CN111582470A - Self-adaptive unsupervised learning image identification method and system based on STDP - Google Patents

Self-adaptive unsupervised learning image identification method and system based on STDP Download PDF

Info

Publication number
CN111582470A
CN111582470A CN202010255290.XA CN202010255290A CN111582470A CN 111582470 A CN111582470 A CN 111582470A CN 202010255290 A CN202010255290 A CN 202010255290A CN 111582470 A CN111582470 A CN 111582470A
Authority
CN
China
Prior art keywords
neuron
neurons
output
adaptive
pulse
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.)
Granted
Application number
CN202010255290.XA
Other languages
Chinese (zh)
Other versions
CN111582470B (en
Inventor
戴琼海
郭彦辰
林星
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Tsinghua University
Original Assignee
Tsinghua University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Tsinghua University filed Critical Tsinghua University
Priority to CN202010255290.XA priority Critical patent/CN111582470B/en
Publication of CN111582470A publication Critical patent/CN111582470A/en
Application granted granted Critical
Publication of CN111582470B publication Critical patent/CN111582470B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/088—Non-supervised learning, e.g. competitive learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a self-adaptive unsupervised learning image identification method and a system based on STDP, wherein the method comprises the following steps of; inputting the input time sequence pulse signal into a pulse neural network of the self-adaptive multi-neuron model, so that the neurons generate output pulses under the combined action of positive stimulation of the input neurons and negative stimulation of the output neurons; carrying out negative feedback regulation on the emissivity of the neuron through a self-balancing function through a self-adaptive neuron model; when the input neurons receive the time sequence pulse signals, decayed positive stimulation is applied to the output neurons, and when any output neuron generates a pulse, the output neurons generate fixed negative stimulation to other output neurons; according to the hebbian theory, the excitatory synaptic connection weights between the bilayer neurons will be altered based on the pulse-time dependent plasticity rules, such that the output neuron will generate a pulse within a window time, at and only when the specific pattern it recognizes appears. This approach may prevent over-output or silence conditions from occurring.

Description

Self-adaptive unsupervised learning image identification method and system based on STDP
Technical Field
The invention relates to the technical fields of computational neurology, computer vision, nonlinear dynamics and the like, in particular to an adaptive unsupervised learning image identification method and system based on STDP (Spike Timing Dependent reliability, impulse neural network).
Background
In the biological brain, hundreds of millions of neurons communicate via synaptic interconnections to manipulate individuals into complex and meticulous biological activities, and this nervous system has been modeled and studied, and artificial neural networks have been of interest worldwide. In recent years, impulse neural networks, as third generation artificial neural networks, have been widely researched due to their remarkable biological similarity and strong computing power in pattern recognition, image processing, computer vision, etc. To date, many scholars have intensively studied various impulse neural network models and learning mechanisms, and these computational models also achieve good results in image classification, decision making and prediction. However, most of the current work is concentrated on a computer software platform, on one hand, the calculation model of the software platform about the impulse neural network is relatively mature due to the work summary of predecessors, and on the other hand, the development difficulty of the software platform from a learning algorithm to the whole neural network system is simple and not less than that of a hardware platform, but the characteristic of serial execution of the software platform is fundamentally not in accordance with the characteristic of parallel processing of biological cerebral neurons, and the problems of long simulation time, poor expansibility and the like are relatively prominent.
It is widely believed that neurons carry out information dissemination in a pulse sequence form, and an STDP learning mechanism is developed by the Hebbian rule, is considered as an important mechanism for brain learning and information storage, and belongs to an unsupervised learning mechanism. STDP achieves network balance by adjusting the pulse time difference of the presynaptic and postsynaptic connections, thus greatly ensuring the stability of the neural network.
In the previously proposed unsupervised image learning based on the pulse network, the stable identification of the output neurons depends on the relatively narrow average emissivity of the input signals, and the environmental requirements are severe. When the average emissivity is too high or too low, the situation that the output or silence cannot be identified occurs.
Disclosure of Invention
The present invention is directed to solving, at least to some extent, one of the technical problems in the related art.
Therefore, an object of the present invention is to provide an adaptive unsupervised learning image recognition method based on STDP, which adds a nonlinear adaptive variable in a neuron model, adjusts the average emissivity of neurons, and prevents the occurrence of over-output or silence, thereby greatly improving robustness and showing stable recognition in a more general input environment.
Another objective of the present invention is to provide an adaptive unsupervised learning image recognition system based on STDP.
In order to achieve the above object, an embodiment of an aspect of the present invention provides an adaptive unsupervised learning image recognition method based on STDP, including the following steps: inputting a time sequence pulse signal; inputting the time sequence pulse signal to a pulse neural network of a preset self-adaptive multi-neuron model, so that the neurons generate output pulses under the combined action of positive stimulation of input neurons and negative stimulation of output neurons; carrying out negative feedback regulation on the emissivity of the neuron through a self-balancing function through the self-adaptive neuron model; when the input neurons receive the time sequence pulse signals, decaying positive stimulation is applied to the output neurons, and when any output neuron generates a pulse, the output neurons generate fixed negative stimulation to other output neurons; according to the hebbian theory, the excitatory synaptic connection weights between the bilayer neurons will be altered based on the pulse-time dependent plasticity rules, such that the output neuron will generate a pulse within a window time, at and only when the specific pattern it recognizes appears.
According to the self-adaptive unsupervised learning image identification method based on the STDP, pulse neurons in a pulse neural network model have negative feedback adjustment variables facing self emissivity, input and output layer neurons are connected through full connection of excitatory synapses with variable weights, and output neurons are connected in pairs through inhibitory synapses with invariable weights; an STDP algorithm based on pulse time difference between neuron layers is adopted in an unsupervised learning mode, and an adaptive variable is added into a model, so that the model can still output stably under the condition that the emissivity of an input signal is more unstable; on the one hand, the design meets the bionic simulation of neuroscience; on the other hand, the algorithm does not change the original pulse neuron model, and only adds a group of negative feedback variables, so that the stability is greatly improved; compared with the traditional unsupervised impulse neural network, the learning accuracy and robustness are greatly improved, and a new thought is provided for realizing the deep multi-layer ultra-large-scale impulse neural network.
In addition, the adaptive unsupervised learning image recognition method based on STDP according to the above-described embodiment of the present invention may further have the following additional technical features:
further, in an embodiment of the present invention, the target signal of the pattern is a fixed time-sequence pulse signal, and appears according to a preset frequency on the premise that time sequences do not overlap.
Further, in an embodiment of the present invention, the neuron adopts a DSSN neuron mathematical model, wherein the DSSN neuron mathematical model is:
Figure BDA0002437071110000021
wherein v represents the cell membrane potential of the neuron, n represents the slowly changing variable of the system, a, b, c, k, p, q and r are model constants, I0For input fixed bias, IstimIs input to the neuron.
Further, in an embodiment of the present invention, the right side of the difference equation of the neuron cell membrane potential v of the DSSN neuron mathematical model is to add an adaptive variable, and the updated mathematical model is:
Figure BDA0002437071110000031
wherein, IdyFor adaptive variables, k0Is a system constant, ftIs the target emissivity.
Further, in one embodiment of the present invention, wherein the time-series pulse signal includes a plurality of fixed target signals and noise signals.
In order to achieve the above object, another embodiment of the present invention provides an adaptive unsupervised learning image recognition system based on STDP
In the self-adaptive unsupervised learning image recognition system based on the STDP, the pulse neurons in the pulse neural network model have negative feedback adjustment variables facing self emissivity, the neurons in the input and output layers are connected by full connection of excitatory synapses with variable weights, and every two neurons in the output layer are connected by inhibitory synapses with invariable weights; an STDP algorithm based on pulse time difference between neuron layers is adopted in an unsupervised learning mode, and an adaptive variable is added into a model, so that the model can still output stably under the condition that the emissivity of an input signal is more unstable; on the one hand, the design meets the bionic simulation of neuroscience; on the other hand, the algorithm does not change the original pulse neuron model, and only adds a group of negative feedback variables, so that the stability is greatly improved; compared with the traditional unsupervised impulse neural network, the learning accuracy and robustness are greatly improved, and a new thought is provided for realizing the deep multi-layer ultra-large-scale impulse neural network.
In addition, the adaptive unsupervised learning image recognition system based on STDP according to the above embodiment of the present invention may also have the following additional technical features:
further, in an embodiment of the present invention, the target signal of the pattern is a fixed time-sequence pulse signal, and appears according to a preset frequency on the premise that time sequences do not overlap.
Further, in an embodiment of the present invention, the neuron adopts a DSSN neuron mathematical model, wherein the DSSN neuron mathematical model is:
Figure BDA0002437071110000032
wherein v represents the cell membrane potential of the neuron, n represents the slowly changing variable of the system, a, b, c, k, p, q and r are model constants, I0For input fixed bias, IstimIs a neuronAnd (4) inputting.
Further, in an embodiment of the present invention, the right side of the difference equation of the neuron cell membrane potential v of the DSSN neuron mathematical model is to add an adaptive variable, and the updated mathematical model is:
Figure BDA0002437071110000041
wherein, IdyFor adaptive variables, k0Is a system constant, ftIs the target emissivity.
Further, in one embodiment of the present invention, wherein the time-series pulse signal includes a plurality of fixed target signals and noise signals.
Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.
Drawings
The foregoing and/or additional aspects and advantages of the present invention will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings of which:
fig. 1 is a flowchart of an adaptive unsupervised learning image recognition method based on STDP according to an embodiment of the present invention;
FIG. 2 is a schematic diagram of a spiking neural network according to an embodiment of the invention;
FIG. 3 is a flow chart of an STDP unsupervised learning algorithm according to an embodiment of the present invention;
FIG. 4 is a graph of neuron emissivity in accordance with an embodiment of the invention;
fig. 5 is a schematic structural diagram of an adaptive unsupervised learning image recognition system based on STDP according to an embodiment of the present invention.
Detailed Description
Reference will now be made in detail to embodiments of the present invention, examples of which are illustrated in the accompanying drawings, wherein like or similar reference numerals refer to the same or similar elements or elements having the same or similar function throughout. The embodiments described below with reference to the drawings are illustrative and intended to be illustrative of the invention and are not to be construed as limiting the invention.
The adaptive unsupervised learning image recognition method and system based on STDP according to the embodiments of the present invention will be described below with reference to the accompanying drawings, and first, the adaptive unsupervised learning image recognition method based on STDP according to the embodiments of the present invention will be described with reference to the accompanying drawings.
Fig. 1 is a flowchart of an adaptive unsupervised learning image recognition method based on STDP according to an embodiment of the present invention.
As shown in fig. 1, the adaptive unsupervised learning image recognition method based on STDP includes the following steps:
in step S101, a timing pulse signal is input.
It will be appreciated that the input to the impulse neural network is a specific time-series impulse signal, which is composed of the target signal and noise. The impulse neural network will be described in detail below, and will not be set forth herein in any greater detail.
The target signal of the graph is a fixed time sequence pulse signal and appears according to a certain frequency on the premise that time sequences are not overlapped. The occurrence of noise follows the poisson process under a certain emissivity, each pulse signal has time information codes, and each layer of neuron can respectively receive and issue pulse signals with accurate time information, so that the time difference between the front pulse and the rear pulse is a main characteristic of the pulse neural network and is also an important parameter of the STDP unsupervised learning algorithm.
In step S102, a time-series pulse signal is input to a pulse neural network of a preset adaptive multi-neuron model, so that the neurons generate output pulses due to the combined action of the positive stimulus input to the neurons and the negative stimulus output between the neurons.
It can be understood that, the time-series pulse signal is input to the pulse neural network of the preset adaptive multi-neuron model, and the neuron generates the output pulse according to the neuron model due to the combined action of the positive stimulation of the input neuron and the negative stimulation of the output neuron.
It should be noted that, as shown in fig. 2, the spiking neural network of the preset adaptive multi-neuron model is a set of two-layer multi-synaptic-delay forward feedback spiking neural network. The input and output layer neurons are connected by full connection of excitatory synapses with variable weights, and the output neurons are connected in pairs by inhibitory synapses with invariable weights. The neuron adopts a DSSN model as an impulse response model. An external input signal will cause a change in the cell membrane potential, and when the membrane potential exceeds a threshold, the neuron will fire a pulse, then enter a refractory period, and will not respond to any external stimulus.
Further, as shown in fig. 3, the embodiment of the present invention trains the network by using an STDP unsupervised learning algorithm. The STDP unsupervised learning algorithm does not depend on real-time error data of synaptic connections and teacher signals, and can realize the training process only according to the time difference between the front pulse and the rear pulse, so that the complexity of the gradient descent algorithm is avoided, and the difficulty is undoubtedly reduced for hardware design. All output neurons can fire at accurate time, namely the training process of the whole impulse neural network is completed, and a locking signal is generated. The whole network training process belongs to unsupervised learning behaviors.
The input layer neuron only participates in image information coding, and sends pulses to the next layer neuron without participating in calculation; when the neuron in the output layer receives an input signal, the potential of the cell membrane will change. The input consists of excitatory signals from the previous layer together with inhibitory signals from the same layer, and the cell membrane potential will change. The DSSN neuron mathematical model is:
Figure BDA0002437071110000051
wherein v represents the cell membrane potential of the neuron, n represents the slowly varying variable of the system, a, b, c, k, p, q, r are model constants, I0For input fixed bias, IstimIs input to the neuron.
In one embodiment of the present invention, the right side of the difference equation of the membrane potential v of the neuron cell of the DSSN model is added with an adaptive variable, and the updated mathematical model is:
Figure BDA0002437071110000061
wherein, IdyFor adaptive variables, k0Is a system constant, ftIs the target emissivity.
Referring to FIG. 3 at the dotted line and FIG. 4, when the mean neuron emittance is below ftWhen, IdyWill increase to increase the emittance, whereas when the average emittance of the neurons is higher than ftWhen, IdyWill be reduced to lower the emissivity. Therefore, the over-emission and silence conditions are prevented, and the function stability can be further kept under different input emissivity.
In step S103, the self-adaptive neuron model performs negative feedback adjustment on the emissivity of the neuron through a self-balancing function.
In step S104, when the input neuron receives the time-series pulse signal, it applies decaying positive stimulation to the output neurons, and when any output neuron appears pulsed, the output neurons generate fixed negative stimulation to other output neurons.
In step S105, according to hebry theory, the excitatory synaptic connection weights between the bilayer neurons will be changed based on the pulse-time dependent plasticity rules, such that the output neurons will generate pulses within the window time, when and only when the specific pattern they recognize appears.
In summary, the embodiment of the invention uses a software platform to design the model of the impulse neural network, and innovatively uses the adaptive variable of negative feedback in the learning and training of the impulse neuron model, thereby realizing specific functions. The impulse neural network model has high robustness in a complex input environment. The emissivity of the hidden layer is ensured in the multilayer unsupervised neural network, the expansibility of the hidden layer is greatly improved, and a new thought is provided for realizing the deep unsupervised pulse neural network.
According to the self-adaptive unsupervised learning image identification method based on the STDP, pulse neurons in a pulse neural network model have negative feedback adjustment variables facing self emissivity, neurons in an input and output layer are connected by full connection of excitatory synapses with variable weights, and two neurons in an output layer are connected by inhibitory synapses with invariable weights; an STDP algorithm based on pulse time difference between neuron layers is adopted in an unsupervised learning mode, and an adaptive variable is added into a model, so that the model can still output stably under the condition that the emissivity of an input signal is more unstable; on the one hand, the design meets the bionic simulation of neuroscience; on the other hand, the algorithm does not change the original pulse neuron model, and only adds a group of negative feedback variables, so that the stability is greatly improved; compared with the traditional unsupervised impulse neural network, the learning accuracy and robustness are greatly improved, and a new thought is provided for realizing the deep multi-layer ultra-large-scale impulse neural network.
Next, an adaptive unsupervised learning image recognition system based on STDP according to an embodiment of the present invention will be described with reference to the drawings.
Fig. 5 is a schematic structural diagram of an adaptive unsupervised learning image recognition system based on STDP according to an embodiment of the present invention.
As shown in fig. 5, the adaptive unsupervised learning image recognition system 10 based on STDP includes: a first input module 100, a second input module 200, a feedback adjustment module 300, an application module 400, and a change module 500.
The first input module 100 is configured to input a timing pulse signal; the second input module 200 is configured to input a time-series pulse signal to a pulse neural network of a preset adaptive multi-neuron model, so that a neuron generates an output pulse due to a combined action of a positive stimulus input to the neuron and a negative stimulus output between the neurons; the feedback adjusting module 300 is configured to perform negative feedback adjustment on the emissivity of the neuron through a self-balancing function through a self-adaptive neuron model; the applying module 400 is used for applying decaying positive stimulation to the output neurons when the input neurons receive the time sequence pulse signals, and the output neurons generate fixed negative stimulation to other output neurons when any output neuron generates pulses; the changing module 500 is used to change the excitatory synaptic connection weights between the neurons in the double layer according to hebry theory based on the pulse-time dependent plasticity rule so that the output neurons generate pulses within the window time when and only when the specific pattern they recognize appears. The system 10 of the embodiment of the invention adds the nonlinear adaptive variable in the neuron model, adjusts the average emissivity of the neuron, and prevents the occurrence of over-output or silence, thereby greatly improving the robustness and showing stable identification in a more universal input environment.
Further, in an embodiment of the present invention, the target signal of the pattern is a fixed time-sequence pulse signal, and appears according to a preset frequency on the premise that time sequences do not overlap.
Further, in an embodiment of the present invention, the neuron uses a DSSN neuron mathematical model, wherein the DSSN neuron mathematical model is:
Figure BDA0002437071110000071
wherein v represents the cell membrane potential of the neuron, n represents the slowly changing variable of the system, a, b, c, k, p, q and r are model constants, I0For input fixed bias, IstimIs input to the neuron.
Further, in one embodiment of the present invention, the right side of the difference equation of the neuron cell membrane potential v of the DSSN neuron mathematical model is to add an adaptive variable, and the updated mathematical model is:
Figure BDA0002437071110000072
wherein, IdyFor adaptive variables, k0Is a system constant, ftIs the target emissivity.
Further, in one embodiment of the present invention, wherein the timing pulse signal includes a plurality of fixed target signals and noise signals.
It should be noted that the foregoing explanation on the embodiment of the adaptive unsupervised learning image recognition method based on STDP is also applicable to the adaptive unsupervised learning image recognition system based on STDP in this embodiment, and is not repeated here.
According to the self-adaptive unsupervised learning image recognition system based on the STDP, pulse neurons in a pulse neural network model have negative feedback adjustment variables facing self emissivity, neurons in an input and output layer are connected by full connection of excitatory synapses with variable weights, and two neurons in an output layer are connected by inhibitory synapses with invariable weights; an STDP algorithm based on pulse time difference between neuron layers is adopted in an unsupervised learning mode, and an adaptive variable is added into a model, so that the model can still output stably under the condition that the emissivity of an input signal is more unstable; on the one hand, the design meets the bionic simulation of neuroscience; on the other hand, the algorithm does not change the original pulse neuron model, and only adds a group of negative feedback variables, so that the stability is greatly improved; compared with the traditional unsupervised impulse neural network, the learning accuracy and robustness are greatly improved, and a new thought is provided for realizing the deep multi-layer ultra-large-scale impulse neural network.
Furthermore, the terms "first", "second" and "first" are used for descriptive purposes only and are not to be construed as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one such feature. In the description of the present invention, "a plurality" means at least two, e.g., two, three, etc., unless specifically limited otherwise.
In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above are not necessarily intended to refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. Furthermore, various embodiments or examples and features of different embodiments or examples described in this specification can be combined and combined by one skilled in the art without contradiction.
Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention, and that variations, modifications, substitutions and alterations can be made to the above embodiments by those of ordinary skill in the art within the scope of the present invention.

Claims (10)

1. An adaptive unsupervised learning image identification method based on STDP is characterized by comprising the following steps:
inputting a time sequence pulse signal;
inputting the time sequence pulse signal to a pulse neural network of a preset self-adaptive multi-neuron model, so that the neurons generate output pulses under the combined action of positive stimulation of input neurons and negative stimulation of output neurons;
carrying out negative feedback regulation on the emissivity of the neuron through a self-balancing function through the self-adaptive neuron model;
when the input neurons receive the time sequence pulse signals, decaying positive stimulation is applied to the output neurons, and when any output neuron generates a pulse, the output neurons generate fixed negative stimulation to other output neurons; and
according to the hebbian theory, the excitatory synaptic connection weights between the bilayer neurons will be altered based on the pulse-time dependent plasticity rules, such that the output neuron will generate a pulse within a window time, at and only when the specific pattern it recognizes appears.
2. The method of claim 1, wherein the target signal of the pattern is a fixed time-sequence pulse signal, and occurs according to a predetermined frequency without overlapping the time sequence.
3. The method of claim 1, wherein the neurons employ a DSSN neuron mathematical model, wherein the DSSN neuron mathematical model is:
Figure FDA0002437071100000011
wherein v represents the cell membrane potential of the neuron, n represents the slowly changing variable of the system, a, b, c, k, p, q and r are model constants, I0For input fixed bias, IstimIs input to the neuron.
4. The method of claim 3, wherein the right side of the difference equation for the neuron cell membrane potential v of the DSSN neuron mathematical model is to add an adaptive variable, and the updated mathematical model is:
Figure FDA0002437071100000012
wherein, IdyFor adaptive variables, k0Is a system constant, ftIs the target emissivity.
5. The method of any one of claims 1-4, wherein the time-series pulse signal comprises a plurality of fixed target signals and noise signals.
6. An adaptive unsupervised learning image recognition system based on STDP, comprising:
the first input module is used for inputting a time sequence pulse signal;
the second input module is used for inputting the time sequence pulse signal to a pulse neural network of a preset self-adaptive multi-neuron model, so that the neurons generate output pulses under the combined action of positive stimulation of the input neurons and negative stimulation of the output neurons;
the feedback adjusting module is used for carrying out negative feedback adjustment on the emissivity of the neuron through a self-balancing function through the self-adaptive neuron model;
the application module is used for applying decaying positive stimulation to the output neurons when the input neurons receive the time sequence pulse signals, and the output neurons generate fixed negative stimulation to other output neurons when any output neuron generates pulses; and
and the changing module is used for changing the excitatory synaptic connection weight between the double-layer neurons based on the pulse time-dependent plasticity rule according to the hebbian theory so that the output neurons generate pulses within the window time when and only when the specific patterns identified by the output neurons appear.
7. The system of claim 6, wherein the target signal of the pattern is a fixed time-sequence pulse signal, and occurs according to a predetermined frequency without overlapping the time sequence.
8. The system of claim 6, wherein the neurons employ a DSSN neuron mathematical model, wherein the DSSN neuron mathematical model is:
Figure FDA0002437071100000021
wherein v represents the cell membrane potential of the neuron, n represents the slowly changing variable of the system, a, b, c, k, p, q and r are model constants, I0For input fixed bias, IstimIs input to the neuron.
9. The system of claim 8, wherein the right side of the difference equation for the neuron cell membrane potential v of the DSSN neuron mathematical model is to add an adaptive variable, and the updated mathematical model is:
Figure FDA0002437071100000022
wherein, IdyFor adaptive variables, k0Is a system constant, ftIs the target emissivity.
10. The system according to any one of claims 6-9, wherein the time-series pulse signal comprises a plurality of fixed target signals and noise signals.
CN202010255290.XA 2020-04-02 2020-04-02 Self-adaptive unsupervised learning image identification method and system based on STDP Active CN111582470B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010255290.XA CN111582470B (en) 2020-04-02 2020-04-02 Self-adaptive unsupervised learning image identification method and system based on STDP

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010255290.XA CN111582470B (en) 2020-04-02 2020-04-02 Self-adaptive unsupervised learning image identification method and system based on STDP

Publications (2)

Publication Number Publication Date
CN111582470A true CN111582470A (en) 2020-08-25
CN111582470B CN111582470B (en) 2023-01-10

Family

ID=72112981

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010255290.XA Active CN111582470B (en) 2020-04-02 2020-04-02 Self-adaptive unsupervised learning image identification method and system based on STDP

Country Status (1)

Country Link
CN (1) CN111582470B (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112749637A (en) * 2020-12-29 2021-05-04 电子科技大学 Distributed optical fiber sensing signal identification method based on SNN

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105659262A (en) * 2013-11-08 2016-06-08 高通股份有限公司 Implementing synaptic learning using replay in spiking neural networks
CN108985447A (en) * 2018-06-15 2018-12-11 华中科技大学 A kind of hardware pulse nerve network system

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105659262A (en) * 2013-11-08 2016-06-08 高通股份有限公司 Implementing synaptic learning using replay in spiking neural networks
CN108985447A (en) * 2018-06-15 2018-12-11 华中科技大学 A kind of hardware pulse nerve network system

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
NANAMI T等: "An FPGA-based cortical and thalamic silicon neuronal network", 《JOURNAL OF ROBOTICS NETWORKING AND ARTIFICIAL LIFE》 *
TAKASHI KOHNO等: "Silicon neuronal networks towards brain-morphic computers", 《NONLINEAR THEORY AND ITS APPLICATIONS》 *
蔺想红等: "脉冲神经网络的监督学习算法研究综述", 《电子学报》 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112749637A (en) * 2020-12-29 2021-05-04 电子科技大学 Distributed optical fiber sensing signal identification method based on SNN
CN112749637B (en) * 2020-12-29 2023-09-08 电子科技大学 SNN-based distributed optical fiber sensing signal identification method

Also Published As

Publication number Publication date
CN111582470B (en) 2023-01-10

Similar Documents

Publication Publication Date Title
Hunsberger et al. Spiking deep networks with LIF neurons
Sporea et al. Supervised learning in multilayer spiking neural networks
Kasiński et al. Comparison of supervised learning methods for spike time coding in spiking neural networks
Bohnstingl et al. Online spatio-temporal learning in deep neural networks
Shrestha et al. Stable spike-timing dependent plasticity rule for multilayer unsupervised and supervised learning
CN111639754A (en) Neural network construction, training and recognition method and system, and storage medium
CN111582470B (en) Self-adaptive unsupervised learning image identification method and system based on STDP
Musakulova et al. Synthesis of the backpropagation error algorithm for a multilayer neural network with nonlinear synaptic inputs
Stanojevic et al. An exact mapping from ReLU networks to spiking neural networks
Allred et al. Stimulating stdp to exploit locality for lifelong learning without catastrophic forgetting
JP7336710B2 (en) Neural network system, learning method and program
CN111582461B (en) Neural network training method and device, terminal equipment and readable storage medium
KR20220075025A (en) Neuromorphic computing device
Humaidi et al. Recognition of arabic characters using spiking neural networks
Gupta et al. Design and analysis of configurable multilayer spiking neural network
Patra et al. A novel approach for pattern recognition
Мигалев et al. Modeling the Learning of a Spiking Neural Network with Synaptic Delays
US20240257943A1 (en) Training method of brain activity state classification model, brain activity state classification method, device, and storage medium
Yousefi et al. Time Divergence-Convergence Learning Scheme in Multi-Layer Dynamic Synapse Neural Networks
Zhang et al. A Bionic Spiking Recurrent Neural Network with Sparse Connections and Dale's Principle
Selma et al. Trajectory estimation and control of vehicle using neuro-fuzzy technique
Munavalli et al. Pattern recognition for data retrieval using artificial neural network
US20230117659A1 (en) Device and method for recognizing image using brain-inspired spiking neural network and computer readable program for the same
Kar et al. Artificial neural networks and learning techniques
Ballini et al. Recurrent fuzzy neural computation: Modeling, learning and application

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
GR01 Patent grant
GR01 Patent grant