CN107609636A - A kind of polygamma function correspondingly exports the design method of feedback function artificial neuron - Google Patents
A kind of polygamma function correspondingly exports the design method of feedback function artificial neuron Download PDFInfo
- Publication number
- CN107609636A CN107609636A CN201710884964.0A CN201710884964A CN107609636A CN 107609636 A CN107609636 A CN 107609636A CN 201710884964 A CN201710884964 A CN 201710884964A CN 107609636 A CN107609636 A CN 107609636A
- Authority
- CN
- China
- Prior art keywords
- activation
- artificial neuron
- activation primitive
- feedback
- neuron
- 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
Landscapes
- Pharmaceuticals Containing Other Organic And Inorganic Compounds (AREA)
- Medicines That Contain Protein Lipid Enzymes And Other Medicines (AREA)
Abstract
A kind of polygamma function correspondingly exports the technical field of the design method of feedback function artificial neuron,It is to belong to artificial intelligence,Bionics,The technical field of circuit design,Major technique is that artificial neuron is inputted by multichannel,When accumulated value is less than minimum threshold values,Artificial neuron,It will not be activated,When cumulative value exceedes the threshold values of setting,Artificial neuron is activated,Threshold values is passed to activation primitive collection,Activation primitive collection sets multiple activation primitives,These activation primitives are activated entirely,Each activation primitive exports oneself operation result,Activation primitive collection connects with end-apparatus is selected,How many activation primitive connects with regard to how many bar,Pass through the setting to selecting end-apparatus,It is feedback that some activation primitives, which can be set,,Some activation primitives are directly passed to next layer of artificial neuron,Some activation primitives therein can also be set,Feedback and directly transmission exist simultaneously.
Description
Technical field
A kind of polygamma function correspondingly exports the technical field of the design method of feedback function artificial neuron, is to belong to artificial intelligence
Can, bionics, the technical field of circuit design, major technique is that artificial neuron is inputted by multichannel, when accumulated value is less than most
During small threshold values, artificial neuron, it will not be activated, when cumulative value exceedes the threshold values of setting, artificial neuron is activated, valve
Value passes to activation primitive collection, and activation primitive collection sets multiple activation primitives, and these activation primitives are activated entirely, each activation
Function exports oneself operation result, and activation primitive collection connects with end-apparatus is selected, and how many activation primitive connects with regard to how many bar, leads to
The setting to selecting end-apparatus is crossed, it is feedback that can set some activation primitives, and some activation primitives are to be directly passed to next layer
Artificial neuron's, some activation primitives therein can also be set, and feedback and directly transmission exist simultaneously.
Background technology
Neuron is the elementary cell for forming brain, and the brain of the mankind is that have thousands of individual neurons according to certain rule
Form, for the mankind in order to simulate human brain, the design to artificial neuron is the most important thing, has artificial neuron to form people
Work network, artificial neural network are a kind of mathematical modulos for the structure progress information processing that application is similar to cerebral nerve cynapse connection
Type.In this model, composition network is coupled to each other between substantial amounts of artificial neuron, i.e. " neutral net ", to reach processing
The purpose of information.A kind of kinetic simulation for the distributed parallel information processing algorithm structure for imitating animal nerve network behavior feature
Type., with multichannel input stimulus are received, the part that " excitement " output is produced when exceeding certain threshold value by weighted sum is dynamic to imitate for it
The working method of thing neuron, and the weight coefficient of the structure being coupled to each other by these neural components and reflection strength of association makes
Its " collective behavior " has the various complicated information processing functions.Particularly it is this macroscopically have robust, it is fault-tolerant, anti-interference,
The formation of the flexible and strong function such as adaptability, self study can not only be updated by component performance, and pass through
Complicated interconnecting relation is achieved, thus artificial neural network is a kind of connection mechanism model, has many of complication system
Key character.Artificial neural network be applied to signal transacting, data compression, pattern-recognition, robot vision, knowledge processing and its
Using prediction, evaluation and the combinatorial optimization problem such as decision problem, scheduling, route planning.It can in Control System Design
For simulating controlled device characteristic, search and study control law, realizing fuzzy and intelligent control, therefore to the design of neuron
Very important, because fairly obvious, the shape of neuron is very more, although the mankind classify it, neuron has
Thousands of kinds, therefore different neurons also possesses different functions, the present invention simply devises a kind of artificial neuron, existing
The design very simple of some neurons is single, is exactly all inputs and multiplied by weight, is then added up, subtract valve
Value, then sets activation primitive, passes to next layer of neuron.
The content of the invention
The brain of people is that many neurons are formed, therefore neuron is the elementary cell of neutral net, fairly obvious, nerve
First enormous amount, just there are the neuron of different shape, structure, physiologic character and function, neuron in the different parts of human body
Shape it is very strange very more, although the mankind classify to it, neuron has millions upon millions of kinds, therefore different nerves
Member also possesses different functions, and the present invention simply devises a kind of artificial neuron, due to existing neuron design very
It is simple single, exactly all inputs and multiplied by weight are added up, threshold values is subtracted, then activation primitive is set, pass to
Next layer of neuron, a network is so formed, and so simple design solves many forefathers of the mankind and can not solved
The problem of, tremendous influence is produced to All Around The World, but a kind of this simply artificial neuron meta structure most simply, in real world
The various shapes of neuron, various functions, therefore will invention various functions artificial neuron design, it is a kind of
Polygamma function correspondingly exports the design method of feedback function artificial neuron, it is characterized in that：Polygamma function correspondingly exports feedback function people
Work neuron is by input, artificial neuron, selects end-apparatus, output end, and feedback line forms, and input is as defeated in neuron
Enter end, receive the input of upper level artificial neuron or the input by other equipment, the effect of artificial neuron is input
Added up after value and multiplied by weight, if cumulative value is less than threshold values, then artificial neuron would not be activated, and not appoint
What reacts, if cumulative value is more than threshold values, then artificial neuron is activated, and cumulative value is just passed to activation primitive
Collect, all activation primitives are activated in activation primitive collection, and each activation primitive exports oneself operation result, activation primitive collection
Connected with end-apparatus is selected, how many activation primitive connects with regard to how many bar, by the setting to selecting end-apparatus, can set some sharp
Function living is feedback, and some activation primitives are directly passed to next layer of artificial neuron, can also set it is therein certain
A little activation primitives, directly feedback and transmission exist simultaneously, and wherein artificial neuron uses following design, and it is made up of 2 parts, and 1
It is accumulator, its effect is exactly that input is added up, if reaches threshold values, if it exceeds threshold values, just cumulative this
Value passes to activation primitive collection, and all activation primitives are activated simultaneously, and 2 activation primitive collection are made up of multiple activation primitives, and one
More than threshold values, these activation primitives are activated the cumulative value of denier entirely, are passed to by respective circuit and select end-apparatus, select end-apparatus use
Following design, it is connected with activation primitive collection, and how many activation primitive is being selected in end-apparatus in advance with regard to how many bar connection line
That circuit is designed to connect with those ports, for example, activation primitive f (x1) it can be exported from o-1.o-2, be fed back to i-
1.1-2, f (x2) are exported by i-4.i-5, and f (x3) is exported by i-6.i-7.i-8.i-9, and f (x4) is exported by i-11.i-9.i-10,
I-11 is directly transferred to next layer of neuron by feedback i-9.i-10.
Brief description of the drawings
Fig. 1 is the structure principle chart that polygamma function correspondingly exports feedback function artificial neuron, i-1.1-2.i-3.i-4.i-
5.i-6.i-7.i-8.i-9.i-10.i-11.i-12 represents input, and this input is a lot, and it is for generation to draw 12 here
Table acts on, and o-1.o-2.o-3.o-4.o-5.i-6.i-7.i-8.i-9.i-10.i-11.i-12 represents output end, this output
End is a lot, and it is for role of delegate to draw 12 here, and a-1 represents artificial neuron, and a-2 represents the insideAccumulator, a-3 generations
Table activation primitive collection, f (x1) .f (x2) .f (x3) .f (x4) represent the activation primitive of the inside, and b-1 is represented and selected end-apparatus, b-2. b-
3.b-4.b-5 represents activation primitive collection and selects the line of end-apparatus, and how many activation primitive is with regard to how many line, each activation letter
The corresponding line of number, b-6.b-7.b-8.b-9 represent feedback line.
Implementation
The brain of people has thousands of neuron, and some neurons are fed back by exporting, to adjust neuron oneself, and
Next layer of neuron can be directly passed to, because this is a kind of common scenario, therefore the present invention is also according to such case, wound
A kind of artificial neuron is made, polygamma function, which correspondingly exports feedback function artificial neuron, to be by input, artificial neuron, select end
Device, output end, feedback line composition, input such as the input of neuron, receive upper level artificial neuron input or
By the input of other equipment, the effect of artificial neuron is added up after value and multiplied by weight input, if cumulative
Value is less than threshold values, then artificial neuron would not be activated, without any reaction, if cumulative value is more than threshold values, then
Artificial neuron is activated, and cumulative value is just passed to activation primitive collection, and all activation primitives are swashed in activation primitive collection
Living, each activation primitive exports oneself operation result, and activation primitive collection connects with end-apparatus is selected, and how many activation primitive just has
How many connections, by the setting to selecting end-apparatus, it is feedback that can set some activation primitives, and some activation primitives are direct
Next layer of artificial neuron is passed to, some activation primitives therein can also be set, feedback and directly transmission exist simultaneously,
Such artificial neuron and the artificial neuron of other functions are networked, form an artificial brain, it is possible to reach imitation
The function of human brain, it is the form that feedback function is correspondingly exported using polygamma function, therefore due to the artificial neuron of the present invention
More functions can be realized, fewer artificial neuron of the invention can be used, reach sufficiently complex network function.
Claims (1)
1. a kind of polygamma function correspondingly exports the design method of feedback function artificial neuron, it is characterized in that：Polygamma function is corresponding to be exported
Feedback function artificial neuron is by input, artificial neuron, selects end-apparatus, output end, feedback line composition, input as
The input of neuron, receive the input of upper level artificial neuron or the input by other equipment, the effect of artificial neuron
It is to be added up after value and multiplied by weight input, if cumulative value is less than threshold values, then artificial neuron would not be by
Activation, without any reaction, if cumulative value is more than threshold values, then artificial neuron is activated, just cumulative value transmission
Give activation primitive collection, all activation primitives are activated in activation primitive collection, and each activation primitive exports oneself operation result,
Activation primitive collection connects with end-apparatus is selected, and how many activation primitive connects with regard to how many bar, can be with by the setting to selecting end-apparatus
It is feedback to set some activation primitives, and some activation primitives are directly passed to next layer of artificial neuron, can also set
Some activation primitives therein are put, feedback and directly transmission exist simultaneously, and wherein artificial neuron is using following design, and it is by 2
Part forms, and 1 is accumulator, and its effect is exactly that input is added up, if reach threshold values, if it exceeds threshold values, just
This cumulative value passes to activation primitive collection, and all activation primitives are activated simultaneously, and 2 activation primitive collection have multiple activation letters
Array is into once cumulative value is more than threshold values, these activation primitives are activated entirely, are passed to by respective circuit and select end-apparatus, selected
End-apparatus is using following design, and it is connected with activation primitive collection, and how many activation primitive is selecting end with regard to how many bar connection line
That circuit is designed in device in advance to connect with those ports, for example, activation primitive f (x1) it can be exported from o-1.o-2, it is anti-
I-1.1-2 is fed to, f (x2) is exported by i-4.i-5, and f (x3) is exported by i-6.i-7.i-8.i-9, and f (x4) is by i-11.i-9.i-
10 outputs, i-11 are directly transferred to next layer of neuron by feedback i-9.i-10.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710884964.0A CN107609636A (en) | 2017-09-26 | 2017-09-26 | A kind of polygamma function correspondingly exports the design method of feedback function artificial neuron |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710884964.0A CN107609636A (en) | 2017-09-26 | 2017-09-26 | A kind of polygamma function correspondingly exports the design method of feedback function artificial neuron |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107609636A true CN107609636A (en) | 2018-01-19 |
Family
ID=61058487
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710884964.0A Pending CN107609636A (en) | 2017-09-26 | 2017-09-26 | A kind of polygamma function correspondingly exports the design method of feedback function artificial neuron |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107609636A (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103455843A (en) * | 2013-08-16 | 2013-12-18 | 华中科技大学 | Feedback artificial neural network training method and feedback artificial neural network calculating system |
| CN106126481A (en) * | 2016-06-29 | 2016-11-16 | 华为技术有限公司 | A kind of computing engines and electronic equipment |
-
2017
- 2017-09-26 CN CN201710884964.0A patent/CN107609636A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103455843A (en) * | 2013-08-16 | 2013-12-18 | 华中科技大学 | Feedback artificial neural network training method and feedback artificial neural network calculating system |
| CN106126481A (en) * | 2016-06-29 | 2016-11-16 | 华为技术有限公司 | A kind of computing engines and electronic equipment |
Non-Patent Citations (4)
| Title |
|---|
| SAHIL ABROL ET AL.: "Implementation of Single Artificial Neuron Using various Activation Functions and XOR Gate on FPGA chip", 《IMPLEMENTATION OF SINGLE ARTIFICIAL NEURON USING VARIOUS ACTIVATION FUNCTIONS AND XOR GATE ON FPGA CHIP》 * |
| 姚茂群 等: "多阈值神经元电路设计及在多值逻辑中的应用", 《计算机学报》 * |
| 陈允平 等: "《人工神经网络原理及其应用》", 31 August 2002 * |
| 韦一 等: "基于多阈值神经元的D型触发器设计", 《浙江大学学报(理学版)》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Parsopoulos et al. | Objective function" stretching" to alleviate convergence to local minima | |
| KR20160076533A (en) | Methods and apparatus for tagging classes using supervised learning | |
| CN107169566A (en) | Dynamic neural network model training method and device | |
| DE69124231T2 (en) | ARTIFICIAL NEURONAL PLANT | |
| CN107220707A (en) | Dynamic neural network model training method and device based on 2-D data | |
| CN105373829B (en) | A kind of full Connection Neural Network structure | |
| CN107516130A (en) | A kind of more threshold values polygamma functions select the design method of end output artificial neuron | |
| CN107609636A (en) | A kind of polygamma function correspondingly exports the design method of feedback function artificial neuron | |
| CN107563505A (en) | A kind of design method of external control implantation feedback artificial neuron | |
| CN107886163A (en) | Single-object problem optimization method and device based on AGN and CNN | |
| CN107895170A (en) | A kind of Dropout regularization methods based on activation value sensitiveness | |
| CN106682651A (en) | Monitoring method for cable insulation leather damage based on neural network | |
| CN107480779A (en) | Design method that is a kind of while exporting polymorphic function artificial neuron | |
| DE60218447T2 (en) | Method for processing test patterns for an integrated circuit | |
| CN107590536A (en) | A kind of graded potential formula selects the design method of end feedback artificial neuron | |
| CN107563501A (en) | A kind of implantable activation primitive selects the design method of end output artificial neuron | |
| CN107657315A (en) | A kind of graded potential formula selects the design method of end output artificial neuron | |
| CN107563506A (en) | A kind of voltage-frequency formula selects the design method that frequency exports artificial neuron | |
| CN107491810A (en) | A kind of design method of more threshold values feedback artificial neurons | |
| CN107633300A (en) | A kind of design method of graded potential formula artificial neuron | |
| CN107480780A (en) | A kind of more threshold values polygamma functions select the design method of output artificial neuron more | |
| CN107578097A (en) | A kind of design method of more threshold values polygamma function feedback artificial neurons | |
| CN107563504A (en) | A kind of double design methods for selecting end output artificial neuron of implantable activation primitive | |
| CN107563503A (en) | A kind of codified selects the design method that threshold values selects function artificial neuron | |
| CN107609640A (en) | A kind of threshold values selects the design method of end graded potential formula artificial neuron |
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: 20180119 |
|
| RJ01 | Rejection of invention patent application after publication |