in the form of SFU software functional unit.
If the integrated unit is realized in the form of SFU software functional unit and is used as independent production marketing or use
When, it can be stored in a computer read/write memory medium.Based on such understanding, technical scheme is substantially
The part to be contributed in other words to prior art or all or part of the technical scheme can be in the form of software products
Embody, the computer software product is stored in a storage medium, including some instructions are causing a computer
Equipment (can be personal computer, server or network equipment etc.) perform each embodiment methods described of the present invention whole or
Part steps.And foregoing storage medium includes：USB flash disk, read-only storage (ROM, Read-Only Memory), arbitrary access are deposited
Reservoir (RAM, Random Access Memory), mobile hard disk, magnetic disc or CD etc. are various can be with store program codes
Medium.
Described above is only the preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, under the premise without departing from the principles of the invention, some improvements and modifications can also be made, these improvements and modifications also should
It is considered as protection scope of the present invention.
Claims (10)
- A kind of 1. convolution algorithm method, it is characterised in that including：During epicycle convolution algorithm is carried out, untreated first number of last round of convolution algorithm is obtained from pending data According to, and the second data are obtained from convolution algorithm module；Epicycle convolution algorithm is carried out according to first data and second data；The partial data of epicycle convolution algorithm is stored in above-mentioned convolution algorithm module, the partial data is rolled up as next round Second data of product computing.
- 2. according to the method for claim 1, it is characterised in that described that last round of convolution algorithm is obtained from pending data Untreated first data include：By mobile convolution algorithm window, first data are extracted, wherein, the stride of the convolution algorithm window movement is less than described The width of convolution algorithm window.
- 3. according to the method for claim 1, it is characterised in that according to first data and second data progress After epicycle convolution algorithm, methods described also includes：The operation result obtained to epicycle convolution algorithm carries out pond computing.
- 4. according to the method for claim 3, it is characterised in that the pond computing includes at least one of：Maximum pond Change computing, average pond computing.
- 5. method according to any one of claim 1 to 4, it is characterised in that the pending data includes crossfire matchmaker Volume data.
- A kind of 6. convolution algorithm device, it is characterised in that including：Acquisition module, for during epicycle convolution algorithm is carried out, last round of convolution algorithm to be obtained from pending data Untreated first data, and the second data are obtained from convolution algorithm module；Convolution algorithm module, for carrying out epicycle convolution algorithm according to first data and second data；And the partial data of storage epicycle convolution algorithm, second data of the partial data as next round convolution algorithm.
- 7. device according to claim 6, it is characterised in that the acquisition module is used to perform following steps：From treating Untreated first data of last round of convolution algorithm are obtained in reason data：By mobile convolution algorithm window, first data are extracted, wherein, the stride of the convolution algorithm window movement is less than described The width of convolution algorithm window.
- 8. device according to claim 6, it is characterised in that also include：Pond module, the operation result for being obtained to epicycle convolution algorithm carry out pond computing.
- A kind of 9. terminal, it is characterised in that including：Mobile convolution algorithm window；Processor, the processor operation program, wherein, for being exported from the mobile convolution algorithm window when described program is run Data perform following processing step：During epicycle convolution algorithm is carried out, last round of volume is obtained from pending data Untreated first data of product computing, and the second data are obtained from convolution algorithm module；According to first data and institute State the second data and carry out epicycle convolution algorithm；The partial data of epicycle convolution algorithm is stored in above-mentioned convolution algorithm module, Second data of the partial data as next round convolution algorithm.
- A kind of 10. storage medium, it is characterised in that the storage medium includes the program of storage, wherein, described program right of execution Profit requires the convolution algorithm method described in any one in 1 to 5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710983505.8A CN107832262A (en) | 2017-10-19 | 2017-10-19 | Convolution algorithm method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710983505.8A CN107832262A (en) | 2017-10-19 | 2017-10-19 | Convolution algorithm method and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107832262A true CN107832262A (en) | 2018-03-23 |
Family
ID=61648801
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710983505.8A Pending CN107832262A (en) | 2017-10-19 | 2017-10-19 | Convolution algorithm method and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107832262A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113128289A (en) * | 2019-12-31 | 2021-07-16 | 深圳云天励飞技术有限公司 | Feature extraction and calculation method and device for face recognition |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150309961A1 (en) * | 2014-04-28 | 2015-10-29 | Denso Corporation | Arithmetic processing apparatus |
| CN106250103A (en) * | 2016-08-04 | 2016-12-21 | 东南大学 | A kind of convolutional neural networks cyclic convolution calculates the system of data reusing |
| CN106951395A (en) * | 2017-02-13 | 2017-07-14 | 上海客鹭信息技术有限公司 | Towards the parallel convolution operations method and device of compression convolutional neural networks |
| CN108073548A (en) * | 2016-11-14 | 2018-05-25 | 耐能股份有限公司 | Convolution algorithm device and convolution algorithm method |
-
2017
- 2017-10-19 CN CN201710983505.8A patent/CN107832262A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150309961A1 (en) * | 2014-04-28 | 2015-10-29 | Denso Corporation | Arithmetic processing apparatus |
| CN106250103A (en) * | 2016-08-04 | 2016-12-21 | 东南大学 | A kind of convolutional neural networks cyclic convolution calculates the system of data reusing |
| CN108073548A (en) * | 2016-11-14 | 2018-05-25 | 耐能股份有限公司 | Convolution algorithm device and convolution algorithm method |
| CN106951395A (en) * | 2017-02-13 | 2017-07-14 | 上海客鹭信息技术有限公司 | Towards the parallel convolution operations method and device of compression convolutional neural networks |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113128289A (en) * | 2019-12-31 | 2021-07-16 | 深圳云天励飞技术有限公司 | Feature extraction and calculation method and device for face recognition |
| CN113128289B (en) * | 2019-12-31 | 2024-01-09 | 深圳云天励飞技术有限公司 | Face recognition feature extraction calculation method and equipment |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN105681628B (en) | A kind of convolutional network arithmetic element and restructural convolutional neural networks processor and the method for realizing image denoising processing | |
| CN107316079A (en) | Processing method, device, storage medium and the processor of terminal convolutional neural networks | |
| CN107145939A (en) | A kind of Neural network optimization and device | |
| CN107203808B (en) | A kind of two-value Convole Unit and corresponding two-value convolutional neural networks processor | |
| CN107749044A (en) | The pond method and device of image information | |
| CN109918201A (en) | The control method and system of task unloading | |
| CN111176820A (en) | Deep neural network-based edge computing task allocation method and device | |
| CN107491787A (en) | Local binarization CNN processing method, device, storage medium and processor | |
| CN107944545A (en) | Computational methods and computing device applied to neutral net | |
| CN111143702A (en) | Information recommendation method, system, device and medium based on graph convolution neural network | |
| CN111723900A (en) | Mapping method of neural network based on many-core processor and computing device | |
| CN109344314A (en) | A kind of data processing method, device and server | |
| CN109325530B (en) | Image classification method, storage device and processing device | |
| CN107766932A (en) | Image processing method and device based on neutral net | |
| CN107169566A (en) | Dynamic neural network model training method and device | |
| CN108320018A (en) | A kind of device and method of artificial neural network operation | |
| CN107832262A (en) | Convolution algorithm method and device | |
| CN110009644B (en) | Method and device for segmenting line pixels of feature map | |
| CN107862380A (en) | Artificial neural network computing circuit | |
| CN107220707A (en) | Dynamic neural network model training method and device based on 2-D data | |
| Chakraborty et al. | Efficient hybrid network architectures for extremely quantized neural networks enabling intelligence at the edge | |
| CN110782001A (en) | Improved method for using shared convolution kernel based on group convolution neural network | |
| CN107748914A (en) | Artificial neural network computing circuit | |
| CN109753319A (en) | A kind of device and Related product of release dynamics chained library | |
| CN107729905A (en) | Image information processing method and device |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180323 |