CN103679265A - MBUN (multi-characteristic bionic unified neuron) model - Google Patents
MBUN (multi-characteristic bionic unified neuron) model Download PDFInfo
- Publication number
- CN103679265A CN103679265A CN201310588813.2A CN201310588813A CN103679265A CN 103679265 A CN103679265 A CN 103679265A CN 201310588813 A CN201310588813 A CN 201310588813A CN 103679265 A CN103679265 A CN 103679265A
- Authority
- CN
- China
- Prior art keywords
- neuron
- function
- characteristic
- represent
- time
- 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
- Medicines That Contain Protein Lipid Enzymes And Other Medicines (AREA)
- Pharmaceuticals Containing Other Organic And Inorganic Compounds (AREA)
- Medicines Containing Material From Animals Or Micro-Organisms (AREA)
Abstract
The invention relates to an MBUN (multi-characteristic bionic unified neuron) model. The MBUN (multi-characteristic bionic unified neuron) model is an artificial neuron unified frame model which can more comprehensively indicate characteristics of biological neuron. Dozens of neuron models are uniformly described, 12 basic characteristics of biological neuron that are generally recognized by biological medicine and engineering science at present are summarized, and accordingly the MBUN model is provided.
Description
The present invention is for making unified description to tens kinds of neuron models, 12 large fundamental characteristics of the universally recognized biological neuron of current biomedical Jie He engineering science circle are summarized in summary, and have proposed based on this kind of artificial neuron's Unified frame model that more comprehensively reflects biological neuron characteristic--the bionical unified neurons of many characteristics (MBUN) model.
For solving the problems of the technologies described above, technical scheme of the present invention is:
Comprehensive all kinds of document and achievement in research, the universally recognized biological neuron of current biomedical Jie He engineering science circle has following 12 large fundamental characteristics, and has provided respectively mathematical notation:
(1) spatial integration effect
On neural network structure, a large amount of different neuronic Nerve Terminals In The Human Skins can arrive same neuronic dendron and form a large amount of cynapses.While having signal to transmit on these dendrons, these signals are all stacking up in cell body transport process simultaneously.This neuron can be integrated the input message of originating on different dendrons, is called spatial integration effect.In MP model, integrate effect and represent by weighted sum function.But integrate effect, whether necessarily with addition, describe, whether have other rules, biologically there is no at present clear and definite conclusion.In the present invention, with spatial integration function S (#), represent this specific character.
(2) time is integrated effect
For the information from same cynapse, the information that neuron can import into different time is integrated.Continued stimulus within a certain period of time will produce Overlay, is called time integration effect.With the time, integrate function T (#) herein and represent this specific character.
(3) threshold property
When neuron is accepted from other neuronic information, film potential when starting by time continuous gradation.When space, the temporal summation of irriate is greater than a certain amount ofly, film potential changes while exceeding certain definite value, and neuron produces the pulse that sudden change is risen, and pulse edge aixs cylinder is to next stage transmission.The characteristic that this film potential of neuron just produces pulse transmission up to certain threshold value is called threshold property.In common MP model, visual threshold value is certain certain value, is directly taken as H.Yet Physiologic Studies points out that in neural network, each neuronic threshold value not immobilizes, not only different neurons can be got different values, but also may change along with the time.As just there being what is called/high threshold cell 0 in the network structure at cerebellum pon place.Use threshold function table TH (#) to represent this variable characteristic herein.
(4) refractory period
After electric pulse just discharges, even if very strong signal arrives, neuron can be not excited yet, is called absolute refractory period during this.After absolute refractory period finishes, within a period of time, excited threshold value is higher than common, makes neuron be difficult to excitement, is called relative refractory period during this.The threshold value having uprised can turn back to normal value gradually.Nearly 1ms of absolute refractory period, and nearly several milliseconds of relative refractory period.Herein with not answering function G (#) to represent this specific character.
(5) tired (adaptability) characteristic
Neuron repeatedly can stimulate input to adapt to some extent lasting stimulation afterwards to external world, and its threshold value increases gradually, causes neuron to be difficult to excitement, needs to change stimulus intensity and just can obtain same response, and this is called neuronic fatigue properties.Its mathematical notation can be adjusted (as adopted time-varying function) in threshold value, weight and output characteristics.Use tired function P (#) to represent herein.
(6) excited and inhibition
A neuron has excited and suppresses two states, and is characterized by potential difference (PD) different inside and outside cell membrane.At holddown, between cell membrane is inside and outside, there is interior negative outer positive potential difference (PD); In excitatory state, in producing, just outer negative opposite potential is poor.The existing excitability of connection between neuron connects, and also has inhibition to connect.In the present invention, use on the occasion of weight and represent that excitability connects, with negative value weight, represent that inhibition connects; To represent on the occasion of input the excited input message that neuron is accepted, with negative value input, represent the inhibition input message that neuron is accepted.
(7) time-delay characteristics
Information exchange is crossed cynapse transmission, neuron a time delay between receiving information to and responding, be generally 0.5 ~ 1ms.The present invention represents with delay function D (#).
(8) output exciting characteristic
Neuron is made action to the various stimulations of input after integrating, and produces pulse output, and this effect is referred to as output and activates or exciting characteristic.From the angle of nervous physiology, the neuron output being excited by activation value has extremely complicated relation, is difficult to a kind of well-determined function representation.Therefore, current all kinds of neuronic output excitation function has various ways, has represented separately a side of neuron output characteristics.Variable excitation function F (#) represents for the present invention, and it can be certain Certain function summary of determining, can be also in time or the family of functions that regulates parameter to change.
(9) conduction attenuation characteristic
Neuron produces action potential under activation, and this current potential passes on another neuron along aixs cylinder with certain speed, and its transfer rate is relevant with diameter and the structure of aixs cylinder.Cable equation is generally obeyed in the transmission of current potential, with attenuated form transmission, and conduction attenuation characteristic that Here it is.The present invention represents with attenuation function A (#).
(10) plasticity that cynapse connects
Connection between cynapse is along with the power of signal and neuronic excitement degree and change.With variable connection weight vector W, represent this specific character herein, and with learning algorithm, carry out renewal and the adjustment of weights.
(11) digital/analog signal conversion characteristic
The information of transmitting on neuron axon is the discrete electrical pulse signal of constant amplitude, constant-breadth, coding, is a digital quantity.But in the release of neurotransmitter and dendron, the variation of film potential is continuous in cynapse, this explanation cynapse has digital-analog signal translation function.When neuronic dendron film potential exceeds certain threshold values, become again electric pulse mode and sent out by aixs cylinder, this procedure declaration neuron has analog to digital function switching signal.With analog to digital conversion function AD (#) and digital-to-analog conversion function DA (#), represent respectively.
(12) new discovery characteristic
The neuron behavior of the current popular various neural network models that contrast Getting, PDP group and pertinent literature were once listed, can find out, they all do not exceed 11 above characteristics.But science is constantly deeply and development, considers and also may find more key property future, represents following newfound characteristic with characteristic function N (#), to promote the versatility of framework and perspective.
Compared with prior art, the invention has the beneficial effects as follows:
(1), the network of this model-composing being when being applicable to solve some problem or solving some problem, can obtain satisfactory result;
(2), model is simple, similar to the neuron of biological nervous system, a lot of characteristics are all described fully;
(3), the capacity of network is relevant with network size, while solving more complicated problem, need adopt larger network, the design of this network becomes training very easily to realize.
Claims (1)
1. more than one kind, characteristic bionic unified neuron model is divided into following components:
Comprehensive all kinds of document and achievement in research, the universally recognized biological neuron of current biomedical Jie He engineering science circle has following 12 large fundamental characteristics, and has provided respectively mathematical notation:
(1) spatial integration effect
On neural network structure, a large amount of different neuronic Nerve Terminals In The Human Skins can arrive same neuronic dendron and form a large amount of cynapses; While having signal to transmit on these dendrons, these signals are all stacking up in cell body transport process simultaneously; This neuron can be integrated the input message of originating on different dendrons, is called spatial integration effect; In MP model, integrate effect and represent by weighted sum function; But integrate effect, whether necessarily with addition, describe, whether have other rules, biologically there is no at present clear and definite conclusion; In the present invention, with spatial integration function S (#), represent this specific character;
(2) time is integrated effect
For the information from same cynapse, the information that neuron can import into different time is integrated; Continued stimulus within a certain period of time will produce Overlay, is called time integration effect; With the time, integrate function T (#) herein and represent this specific character;
(3) threshold property
When neuron is accepted from other neuronic information, film potential when starting by time continuous gradation; When space, the temporal summation of irriate is greater than a certain amount ofly, film potential changes while exceeding certain definite value, and neuron produces the pulse that sudden change is risen, and pulse edge aixs cylinder is to next stage transmission; The characteristic that this film potential of neuron just produces pulse transmission up to certain threshold value is called threshold property; In common MP model, visual threshold value is certain certain value, is directly taken as H; Yet Physiologic Studies points out that in neural network, each neuronic threshold value not immobilizes, not only different neurons can be got different values, but also may change along with the time; As just there being what is called/high threshold cell 0 in the network structure at cerebellum pon place; Use threshold function table TH (#) to represent this variable characteristic herein;
(4) refractory period
After electric pulse just discharges, even if very strong signal arrives, neuron can be not excited yet, is called absolute refractory period during this; After absolute refractory period finishes, within a period of time, excited threshold value is higher than common, makes neuron be difficult to excitement, is called relative refractory period during this; The threshold value having uprised can turn back to normal value gradually; Nearly 1ms of absolute refractory period, and nearly several milliseconds of relative refractory period; Herein with not answering function G (#) to represent this specific character;
(5) tired (adaptability) characteristic
Neuron repeatedly can stimulate input to adapt to some extent lasting stimulation afterwards to external world, and its threshold value increases gradually, causes neuron to be difficult to excitement, needs to change stimulus intensity and just can obtain same response, and this is called neuronic fatigue properties; Its mathematical notation can be adjusted (as adopted time-varying function) in threshold value, weight and output characteristics; Use tired function P (#) to represent herein;
(6) excited and inhibition
A neuron has excited and suppresses two states, and is characterized by potential difference (PD) different inside and outside cell membrane; At holddown, between cell membrane is inside and outside, there is interior negative outer positive potential difference (PD); In excitatory state, in producing, just outer negative opposite potential is poor; The existing excitability of connection between neuron connects, and also has inhibition to connect; In the present invention, use on the occasion of weight and represent that excitability connects, with negative value weight, represent that inhibition connects; To represent on the occasion of input the excited input message that neuron is accepted, with negative value input, represent the inhibition input message that neuron is accepted;
(7) time-delay characteristics
Information exchange is crossed cynapse transmission, neuron a time delay between receiving information to and responding, be generally 0.5 ~ 1ms; The present invention represents with delay function D (#);
(8) output exciting characteristic
Neuron is made action to the various stimulations of input after integrating, and produces pulse output, and this effect is referred to as output and activates or exciting characteristic; From the angle of nervous physiology, the neuron output being excited by activation value has extremely complicated relation, is difficult to a kind of well-determined function representation; Therefore, current all kinds of neuronic output excitation function has various ways, has represented separately a side of neuron output characteristics; Variable excitation function F (#) represents for the present invention, and it can be certain Certain function summary of determining, can be also in time or the family of functions that regulates parameter to change;
(9) conduction attenuation characteristic
Neuron produces action potential under activation, and this current potential passes on another neuron along aixs cylinder with certain speed, and its transfer rate is relevant with diameter and the structure of aixs cylinder; Cable equation is generally obeyed in the transmission of current potential, with attenuated form transmission, and conduction attenuation characteristic that Here it is; The present invention represents with attenuation function A (#);
(10) plasticity that cynapse connects
Connection between cynapse is along with the power of signal and neuronic excitement degree and change; With variable connection weight vector W, represent this specific character herein, and with learning algorithm, carry out renewal and the adjustment of weights;
(11) digital/analog signal conversion characteristic
The information of transmitting on neuron axon is the discrete electrical pulse signal of constant amplitude, constant-breadth, coding, is a digital quantity; But in the release of neurotransmitter and dendron, the variation of film potential is continuous in cynapse, this explanation cynapse has digital-analog signal translation function; When neuronic dendron film potential exceeds certain threshold values, become again electric pulse mode and sent out by aixs cylinder, this procedure declaration neuron has analog to digital function switching signal; With analog to digital conversion function AD (#) and digital-to-analog conversion function DA (#), represent respectively;
(12) new discovery characteristic
The neuron behavior of the current popular various neural network models that contrast Getting, PDP group and pertinent literature were once listed, can find out, they all do not exceed 11 above characteristics; But science is constantly deeply and development, considers and also may find more key property future, represents following newfound characteristic with characteristic function N (#), to promote the versatility of framework and perspective.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310588813.2A CN103679265A (en) | 2013-11-21 | 2013-11-21 | MBUN (multi-characteristic bionic unified neuron) model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310588813.2A CN103679265A (en) | 2013-11-21 | 2013-11-21 | MBUN (multi-characteristic bionic unified neuron) model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103679265A true CN103679265A (en) | 2014-03-26 |
Family
ID=50316749
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310588813.2A Pending CN103679265A (en) | 2013-11-21 | 2013-11-21 | MBUN (multi-characteristic bionic unified neuron) model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103679265A (en) |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106897769A (en) * | 2017-01-20 | 2017-06-27 | 清华大学 | The neuronal messages processing method and system of window are drawn with depth time |
| CN107527089A (en) * | 2017-10-10 | 2017-12-29 | 胡明建 | It is a kind of that the design method for pressing counting to be used as identification artificial neuron is adopted using the time |
| CN107545304A (en) * | 2017-09-16 | 2018-01-05 | 胡明建 | A kind of design method for changing activation primitive artificial neuron according to network demand |
| CN107563506A (en) * | 2017-09-27 | 2018-01-09 | 胡明建 | A kind of voltage-frequency formula selects the design method that frequency exports artificial neuron |
| CN107563503A (en) * | 2017-09-14 | 2018-01-09 | 胡明建 | A kind of codified selects the design method that threshold values selects function artificial neuron |
| CN107578096A (en) * | 2017-09-21 | 2018-01-12 | 胡明建 | A kind of voltage-frequency formula selects the design method of end artificial neuron |
| CN107578097A (en) * | 2017-09-25 | 2018-01-12 | 胡明建 | A kind of design method of more threshold values polygamma function feedback artificial neurons |
| CN107609640A (en) * | 2017-10-01 | 2018-01-19 | 胡明建 | A kind of threshold values selects the design method of end graded potential formula artificial neuron |
| CN107633299A (en) * | 2017-09-26 | 2018-01-26 | 胡明建 | A kind of design method of voltage-frequency formula artificial neuron |
| CN111401540A (en) * | 2020-03-09 | 2020-07-10 | 北京航空航天大学 | Neuron model construction method and neuron model |
| CN112101535A (en) * | 2020-08-21 | 2020-12-18 | 中国科学院深圳先进技术研究院 | Signal processing method of pulse neuron and related device |
-
2013
- 2013-11-21 CN CN201310588813.2A patent/CN103679265A/en active Pending
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018133569A1 (en) * | 2017-01-20 | 2018-07-26 | 清华大学 | Neuron information processing method and system having deep time windowing |
| CN106897769A (en) * | 2017-01-20 | 2017-06-27 | 清华大学 | The neuronal messages processing method and system of window are drawn with depth time |
| CN107563503A (en) * | 2017-09-14 | 2018-01-09 | 胡明建 | A kind of codified selects the design method that threshold values selects function artificial neuron |
| CN107545304A (en) * | 2017-09-16 | 2018-01-05 | 胡明建 | A kind of design method for changing activation primitive artificial neuron according to network demand |
| CN107578096A (en) * | 2017-09-21 | 2018-01-12 | 胡明建 | A kind of voltage-frequency formula selects the design method of end artificial neuron |
| CN107578097A (en) * | 2017-09-25 | 2018-01-12 | 胡明建 | A kind of design method of more threshold values polygamma function feedback artificial neurons |
| CN107633299A (en) * | 2017-09-26 | 2018-01-26 | 胡明建 | A kind of design method of voltage-frequency formula artificial neuron |
| CN107563506A (en) * | 2017-09-27 | 2018-01-09 | 胡明建 | A kind of voltage-frequency formula selects the design method that frequency exports artificial neuron |
| CN107609640A (en) * | 2017-10-01 | 2018-01-19 | 胡明建 | A kind of threshold values selects the design method of end graded potential formula artificial neuron |
| CN107527089A (en) * | 2017-10-10 | 2017-12-29 | 胡明建 | It is a kind of that the design method for pressing counting to be used as identification artificial neuron is adopted using the time |
| CN111401540A (en) * | 2020-03-09 | 2020-07-10 | 北京航空航天大学 | Neuron model construction method and neuron model |
| CN112101535A (en) * | 2020-08-21 | 2020-12-18 | 中国科学院深圳先进技术研究院 | Signal processing method of pulse neuron and related device |
| CN112101535B (en) * | 2020-08-21 | 2024-04-09 | 深圳微灵医疗科技有限公司 | Signal processing method of impulse neuron and related device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103679265A (en) | MBUN (multi-characteristic bionic unified neuron) model | |
| EP4235671A3 (en) | System for converting neuron current into neuron current-based time pulses in an analog neural memory in a deep learning artificial neural network | |
| CN104899640A (en) | Simulator and method for neural network | |
| CN104335224A (en) | Neuron circuit and method | |
| Carlson et al. | Neural innovations and the diversification of African weakly electric fishes | |
| Bakkum et al. | MEART: the semi-living artist | |
| Ambroise et al. | Biomimetic neural network for modifying biological dynamics during hybrid experiments | |
| Clark et al. | Music biology: all this useful beauty | |
| Benjamin et al. | Neurogrid simulates cortical cell-types, active dendrites, and top-down attention | |
| CN117275568A (en) | Primary auditory cortex neuron cell release rate curve simulation method and device | |
| CN205268385U (en) | Artifical cochlea simulation system | |
| Van Ooyen et al. | Activity-dependent neurite outgrowth and neural network development | |
| Niven et al. | Reuse of identified neurons in multiple neural circuits | |
| Ricart | Neuromodulatory mechanisms in neural networks and their influence on interstimulus interval effects in Pavlovian conditioning | |
| CN107480779A (en) | Design method that is a kind of while exporting polymorphic function artificial neuron | |
| Weisman et al. | A comparative analysis of auditory perception in humans and songbirds: A modular approach | |
| CN107563506A (en) | A kind of voltage-frequency formula selects the design method that frequency exports artificial neuron | |
| Guo et al. | The unique characteristics of on and off retinal ganglion cells: a modeling study | |
| Sadananda et al. | If you fire together, you wire together | |
| CN107545304A (en) | A kind of design method for changing activation primitive artificial neuron according to network demand | |
| CN107480776A (en) | A kind of design method of briquettability artificial neuron | |
| CN107516130A (en) | A kind of more threshold values polygamma functions select the design method of end output artificial neuron | |
| Lyuty et al. | Nietzsche’s philosophy as a creation of concepts (XVI Kyiv-Mohyla Seminar on the History of Philosophy) | |
| Morse et al. | Enhanced cochlear implant coding using multiplicative noise | |
| Snihur | PERCEPTION AND ANALYZATION OF EMOTIONAL BODY LANGUAGE |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140326 |