CN101860357B - Method for weight control and information integration by utilizing time encoding - Google Patents
Method for weight control and information integration by utilizing time encoding Download PDFInfo
- Publication number
- CN101860357B CN101860357B CN2010102015893A CN201010201589A CN101860357B CN 101860357 B CN101860357 B CN 101860357B CN 2010102015893 A CN2010102015893 A CN 2010102015893A CN 201010201589 A CN201010201589 A CN 201010201589A CN 101860357 B CN101860357 B CN 101860357B
- Authority
- CN
- China
- Prior art keywords
- circuit
- neuron
- weight
- dendritic
- 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.)
- Expired - Fee Related
Links
Images
Abstract
The invention discloses a method for weight control and information integration by utilizing time encoding, which comprises the following steps: utilizing a dendritic circuit to control the spatial distribution of weight; utilizing the dendritic circuit to realize the corresponding relationship between characteristic parameters and the weight; and utilizing the dendritic circuit to control the time distribution of the weight. The invention utilizes the organic combination of input signals distributed in time and space to realize the dynamic distribution control of the weight, utilizes the noise margin characteristics of a CMOS inverter to change the output pulse frequency of a neuron circuit, so that the digital signals increase simulated change parameter. The invention helps to understand and simulate information processing way of biological neurons.
Description
Technical field
The present invention relates to the neuron circuit technical field, relate in particular to a kind of time encoding control weight and information integrated method utilized.
Background technology
Biological neuron has the function of permissive cell external stimulus signal, can receive the nerve impulse from other nerve cell aixs cylinder tips through cynapse.Nerve impulse is derived from action potential, and action potential is formed at axon hillock, is the coefficient results of all cynapses.Cynapse is to realize through the integration function of biological neuron for the effect of nerve impulse; This information integrated function is divided into time integration function and spatial integration function, respectively for carrying out integrated treatment from neuronic action potential of the other biological in different time and space or nerve impulse.The time integration means that biological neuron has the ability of carrying out integrated treatment for the neuronic nerve impulse from the other biological of same cynapse different time.Spatial integration means that biological neuron has the ability of carrying out integrated treatment for from the nerve impulse of other biological neuron of same time of not homo-synapse.The parallel processing that information is simple lacks hierarchical organization, and will cause " stack disaster ".The serial process that the coding outside stimulus is stressed information is provided in the neuron pulse, has the hierarchical organization characteristic.
The nerve impulse of biological neuron is a kind of action pulse signal.Action pulse has on the time and the variation on the amplitude.The action pulse sequence generally all is a time-continuous signal, and its information transmission has " simulation " signal characteristic.Neuron discharge behavior is to describe neuronic activity through the change of the frequency of discharging in the time windows (spike potential density), i.e. " frequency coding " traditionally.Yet, be not to have strict one-to-one relationship between the discharge sequence of stimulation incident and neuron reaction, same stimulation can induce different reaction discharge sequences on same neuron.Adopt the average method of probability just the activity of single neuron under spread effect to be described with the discharge number mean value in the unit interval, this is the statistics to the discharge time of the discharge sequence of being sampled.In fact, the action potential that produces continuously in the same neuron is not a completely random.Any physical structure all has certain inertia, just memory.Really, for most of neurons, its degree of depolarization determined overall spike potential granting and provided probability, and the variation of the change of film potential and granting rate all needs the time, can not accomplish immediately.During this period of time, more or less all can show interdependency between the spike in succession, its degree of dependence depends on open ion channel number.Be not that so strong repeatability is arranged the discharge time of nerve impulse, and time dependent discharge frequency is the sign of discharge sequence general characteristic.The distribution obeys index distribution of the granting rate of nerve impulse, the information that exponential distribution transmits is maximum.For the spike potential sequence, if its incidence has changed, the duration of individual event is still constant, and the number that incident takes place in it and unit interval so is irrelevant.Therefore, the description that " frequency coding " carries out discharge is also being implied and can described neuron activity property through the discharge frequency that random time changes, and this has just proposed the notion of " time encoding ".What " time encoding " showed is the time interval information of spike in the instant time.If the time interval is short, real-time frequency is just big; Otherwise if the time interval is long, real-time frequency is just little.Time interval distribution possibility between the discharge pulse can be investigated through the joint probability of two pulse generations.What " frequency coding " showed is discharge sequence general characteristic, and what " time encoding " showed is the instant characteristic of discharge sequence.The synchronized oscillation phenomenon shows that " time encoding " has the impulse phase coherence property.The phase difference that the diverse location cell is provided roughly and the distance between the position field proportional.The position the closer to, differ more little.Like this, the order of locus just can be coded by neuronic action potential order.In " time encoding "; A neuron then is to send an instruction in the big inventory by the output mode of many time of comforming precision encoding is made in the variation of the single characteristics of stimulation, and " time encoding " will contribute to the parallel integration problem with distributed information processing.
A basic problem in the Neuscience is how the information of same neuron activity relevant with behavior to be mapped, and just how the understanding of things and course of reaction are encoded in brain, express and process to external world.How to understand neural coded system and just become key of dealing with problems and the approach of extricating oneself from a plight.Only regard neuron the notion of average discharge integrator as, receive the challenge of increasing new experimental fact, the neural coding problem is aroused the enthusiasm that the scientist of different field participates in exploratory development.
Neuron is as neural basic function unit, and can experience stimulates and the conduction excitement.The electricity Physiological Experiment has shown that neuron has the non-linear of height, in Different Ca
2+The extraneous galvanic current stimulation of ion concentration or different amplitudes can show abundant discharge mode down, for example the cutting edge of a knife or a sword discharge and bunch discharge of the discharge of the cutting edge of a knife or a sword in cycle and a bunch discharge, chaos.In single neuron coding theory, people mainly investigate the relation between neuronic discharge activities and the environmental stimuli input, seldom are concerned about the influence that other neuron activity brings.At present, increasing electrophysiology evidence shows that neuronic activity depends in the loop or the first activity of related neural in the colony to a great extent.In whole nervous system, the transmission of nerve impulse often will be accomplished through the mode of coupling by plural at least neuron.Scientists has been observed neuronic simultaneous shot pattern in the visual cortex of the cat of anaesthetizing.The relevant character of consciousness shows with the mode of synchronous granting in the video, shows as synchronous granting with the corresponding neuron of same object characteristic, then provides with uncorrelated mode with the corresponding neuron of different objects.In the signal transduction process in neuronal circuit; Each part of information is not to be the individual cells absolute coding; But encode through the activity of cell colony, this process is called " colony's coding ", and it is to constitute the basic process that nerve information transmits.A neuron can not be accomplished the time encoding to the continuous discharge pulse, and neuron colony can reflect common cynapse stream in a synchronous manner.So neuron is the common completion of neuron colony to information processing processing, the motor pattern of neuron colony is very important to the transmission of information.
The neuron models of pulse coupling help to disclose and explain viewed some neural synchronia in the test.Interneuronal coupled characteristic is to be caused by input conductance conduction of current effect between this neuron presynaptic and adjacent neurons dendron postsynaptic; And the control that these input conductances are originally experienced the neuron pulse voltage makes this neural metamessage lead this voltage-controlled characteristic by the cynapse electricity and is sent to adjacent neurons.The equivalent conductance of ion channel influences each other on the synapse, and has neuron voltage dependence (voltage-controlled) characteristic.Neuron is in order to carry out different functions, and its coupled modes also show various types of attachment, and what different connection form should be to coupled neural unit has different effects synchronously.The synchronizing process of coupled system is very complicated, and also there is the result of time lag just in the speed system interaction process in this.The stationary problem of coupled neural metasystem is the key of its information processing.In the time lag ubiquity ecosystem, the appearance of time lag has increased interneuronal synchronous effect just.Reduce system speed adaptively through time lag, the neuron of two weak couplings is regulated initial condition, reaches optimum stiffness of coupling and obtains synchronous.Different types of attachment has different effects to coupled synchronization, and the stiffness of coupling that the chain type connection needs is maximum, secondly be that ring type connects, and overall situation coupling needs minimum stiffness of coupling just can realize fully synchronously.In the Hopfield network, use " Hebb rule " to regulate the connection weight between the neuron, if two unit have identical output, the weight that interconnects between them is energized; If they have opposite output, then weight is weakened.After the output of constantly regulating each unit, what network disclosed is the stable contact of unit activity.Finally it will only with " memory " of its storage approaching information recover this memory from some effectively.The initial connection of neural net is by genetic mechanism control, and neuron unavoidably constantly changes in time delay and processing procedure, can affirm that almost biological evolution just is based upon on these changes and the time delay, and therefrom benefit.True neuron life period inevitably postpones and the continuing to optimize of processing procedure.
Summary of the invention
The technical problem that (one) will solve
In view of this, in order " hierarchical organization " characteristic to be provided to the parallel input signal of neuron, main purpose of the present invention is to provide a kind of time encoding control weight and information integrated method utilized.
(2) technical scheme
For achieving the above object, the invention provides a kind of time encoding control weight and information integrated method utilized, this method comprises:
Utilize the cynapse of dendritic circuit to realize spatial distribution of weight control, the cynapse of this dendritic circuit is made up of the PMOS transistor that breadth length ratio has one group of parallel connection of multiple relation;
Utilize dendritic circuit to realize the corresponding relation of characteristic parameter and weights; And
Multichannel timing control signal through the input dendritic circuit carries out gating to the PMOS transistor of parallel connection, changes the noise margin of input stage inverter, and then realizes the control that distributes of weights time;
Wherein, the said corresponding relation that utilizes dendritic circuit to realize characteristic parameter and weights is to constitute the dendritic circuit with CMOS inverter structure by one group of parallelly connected PMOS transistor and nmos pass transistor series connection; This dendritic circuit characteristic parameter is the transistorized breadth length ratio of PMOS of parallel connection, and weights are the transistorized current strength of the PMOS that flows through; The transistorized current strength of the PMOS that flows through depends on the spatial distribution of different breadth length ratios, makes this dendritic circuit characteristic parameter and weights form corresponding relation;
The time coding signal of multiterminal input selects corresponding PMOS transistor to constitute various combination, and corresponding effectively breadth length ratio changes, thereby changes the transistorized electric current spatial distribution of PMOS of flowing through, and makes time encoding and weights form corresponding relation.
In the such scheme, said multichannel timing control signal is made up of outside stimulus signal and internal feedback signal, and this multichannel timing control signal is imported into the cynapse end of neuron dendritic circuit.
In the such scheme, said internal feedback signal comes from the dynamic memory delay unit that is connected with the first circuit output end of pulse coupled neural.
(3) beneficial effect
Can find out that from technique scheme the present invention has following beneficial effect:
1, the present invention utilizes the organic assembling that input signal distributes in the time and space, has realized the DYNAMIC DISTRIBUTION control of weight.
2, the present invention utilizes the noise margin characteristic of CMOS inverter to change the output pulse frequency of neuron circuit, makes digital signal increase the analog variation parameter.
3, the method for the time encoding of neuron signal provided by the invention and integration helps to understand and the simulation information processing way of biological neurons.
Description of drawings
Fig. 1 is time encoding control weight and the information integrated method flow diagram of utilizing provided by the invention;
Fig. 2 is the structure chart that comprises the Hopfield network of n neuron (Cell) and respective stored and delay cell (M);
Fig. 3 is pulse coupling CMOS neuron (Cell) circuit diagram;
Fig. 4 is dynamic memory delay cell (M) circuit diagram;
Fig. 5 is the PMOS three end input stage circuit domains with different breadth length ratios.
Embodiment
For making the object of the invention, technical scheme and advantage clearer, below in conjunction with specific embodiment, and with reference to accompanying drawing, to further explain of the present invention.
Weight control with information integrated relate to interneuronally be connected, study and storing process.Artificial neural net thinks to have the large-scale parallel distributed processor that neuron can constitute the storage Heuristics and make it available characteristic.Bonding strength between the intrinsic nerve unit, i.e. the weights of cynapse are used to store the knowledge of obtaining.The learning process of neural net promptly is the process that changes system's weights by certain orderly mode.Biological neurology research shows, when nerve impulse when axilemma reaches axon ends, promptly trigger the current potential gated calcium channel opening on the presynaptic membrane, extracellular Ca
2+It is inboard that ion gets into presynaptic membrane; Trigger the synaptic versicle release neurotransmitters, and corresponding receptors bind with postsynaptic membrane, cause with the chemically-gated channel of acceptor coupling open; Make corresponding ion turnover; Thereby change the distribution situation of postsynaptic membrane both sides ion, excitement or inhibition occur and change, and then influence the activity of postsynaptic neuron.Ca
2+The effect of trigger is played in the inflow of ion, activates some gene and for a long time neuron is modified, and these genes can increase the efficient of mediator, the number of acceptor, even increases the efficient that acceptor is opened ion channel.Molecular biology has been found a kind of hereditary information pass through mechanism, makes neuron possibly be the radical change of gene expression institute.By the nervous activity that study causes, can change the trickle chemical constitution of the ribonucleic acid of those relevant with it inside neurons.
According to above neuron models and electrophysiological characteristics, the present invention imitates Ca
2+The triggering gate action of ion; Utilize the open one group of parallel P type MOS transistor of various signals combination gating at pulse coupled neural unit circuit input end with different breadth length ratios combinations, and with complementation have fixedly that breadth length ratio N type MOS transistor constitutes the input stage inverter structure.The input signal combination is different, and the noise margin of inverter will change, thereby influence the pulse train output frequency of pulse coupling circuit.The different breadth length ratio combinations of input P type MOS transistor, relevant with the weight of the first circuit of pulse coupled neural, can affirm that the noise margin of input stage inverter is big more, the coupling weight of pulse coupled neural unit circuit is big more.
In order to realize the distributed memory storage of weight; The present invention's design connects the dynamic storage cell of a single NMOS pipe and an electric capacity at pulse coupled neural unit circuit output end; The vesicle of the release neurotransmitters of imitation presynaptic membrane; Capacitive charge storage voltage will trigger two-stage inverter delay unit, and the output signal feedback is to the input of adjacent pulse coupled neural unit circuit.According to the Hopfield network model, we can constitute a multiloop reponse system with this neuron circuit with input and output feedback.The quantity of feedback loop equals neuronal quantity.Each neuronic output all is fed back to each other in network neuron, has avoided the self feed back in the network.Can realize that through the pulse signal coupling self adaptation is synchronous between interconnected neuron, improve stiffness of coupling, the realization neuron colony is accomplished information processing jointly.
Based on above-mentioned realization principle, Fig. 1 shows time encoding control weight and the information integrated method flow diagram of utilizing provided by the invention, and this method specifically comprises the steps:
Step 1: utilize dendritic circuit to realize spatial distribution of weight control;
Step 2: utilize dendritic circuit to realize the corresponding relation of characteristic parameter and weights; And
Step 3: utilize dendritic circuit to realize the time distribution control of weights.
Wherein, utilizing dendritic circuit to realize spatial distribution of weight control described in the step 1, is to utilize the cynapse of dendritic circuit to realize spatial distribution of weight control, and the cynapse of this dendritic circuit is made up of the PMOS transistor that breadth length ratio has one group of parallel connection of multiple relation.
Utilize dendritic circuit to realize the corresponding relation of characteristic parameter and weights described in the step 2; Be to connect by PMOS transistor AND gate one nmos pass transistor of one group of parallel connection to constitute input stage CMOS inverter, realize the corresponding relation of noise margin and weights by this input stage CMOS inverter.
Utilize dendritic circuit to realize the time distribution control of weights described in the step 3; Be that multichannel timing control signal through the input dendritic circuit carries out gating to the PMOS transistor of parallel connection; Change the noise margin of input stage inverter, and then realize the distribution control of weights time.This multichannel timing control signal is made up of outside stimulus signal and internal feedback signal, and this multichannel timing control signal is imported into the cynapse end of neuron dendritic circuit.The internal feedback signal comes from the dynamic memory delay unit that is connected with the first circuit output end of pulse coupled neural.
Hopfield network shown in Figure 2 comprises one group of neuron and one group of corresponding unit storage delay, constitutes a multiloop reponse system.Basically, each neuron all is a computing unit, accepts to add the feedback input of input and other nodes simultaneously, also all directly exports to the external world.The quantity of feedback loop equals neuronal quantity, and the output of each neuron Cell all is fed back to each other in network neuron through a storage delay unit M.Consider the convergence and the stability problem of feedback network, do not have self feed back in the network.Can realize that through the pulse signal coupling self adaptation is synchronous between interconnected neuron, improve stiffness of coupling, the realization neuron colony is accomplished information processing jointly.
Pulse coupling CMOS neuron circuit shown in Figure 3, this circuit is connected and composed by dendritic circuit, integration summer and pulse generating circuit three parts successively.The characteristics of this pulse coupled neural unit circuit are that output and input are pulse train bursts, and the device of this circuit is the CMOS transistor.Dendritic circuit is in series through the drain terminal node by one a group of parallelly connected P type MOS transistor and a N type MOS transistor and constitutes cmos circuit, the source end input pulse voltage signal of P type MOS transistor; The integration summer is by a capacitor C
∑Constitute, this electric capacity is connected with the drain terminal node of N type MOS transistor with P type in the dendritic circuit, accumulates weighted current formation trigger voltage signal; Pulse generating circuit forms feedback loop by the CMOS inverter and the dendron cmos circuit of even number series connection, produces pulse train bursts output, and the frequency of output pulse sequence string receives the modulation of input voltage pulse signal.
Dynamic memory delay unit M shown in Figure 4 is by a single NMOS pipe and a capacitor C
wConstitute, the grid of NMOS pipe is connected capacitor C with the first circuit output end of pulse coupled neural
wThe integration stored charge, the distributed memory storage of realization weight, M output signal feedback is to the input of adjacent pulse coupled neural unit circuit.
Embodiment:
Utilize the open one group of parallel P type MOS transistor of various signals combination gating at pulse coupled neural unit circuit input end with different breadth length ratios combinations, and with complementation have fixedly that breadth length ratio N type MOS transistor constitutes the input stage inverter structure.The input signal combination is different, and the noise margin of inverter will change, thereby influence the pulse train output frequency of pulse coupling circuit.The different breadth length ratio combinations of input P type MOS transistor, relevant with the weight of the first circuit of pulse coupled neural, can affirm that the noise margin of input stage inverter is big more, the coupling weight of pulse coupled neural unit circuit is big more.
As shown in Figure 5, the ratio of the wide length (W/L) of a nmos pass transistor of the PMOS transistor AND gate of one group of parallel connection series connection is the electric current weighting ratio in the dendritic circuit, and wherein the breadth length ratio of the PMOS of parallel connection pipe is the binary system relation.Through the sequential combination of input frequency multiplication square wave voltage waveform, obtain different input current weight gating combinations.Along with the increase of time, PMOS passage and quantity that dendron is partly opened change successively, and parallelly connected PMOS manages equivalent grid width and increased progressively by 1 μ m to 7 μ m successively, and general effect is ∑ (W/L)
p/ (W/L)
nMove closer to 3 but, thereby make the noise margin of one-level inverter move closer to the maximum noise tolerance limit, and the summation of the weighted integral of electric current increases gradually, to capacitor C all the time less than 3
∑Charging rate speed, cause the output waveform frequency to increase.For the multiterminal input, when the noise margin of first order inverter was maximum, this moment, output pulse frequency reached the highest.
Above-described specific embodiment; The object of the invention, technical scheme and beneficial effect have been carried out further explain, and institute it should be understood that the above is merely specific embodiment of the present invention; Be not limited to the present invention; All within spirit of the present invention and principle, any modification of being made, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (3)
1. one kind is utilized time encoding control weight and information integrated method, it is characterized in that this method comprises:
Utilize the cynapse of dendritic circuit to realize spatial distribution of weight control, the cynapse of this dendritic circuit is made up of the PMOS transistor that breadth length ratio has one group of parallel connection of multiple relation;
Utilize dendritic circuit to realize the corresponding relation of characteristic parameter and weights; And
Multichannel timing control signal through the input dendritic circuit carries out gating to the PMOS transistor of parallel connection, changes the noise margin of input stage inverter, and then realizes the control that distributes of weights time;
Wherein, the said corresponding relation that utilizes dendritic circuit to realize characteristic parameter and weights is to constitute the dendritic circuit with CMOS inverter structure by one group of parallelly connected PMOS transistor and nmos pass transistor series connection; This dendritic circuit characteristic parameter is the transistorized breadth length ratio of PMOS of parallel connection, and weights are the transistorized current strength of the PMOS that flows through; The transistorized current strength of the PMOS that flows through depends on the spatial distribution of different breadth length ratios, makes this dendritic circuit characteristic parameter and weights form corresponding relation;
The time coding signal of multiterminal input selects corresponding PMOS transistor to constitute various combination, and corresponding effectively breadth length ratio changes, thereby changes the transistorized electric current spatial distribution of PMOS of flowing through, and makes time encoding and weights form corresponding relation.
2. time encoding control weight and the information integrated method utilized according to claim 1; It is characterized in that; Said multichannel timing control signal is made up of outside stimulus signal and internal feedback signal, and this multichannel timing control signal is imported into the cynapse end of neuron dendritic circuit.
3. time encoding control weight and the information integrated method utilized according to claim 2 is characterized in that said internal feedback signal comes from the dynamic memory delay unit that is connected with the first circuit output end of pulse coupled neural.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010102015893A CN101860357B (en) | 2010-06-09 | 2010-06-09 | Method for weight control and information integration by utilizing time encoding |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2010102015893A CN101860357B (en) | 2010-06-09 | 2010-06-09 | Method for weight control and information integration by utilizing time encoding |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN101860357A CN101860357A (en) | 2010-10-13 |
| CN101860357B true CN101860357B (en) | 2012-04-11 |
Family
ID=42946046
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2010102015893A Expired - Fee Related CN101860357B (en) | 2010-06-09 | 2010-06-09 | Method for weight control and information integration by utilizing time encoding |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN101860357B (en) |
Families Citing this family (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9147155B2 (en) * | 2011-08-16 | 2015-09-29 | Qualcomm Incorporated | Method and apparatus for neural temporal coding, learning and recognition |
| CN103954659B (en) * | 2014-04-03 | 2016-08-31 | 西安电子科技大学 | A kind of method of dynamic real-time measurement cell membrane potential |
| US10423879B2 (en) | 2016-01-13 | 2019-09-24 | International Business Machines Corporation | Efficient generation of stochastic spike patterns in core-based neuromorphic systems |
| US20170330069A1 (en) * | 2016-05-11 | 2017-11-16 | Kneron Inc. | Multi-layer artificial neural network and controlling method thereof |
| CN112434802B (en) | 2020-11-06 | 2022-05-20 | 北京大学 | Method for realizing self-adaptive random pulse neuron based on ferroelectric transistor |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1139317A (en) * | 1995-04-26 | 1997-01-01 | 松下电器产业株式会社 | Logic circuit |
| US5654645A (en) * | 1995-07-27 | 1997-08-05 | Cypress Semiconductor Corp. | Buffer with controlled hysteresis |
-
2010
- 2010-06-09 CN CN2010102015893A patent/CN101860357B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN1139317A (en) * | 1995-04-26 | 1997-01-01 | 松下电器产业株式会社 | Logic circuit |
| US5654645A (en) * | 1995-07-27 | 1997-08-05 | Cypress Semiconductor Corp. | Buffer with controlled hysteresis |
Also Published As
| Publication number | Publication date |
|---|---|
| CN101860357A (en) | 2010-10-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Liu et al. | Embedding multiple trajectories in simulated recurrent neural networks in a self-organizing manner | |
| Shouval et al. | Spike timing dependent plasticity: a consequence of more fundamental learning rules | |
| US8250011B2 (en) | Autonomous learning dynamic artificial neural computing device and brain inspired system | |
| CN101860357B (en) | Method for weight control and information integration by utilizing time encoding | |
| Li et al. | Using a million cell simulation of the cerebellum: network scaling and task generality | |
| JP2016538632A (en) | Method and apparatus for tagging a class using supervised learning | |
| US20110145179A1 (en) | Framework for the organization of neural assemblies | |
| JP2017513127A (en) | Training, recognition, and generation in a spiking deep belief network (DBN) | |
| CN101770560B (en) | Information processing method and device for simulating biological neuron information processing mechanism | |
| JP2016539407A (en) | Causal saliency time inference | |
| KR20160084401A (en) | Implementing synaptic learning using replay in spiking neural networks | |
| Wu et al. | Regulating the nonlinear dynamics of olfactory cortex | |
| Gerstner | Hebbian learning and plasticity | |
| KR20160135206A (en) | Analog signal reconstruction and recognition via sub-threshold modulation | |
| Ponulak | Supervised learning in spiking neural networks with ReSuMe method | |
| CN209216149U (en) | The bionical circuit of neuron and neuromorphic system | |
| Ramezani et al. | Information theoretical analysis of synaptic communication for nanonetworks | |
| JP2017515207A (en) | Plastic synapse management | |
| JP2017511896A (en) | Auditory source separation in spiking neural networks | |
| Paugam-Moisy et al. | A supervised learning approach based on STDP and polychronization in spiking neuron networks | |
| JP2016540197A (en) | Doppler effect processing in neural network model | |
| Song et al. | Modeling elucidates how refractory period can provide profound nonlinear gain control to graded potential neurons | |
| Liljenström | 4711-Is grandma a strange attractor?,” | |
| JP2017509956A (en) | Method for converting values to spikes | |
| LI et al. | Research on learning algorithm of spiking neural network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| C17 | Cessation of patent right | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120411 Termination date: 20130609 |