JP2002150258A - Circuit using neuron model and information processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a circuit of a neuron model further approaching to an information processing mechanism of a living body, various types of circuits applying the neuron model circuit, and an information processing method. SOLUTION: This circuit using the neuron model is constituted of olfactory cell models (input parts) 14 and 15 receiving an input signal and a closed loop 1 composed of a plurality of neuron model units 2 having input/output characteristics following the Hodgkin-Huxley equation and transmitting saltation after inhibition. The closed loop 1 includes inhibitory neuron models 11 and 12 and forms rhythm patterns to autonomous vibration so as to provide various operations.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

TECHNICAL FIELD The present invention relates to visual, auditory, taste,
The present invention relates to a neuron model circuit and an information processing method capable of processing information such as olfactory and tactile sensations corresponding to a living body's senses and perceptions. , Flip-flop circuit,
The present invention relates to an active filter and an information processing method.

[0002]

2. Description of the Related Art Neurons are responsible for information processing in the nervous system and brain of a living body. Neurons are cells specialized for information processing, and neurons are connected to each other, and the connection site is called a synapse.

When the potential of a neuron population is measured by a microelectrode method or an optical measurement method using a potential-sensitive dye, potential oscillations of a specific cycle are observed in sensory / perceptual centers and motor centers. This potential oscillation is thought to be caused by an intrinsic neuron circuit, which is found in the movement center in a circuit that controls repetitive movements such as gastric peristalsis, and the repetitive movement occurs according to the oscillation cycle of this circuit . Such a repetitive motion is observed not only at the motor center but also at the sensory / perceptual center as a vibration phenomenon of the neuron circuit, and its vibration cycle is unique.

[0004] The circuit that causes the vibration phenomenon is particularly a central pattern generator (CentralPattern Generator) in the movement center.
By incorporating such a pattern generator, the creation of a so-called neurochip that can realize sensory / perceptual information processing in the brain will be realized.

[0005]

One of the research fields of a neurochip is to create a neural circuit configuration on a substance other than a living body such as silicon. As an example, a simulation such as swimming of a hill is performed. Some reports have been developed on silicon chips (for example, “A Si
licon Model of the Hirude Swim Oscillator "S. Wolpe
rt, WO Friesen, and AJLaffely (2000) IEEE Eng.
Med. Biol. 19 64-75).

In such an attempt, a buffer or amplifier is formed on silicon to form a circuit simulating a central pattern generator, and a pair of inhibitory neurons 1 shown in FIG.
A circuit simulating the operation corresponding to the set of 00, 101, and the five inhibitory neurons 111, 112, 11 shown in FIG.
An oscillation circuit is formed in which 3, 114, and 115 are arranged in a ring to sequentially transmit potential fluctuations. Here, VLSI (Very Large Scale Integrated Circui
t) Incorporating technology and positioning it as an extension of the current semiconductor chip.

However, the senses belonging to the actual living body
The information processing in the brain of perception is extremely complicated, and a circuit in which a buffer and an amplifier are simply combined on silicon as shown in FIG. 41 has a circuit response characteristic that is too simple and insufficient. The necessity of simulating the information processing mechanism of a living body is left in a neural network or a neurochip.

It is therefore an object of the present invention to provide a circuit of a neuron model which is much closer to the information processing mechanism of an actual living body, various circuits to which the circuit of the neuron model is applied, and an information processing method.

[0009]

The present inventors have noticed that the behavior of the synapse of a living body as a membrane potential can be simulated by the Hodgkin-Huxley equation (or Hodgkin-Huxley equation). It is found that the application of the information processing to the information processing such as sensation and perception makes it possible to construct a neuron model circuit closer to the living body. It is characterized in that it is processed in a form similar to a living body due to its configuration.

A circuit of a neuron model as an example of the present invention comprises an input unit for receiving an input signal, and a plurality of neuron model units whose input / output characteristics follow the Hodgkin-Huxley equation. And a closed loop for transmitting.

The input section is a contact point with the outside of the circuit, and receives, for example, signals corresponding to any of smell, sight, hearing, taste, touch, memory, recognition, exercise, and a combination thereof. This input is connected to a closed loop, and the input signal is transmitted to the closed loop.

The closed loop is a circuit element for generating a periodic vibration simulating a rhythm pattern in the central pattern generator. In the present invention, the input / output characteristics of the closed loop follow the Hodgkin-Huxley equation. This Hodgkin-Huxley equation is an equation representing the action potential action in a living body. L. Hodgkin and A. F.
Na derived from experimental results of membrane current measurement by Huxley
^The conductivity versus the amount of ⁺ , K ⁺ ions is a function of membrane potential and time. Specifically, it is expressed by the following equation (1).

[0013]

(Equation 1)

According to the Hodgkin-Huxley equation, the case where the nerves are simultaneously excited and the manner in which the excited potential is transmitted are shown in the equations. In the present invention,
As a configuration for indicating the input / output characteristics according to the Hodgkin-Huxley equation, the input / output characteristics may be configured as a unit of a plurality of neuron models according to the Hodgkin-Huxley equation. Hodgkin using a semiconductor element
A unit of a plurality of neuron models according to the Huxley equation may be configured.

The circuit of the neuron model of the present invention simulates the behavior of a biological synapse as a membrane potential, and forms a rhythm pattern accompanied by postinhibitory rebound. The jump after inhibition is a firing phenomenon that occurs when the hyperpolarization returns to the resting membrane potential. In a living body, the inhibited neuron undergoes hyperpolarization. In the closed loop consisting of units of a plurality of neuron models whose input / output characteristics follow the Hodgkin-Huxley equation together with the jump after suppression, the mutually inhibitory neuron connection contributes to the generation of the vibration rhythm pattern in the central pattern generator. Mutual inhibitory neuron connections are formed by at least one pair of inhibitory neuron models, and a greater number of inhibitory neuron models may be connected in a ring to generate a rhythm pattern. An excitatory neuron model may be combined with a part of the closed loop. By combining the excitatory neuron model, it is possible to act so as to weaken a steady autonomous oscillation by a simple mutual inhibition circuit.

The circuit of the above-described neuron model is as follows.
Each of them can be used as a network of a neuron model circuit, a flip-flop circuit, and an active filter. The circuit network of the neuron model is a network in which the circuits of a plurality of neuron models are arranged in a network and connected, and the synapse connection between the circuits of each neuron model may have regularity. It may be something. In this case, depending on the connection, a plurality of closed loops are built in the network of the neuron model, the rhythm patterns generated are further diversified, and furthermore, information processing close to the information processing mechanism of sensory / perception of a living body is performed. Becomes possible.

Further, the flip-flop circuit of the present invention comprises:
An input unit for receiving an input signal, and a closed loop connected to the input unit and transmitting a jump after suppression, comprising a plurality of neuron models whose input / output characteristics follow the Hodgkin-Huxley equation, and a flip-flop for the input signal. The circuit is characterized by performing a response. According to the experimental example conducted by the present inventors, the circuit of the neuron model returns to the original autonomous oscillation when the potential stimulus stops as a response, and to a state of another phase in a special case. In some cases, the phase shifter operates to maintain the phase, and the flip-flop circuit is configured by using the latter circuit operation.

Further, the active filter of the present invention comprises:
An input unit for receiving an input signal, and having a closed loop connected to the input unit and transmitting a jump after suppression, comprising a plurality of neuron models whose input / output characteristics conform to the Hodgkin-Huxley equation, and different outputs depending on the input signal A signal is obtained. As described above, the input signal can be a sensory / perceptual information signal, and the active filter of the present invention can sort the information signal while utilizing autonomous vibration to identify information.

Further, according to the information processing method of the present invention, a required input signal is input to the input unit, and the input / output
A signal including a jump after suppression is transmitted as a closed-loop signal transmission composed of a plurality of neuron models according to the Huxley equation and connected to the input unit, and an output signal is extracted from a part of the closed loop. In this information processing method, information processing can be advanced by using a rhythm pattern generated in a closed loop and responding to the input signal.

[0020]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a diagram showing an embodiment of a circuit of a neuron model of the present invention. Each neuron model consists of units of a neuron model. A pair of inhibitory neuron models 11 and 12 are connected to each other via an inhibitory synapse 16, and the inhibitory synapse 16
A pair of inhibitory neuron models 11 connected via
12 constitutes the closed loop 1. The portion including the closed loop formed by the pair of inhibitory neuron models 11 and 12 is the basic configuration of the circuit of the neuron model of the present embodiment. Is provided with an excitatory neuron model 13 as an output neuron. By using the excitatory neuron model 13, a rhythm pattern by autonomous oscillation can be diversified. The pair of inhibitory neuron models 11 and 12 are connected via excitatory synapses. Olfactory cell models 14 and 15 are connected to the pair of inhibitory neuron models 11 and 12, respectively. These olfactory cell models 14 and 15 function as a pair of input neuron models for the pair of inhibitory neuron models 11 and 12 constituting the closed loop 1, that is, an input unit. In the present embodiment, potential stimuli simulating a perceptual signal are applied. Can be

Each of the pair of inhibitory neuron models 11 and 12 constituting the closed loop 1 is composed of a plurality of units 2 of neuron models, and each unit 2 of each neuron model follows the Hodgkin-Huxley equation. 9 shows input / output characteristics, that is, dynamics of a membrane potential. Specifically, the Hodgkin-Huxley equation is expressed by the following equation (equivalent to the above-described equation 1).

[0022]

(Equation 2) Here, the Hodgkin-Huxley equation specifically expresses a temporal change of ions (Na, K) entering and exiting through an ion channel in a nerve cell membrane, and describes an equation describing a relationship between three currents including a leak current and a membrane current. It is. The Hodgkin-Huxley equation has a diffusion term d
The following equation (3) to which (∇ ² V _m ) / _Ra is added is generally called a cable equation, and describes a behavior in which an action potential propagates through a nerve.

(Equation 3)

The unit 2 of each neuron model according to the Hodgkin-Huxley equation is a minimum unit obtained by discretizing the cable equation of the above equation 3, and has a circuit structure having an input stage and an output stage. It has a structure in which one or more inputs are integrated, and a structure in which the aggregated portion is expanded to one or more outputs in an output stage. Signal transmission in the unit 2 of the neuron model is performed by propagation of firing of the membrane potential. The membrane potential usually shows around -60 mV. When the membrane potential is depolarized (increased) by a signal input and exceeds the threshold value, the membrane potential rapidly rises to about +40 mV within a few ms. Firing this phenomenon (Firing, or excitement (Exc
itation)).

Here, regarding the relationship between the unit 2 of the neuron model and the inhibitory neuron models 11, 12 and the excitatory neuron model 13, each of the inhibitory neuron models 11, 12 and the excitatory neuron model 13 is a single neuron. It is also possible to adopt a structure composed of model units 2, and as shown in the figure, a plurality of neuron model units 2 can form one group to constitute each neuron. For example, when a circuit of the neuron model circuit of the present embodiment or a circuit network of a neuron model circuit to be described later is simulated by computer software or a semiconductor element, the unit 2 of one neuron model is formed. For other neuron model units, the position may be changed and copied, and a complex neuron model circuit or a network of neuron model circuits can be easily configured. In addition, the unit 2 of each neuron model according to the Hodgkin-Huxley equation can also be a constituent element of the olfactory cell models 14, 15 and the synapse 16 connecting the neuron models.

In the circuit of the neuron model of the present embodiment having such a circuit configuration, a periodic rhythm pattern is generated in the closed loop 1 formed by the interconnecting portions of the pair of inhibitory neuron models 11 and 12. This rhythm pattern is a signal form accompanied by a postinhibitory rebound, and a periodic rhythm pattern is generated in such a form that the post-suppression jump is sequentially transmitted between the neuron models. The jump after inhibition is a firing phenomenon that occurs when the hyperpolarization returns to the resting membrane potential. In a living body, the inhibited neuron undergoes hyperpolarization.

In the circuit of the neuron model of the present embodiment, the dynamics of the synaptic membrane potential typically follows the Hodgkin-Huxley equation. This synapse can typically be selected from three types: electrical synapses (gap junctions), excitatory synapses, and inhibitory synapses.
In the electrical synapse (gap connection), signals are transmitted bidirectionally between neuron models, and the input stage and the output stage are connections that can be substituted for each other. Excitatory synapses transmit signals in only one direction, and the membrane potential at the input stage of the receiving neuron model depolarizes (rises). At the time of connection by the excitatory synapse, the membrane potential indicates firing if the membrane potential exceeds a threshold value at which firing occurs. Finally, a signal is transmitted to the inhibitory synapse only in one direction, and the membrane potential of the input stage of the receiving neuron model is hyperpolarized (decreased).
This hyperpolarization returns to a steady state with time according to the Hodgkin-Huxley equation. However, the threshold value is lowered according to the Hodgkin-Huxley equation due to hyperpolarization. Therefore, when the membrane potential returns to the steady state, if the decreased threshold value is exceeded, the membrane potential indicates ignition, that is, a jump after suppression.

In the circuit of the neuron model of the present embodiment, the olfactory cell models 14 and 15 are used as input parts of the circuit of the neuron model. This olfactory cell model 1
Nos. 4 and 15 are nothing but a set of neuron models according to the Hodgkin-Huxley equation (neuron model). The reason why the olfactory cell models 14 and 15 are used as the input unit of the circuit of the neuron model in the present embodiment is that the research on the sensation / perception center of the olfaction is the most common knowledge in the field of current neuroanatomy and neurophysiology. This is a result of the idea of configuring a circuit for handling odor information by imitating a living body, and the circuit of the neuron model of the present embodiment is composed of olfactory cell models 14 and 15
Other perceptual / sensory circuits may be used as the circuits which can be replaced with the above. In addition, the olfactory cells in the living body have a potential of 20 to 30 Hz as the potential of the olfactory lobe due to odor reception.
Is generated, and this frequency is not related to the type of smell. Then, the odor information transmitted from the olfactory cells is sent to the olfactory lobe, where the odor information is spatiotemporally coded according to the rhythm pattern, and sent to the mushroom body.

The information that can be processed in the present embodiment is not limited to information relating to olfaction, but is a signal corresponding to any of visual, auditory, taste, tactile, memory, cognitive, and movement, and a combination thereof. Is also good. The circuit of the neuron model of the present embodiment includes a plurality of neuron models 11, 1
2 has a neuron (input neuron) model for receiving an input signal from the outside of the circuit and a neuron (output neuron) model for transmitting an output signal to the outside of the circuit, using a circuit connected to each other by an inhibitory synapse 16 as a basic circuit. ing. The input neuron model connects to the basic circuit at an excitatory synapse, and the output neuron model connects to the basic circuit at an inhibitory or excitatory synapse. As a modification, instead of a circuit of a neuron model connected by mutually inhibitory synapses, a circuit of a cyclic neuron model including a plurality of neuron models having a connection of inhibitory synapses in one direction may be used. . Also, a neuron model with excitatory synaptic connections may be inserted between inhibitory synaptic connections.

Next, a circuit network of a neuron model constructed using this embodiment will be described with reference to FIG. The network of the neuron model of FIG. 2 has a structure in which a plurality of inhibitory neuron models 20 to 27 and a plurality of excitatory neuron models 31 to 33 are connected in a network via required synapses. A plurality of closed loops are multiplexed in the circuit network of this neuron model.
For example, a pair of an inhibitory neuron model 21 and an inhibitory neuron model 24, a pair of an inhibitory neuron model 24 and an inhibitory neuron model 25, and an inhibitory neuron model 2
2. An annular closed loop constituted by a set of an excitatory neuron model 33, an inhibitory neuron model 27, and an inhibitory neuron model 25 is multiplexed via synapses. Even in such a circuit network configuration, it is assumed that a rhythm pattern is generated in each closed loop, and the entire network of the neuron model functions as a complex that is closer to the living body and can perform various signal processing. .

Next, based on experiments conducted by the inventor, the circuit of the neuron model of the present invention is shown in FIGS.
This will be further described with reference to FIG. The experiment was conducted for five circuit patterns from circuit A to circuit E. 3 to 7 are circuit diagrams of circuits A to E, respectively. FIGS. 8 to 14 are response waveform diagrams of the output neuron model of the circuit A. FIGS. 15 to 17 are output neuron models of the circuit B. 18 to 23 are response waveform diagrams of the output neuron model of the circuit C, FIGS. 24 to 30 are response waveform diagrams of the output neuron model of the circuit D, and FIGS. 31 to 38. FIG. 9 is a response waveform diagram of the output neuron model of the circuit E.

FIG. 3 shows the configuration of the circuit A. A pair of inhibitory neuron models 41 and 42 mutually inhibit inhibitory synapse 4
3 and 44, and these inhibitory neuron models 41 and 42 and inhibitory synapses 43 and 44 form a closed loop. Each of the inhibitory neuron models 41 and 42 is formed from 5 × 5 neuron model units. The unit of each neuron model is a circuit whose input / output characteristics follow the Hodgkin-Huxley equation.

A pair of olfactory cell models 46 and 47 serve as an input neuron model and a pair of inhibitory neuron models 4
1, 42. The olfactory cell models 46 and 47 are also a set of neuron models according to the Hodgkin-Huxley equation (neuron model), and are connected to a pair of inhibitory neuron models 41 and 42 via excitatory synapses 48 and 49, respectively. As an output neuron model, an excitatory neuron model 45 is configured to couple to a pair of inhibitory neuron models 41 and 42 via excitatory synapses 50 and 51, respectively.

The circuit operation of the circuit A will be described.
Circuit A shows a steady operation, which is a periodic oscillation in a closed loop, and an operation in response to an input signal. First, with respect to the steady operation, when the inhibitory neuron model 42 was subjected to potential stimulation, it was found that the output neuron model oscillates at a constant period after that, without applying a potential stimulus.

Next, in order to input a signal to the circuit of the neuron model, potential stimulation was simultaneously applied to both input neuron models of the olfactory cell models 46 and 47. 3000, 5000, 8000, 100
00, 12000, 14000, 20000 steps were given 10 times from the 250,000th step. When the interval was 20,000 steps, no change was observed in the oscillation cycle of the output neuron model. 3000, 500
0, 8000, 10000, 12000, and 140
In the case of 00, the period changed to a period synchronized with twice the period of the input signal. Further, it was observed that when the input of the signal was interrupted, the vibration cycle returned to the original one. FIG. 8 is a response waveform diagram of the output neuron model of the circuit A when a potential stimulus is applied at 3000 step intervals, and FIG. 9 is a response waveform of the output neuron model of the circuit A when a potential stimulus is applied at 5000 step intervals. FIG. 10 is a response waveform diagram of the output neuron model of the circuit A when a potential stimulus is applied at an interval of 8000 steps. FIG. 11 is an output neuron model of the circuit A when a potential stimulus is applied at an interval of 10,000 steps. FIG. 12 is a response waveform diagram of the model A. FIG. 12 is a response waveform diagram of the output neuron model of the circuit A when a potential stimulus is applied at an interval of 12000 steps. FIG. 13 is a circuit diagram when a potential stimulus is applied at an interval of 14000 steps. FIG. 14 is a response waveform diagram of the output neuron model of circuit A. FIG. 14 shows the output neuron model of circuit A when a potential stimulus is applied at 20,000 step intervals. It is a response waveform chart of a Le.

As described above, the circuit A used in the experiment
No response to the input signal at the interval of 000 steps, 3000
In response to an input signal at ~ 14000 step intervals, an output signal tuned to it was shown. It has a response or non-response selectivity to the interval of the input signal, and changes to one of two phases of response or non-response. Furthermore, when the input signal was interrupted, it returned to the original cycle autonomously. From this, it can be seen that in addition to having selectivity for the input signal interval, it also functions as an active filter that performs autonomous return to the original state when the input signal is interrupted. It has been found possible to identify the signal. In this specification, an active filter is a filter having a function of switching a phase state. When an input signal is input, the active filter shifts from one phase state to another phase state. Used as a return to.

FIG. 4 shows the configuration of the circuit B. A pair of inhibitory neuron models 53 and 54 mutually inhibit inhibitory synapse 5
The inhibitory neurons 53 and 54 and inhibitory synapses 55 and 56 form a closed loop. Each of the inhibitory neuron models 53 and 54 is formed from 5 × 5 neuron model units. The unit of each neuron model is a circuit whose input / output characteristics follow the Hodgkin-Huxley equation.

A pair of olfactory cell models 60 and 61 serve as an input neuron model and a pair of inhibitory neuron models 5
3, 54. The olfactory cell models 60 and 61 are also a set of neuron models according to the Hodgkin-Huxley equation (neuron model), and are connected to a pair of inhibitory neuron models 53 and 54 via excitatory synapses 62 and 63, respectively. As an output neuron model, an excitatory neuron model 58 is configured to be coupled to one inhibitory neuron model 53 via excitatory synapses 58 and 59, respectively. That is, the configuration of the circuit B is different from that of the circuit A shown in FIG. 3 in that the excitatory neuron model 58 is connected to only one inhibitory neuron model 53, and the other circuit configurations are the same.

The circuit operation of the circuit B will be described.
The circuit B, like the circuit A, shows a steady operation, which is a periodic oscillation in a closed loop, and an operation in response to an input signal. First, regarding the steady operation, the inhibitory neuron model 54
Was stimulated with a potential, and it turned out that it exhibited an autonomous vibration and vibrated at a steady period without applying a potential stimulus.

Next, in order to input a signal to the circuit of the neuron model, both input neurons of the olfactory cell models 60 and 61 were simultaneously subjected to potential stimulation. The intervals between the potential stimulations were 10,000, 12,000, and 20,000 steps, and were given 10 times from the 250,000th step. When the interval was 20,000 steps, no change was observed in the oscillation cycle of the output neuron model. 10,000 and 120
In the case of 00, the period changed to a period synchronized with twice the period of the input signal. When the input of the signal was interrupted, it returned to the original cycle.
FIG. 15 is a response waveform diagram of an output neuron model of the circuit B when a potential stimulus is applied at an interval of 10,000 steps.
FIG. 16 is a response waveform diagram of an output neuron model of the circuit B when a potential stimulus is applied at an interval of 12000 steps.
FIG. 17 is a response waveform diagram of the output neuron model of the circuit B when a potential stimulus is applied at an interval of 20,000 steps.

As described above, the circuit B manufactured is the circuit A
In the same manner as in the above operation, the apparatus did not respond to the input signal at the interval of 20,000 steps, responded to the input signal at the intervals of 10,000 steps and 12,000 steps, and showed the output signal synchronized therewith. It has a response or non-response selectivity with respect to the interval of the input signal and changes to one of two phases of response or non-response. Further, when the input signal was interrupted, it returned to the original cycle autonomously. In addition to having selectivity for the interval of the input signal,
It functioned as an active filter that autonomously returned to its original state when the input signal was interrupted. It has been found that it is possible to identify the input signal by the change to two phases.

Next, the configuration of the circuit C will be described. FIG.
As shown in FIG. 2, one excitatory neuron model 67 is laid out so as to sandwich a pair of inhibitory neuron models 65 and 66. The connection between the excitatory neuron model 67 and the inhibitory neuron model 65 is established by an excitatory synapse. 69 and a combination of inhibitory synapses 68,
The connection between the excitatory neuron model 67 and the inhibitory neuron model 66 is connected from a combination of the excitatory synapse 70 and the inhibitory synapse 71. The closed loop includes one excitatory neuron model 67 and a pair of inhibitory neuron models 65 and 66. Each of the inhibitory neuron models 65 and 66 and the excitatory neuron model 67 is formed from a unit of 5 × 5 neuron models. The unit of each neuron model is a circuit whose input / output characteristics follow the Hodgkin-Huxley equation.

The pair of olfactory cell models 72 and 73 serve as input neuron models as a pair of inhibitory neuron models 6.
5 and 66 are connected. The olfactory cell models 72 and 73 are also a set of neuron models according to the Hodgkin-Huxley equation (neuron model), and are connected to a pair of inhibitory neuron models 65 and 66 via excitatory synapses 74 and 75, respectively.

This circuit C also shows a steady operation, which is a periodic oscillation in a closed loop, and an operation in response to an input signal. First, with respect to the steady operation, when the inhibitory neuron model 66 was stimulated with a potential, the autonomic oscillation was exhibited without applying a potential stimulus thereafter, and it was found that the output neuron model oscillated at a steady period.

Next, in order to input a signal to the circuit of the neuron model, both input neuron models of the olfactory cell models 72 and 73 were simultaneously subjected to potential stimulation. Potential stimulation intervals were 10,000, 11000, 12000, 130
00, 14000, 15000 steps, 250
It was given 10 times from the 000th step. 15000 intervals
In the case of the step, no change was observed in the oscillation cycle of the output neuron model. In the case of 10000 to 12000, the period changed to a period synchronized with twice the period of the input signal.
When the input of the signal was interrupted, it returned to the original cycle. 1300
At the 0-step interval, a period shift from the steady vibration was observed, but it did not reach twice the input signal and became out of synchronization. When an input signal was given at an interval of 14000 steps, synchronization to twice the period of the input signal and the original stationary period were alternately observed. However, in this case as well, when the cycle of the signal was interrupted, it returned to the original cycle. In addition, FIG.
FIG. 19 is a response waveform diagram of the output neuron model of the circuit C when a potential stimulus is applied at step intervals.
FIG. 20 is a response waveform diagram of an output neuron model of the circuit C when a potential stimulus is applied at step intervals.
FIG. 9 is a response waveform diagram of an output neuron model of a circuit C when a potential stimulus is applied at step intervals. FIG.
FIG. 22 is a response waveform diagram of an output neuron model of the circuit C when a potential stimulus is given at 00 step intervals.
FIG. 23 is a response waveform diagram of the output neuron model of the circuit C when a potential stimulus is given at 00 step intervals.
FIG. 9 is a response waveform diagram of an output neuron model of the circuit C when a potential stimulus is applied at 00 step intervals.

As described above, the prepared circuit C is 1500
It did not respond to the input signal at the 0-step interval, but responded to the input signal at the step interval from 10,000 to 13000. For an input signal at 14000 step intervals, a non-response type and a tuned response to the input signal are shown as an intermediate type. Selective response, non-response, and intermediate response to the signal interval,
It changed to one of its three phases. Furthermore, when the input signal was interrupted, it returned to the original cycle autonomously. In addition to having selectivity with respect to the interval between input signals, it has been found that the filter functions as an active filter that performs autonomous return to the original state when the input signal is interrupted. Further, from the oscillation waveform of the circuit C, if the inhibitory synaptic connections shown in the circuits A and B do not have the basic structure in which the inhibitory synaptic connections are mutually formed, the inhibitory synaptic connections are functionally formed.
It has been found that a similar filter function can be provided. It turns out that the input signal can be identified by the change to three phases,
It has been found that the configuration of the circuit C can function as an active filter.

As shown in FIG. 6, the circuit D has a basic circuit configuration in which three neuron models are arranged in a ring and connected in one direction by inhibitory synapses. The neuron forming the closed loop is a suppressive neuron model 77.
, An inhibitory neuron model 78, and an inhibitory neuron model 79. These are three inhibitory synapses 8
0, 81, and 82, and the signal transmission direction is counterclockwise. Each inhibitory neuron model 7
7, 78 and 79 are each formed from 5 × 5 neuron model units. The unit of each neuron model is a circuit whose input / output characteristics follow the Hodgkin-Huxley equation.

An excitatory neuron model 83 is connected to the inhibitory neuron model 77 and the inhibitory neuron model 79 forming a closed loop. The connection between the excitatory neuron model 83 and the inhibitory neuron model 77 and the inhibitory neuron model 79 is formed by an excitatory synapse 8.
4, 85. Olfactory cell models 88, 89, and 87 are connected as input neurons to the inhibitory neuron model 77, the inhibitory neuron model 78, and the inhibitory neuron model 79 that form a closed loop, respectively. The olfactory cell models 88, 89, and 87 are also a set of neuron models according to the Hodgkin-Huxley equation (neuron model), and each excitatory synapse 90,
It connects via 91,86.

It has been found that this circuit D also exhibits a steady operation, which is a periodic oscillation in a closed loop, and an operation in response to an input signal. First, regarding the steady operation, when the inhibitory neuron model 78 was stimulated with a potential,
The output neuron model showed an autonomous oscillation without applying a potential stimulus, and it was found that the output neuron model oscillated at a steady period with a slight fluctuation.

Next, in order to input a signal to the circuit of the neuron model, potential stimuli were simultaneously applied to all of the input neuron models of the olfactory cell models 88, 89, and 87. The intervals of the potential stimulation are 10,000, 11000, 12000,
13000, 14000, 15000 steps,
It was given 10 times from the 250,000th step. 14 intervals
In the case of 000 steps and 15000 steps, no change was found in the oscillation cycle of the output neuron model. In the case of 10,000 to 13000, the period changed to a period synchronized with twice the period of the input signal. When the input of the signal was interrupted, it returned to the original cycle. FIG. 24 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus is applied at an interval of 10000 steps, and FIG. 25 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus is applied at an interval of 11000 steps. FIG. 26 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus is applied at an interval of 12000 steps. FIG. 27 shows 1300
FIG. 28 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus is applied at an interval of 0 steps.
FIG. 29 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus is applied at 0 step intervals.
FIG. 9 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus is applied at an interval of 0 steps.

As described above, the prepared circuit D is 1000
In response to an input signal with a step interval from 0 to 13000, an output signal tuned to it was shown. 14000 and 1
No response was shown for an input signal at 5000 step intervals. Circuit D has a selectivity of response and intermediate response to the signal interval, and has changed to one of its two phases. Furthermore, when the input signal was interrupted, it returned to the original cycle autonomously. In addition to having selectivity for the interval between input signals, it also functioned as an active filter that autonomously returned to the original state when the input signal was interrupted. It has been found from the configuration of the circuit D that the active filter function similar to the reciprocal inhibitory synapse connection can be realized by the annular inhibitory synapse connection.

When signals are input at 15000 step intervals, the olfactory cell models 88 and 8 which are three input synapses
FIG. 30 shows the results when the potential stimuli to 9, 87 were shifted in order of 5000 steps. When a signal is input, the period of the vibration is equally spaced, and even if the input is given at the same step interval, the response result can be changed depending on the way of inputting the signal to the input neuron model. It has been found that the interruption of the signal returns to the original period of vibration. In the circuit D, it was possible to realize an active filter that changes into two phases depending on the difference in the way of giving an input signal. It has been found that a change to three phases is possible if the way of inputting the signal is included, whereby the input signal can be identified.

The circuit E, as shown in FIG. 7, forms a closed loop by arranging three neuron models in a ring and connecting the inhibitory synapses and the excitatory synapses. The neuron models constituting the closed loop are an inhibitory neuron model 91, an inhibitory neuron model 92, and an excitatory neuron model 79. Inhibitory neuron model 91
Between the inhibitory neuron model 92 and the inhibitory synapse 9
4, the inhibitory neuron model 91 and the excitatory neuron model 93 are connected by an inhibitory synapse 96, and the excitatory neuron model 93 and the inhibitory neuron model 92 are connected by an excitatory synapse 95. .
Each of the inhibitory neuron models 91 and 92 and the excitatory neuron model 93 are formed from 5 × 5 neuron model units. The unit of each neuron model is a circuit whose input / output characteristics follow the Hodgkin-Huxley equation.

A pair of olfactory cell models 97 and 98 are inhibitory neuron models 91 and 92 as input neuron models.
It is connected to the. The olfactory cell models 97 and 98 are also a set of neuron models according to the Hodgkin-Huxley equation (neuron model), and the inhibitory neuron models 91 and 9 via excitatory synapses 99 and 100, respectively.
2

Also in this circuit E, a steady operation as a periodic oscillation in a closed loop and an operation in response to an input signal are shown. First, as for the steady operation, when the inhibitory neuron model 92 was stimulated with a potential, the autonomic oscillation was shown without applying the potential stimulus thereafter, and it was found that the output neuron model oscillated at a regular cycle.

Next, in order to input a signal to the circuit of the neuron model, potential stimulation was simultaneously applied to both the input neuron models of the olfactory cell models 97 and 98. Potential stimulation intervals were 10,000, 11000, 12000, 130
00, 14000, 15000 steps, 250
It was given 10 times from the 000th step. 12000 spacing
In the case of steps and 13000 steps, the cycle of the output neuron model changed to a cycle synchronized with twice the cycle of the input signal. When the input of the signal was interrupted, it returned to the original cycle. When an input signal was applied at 14000 and 15000 step intervals, vibrations with a long cycle and a short cycle were alternately repeated. When the input of the signal was interrupted, it returned to the original cycle.
In the case of the 10000 step interval, the period changes to a period synchronized with twice the period of the input signal, and even if the input of the signal is interrupted, the vibration of the period twice as long as the period of the input signal is maintained. FIG. 31 is a response waveform diagram of an output neuron model of the circuit E when a potential stimulus is applied at an interval of 10000 steps, and FIG. 33 is a response waveform diagram of an output neuron model of the circuit E when a potential stimulus is applied at an interval of 11000 steps. FIG. 34 is a response waveform diagram of an output neuron model of the circuit E when a potential stimulus is applied at an interval of 12000 steps. FIG. 35 is a response waveform diagram of the output neuron model of the circuit E when a potential stimulus is applied at 13000 step intervals. FIG. 36 is a response waveform of the output neuron model of the circuit E when a potential stimulus is applied at 14000 step intervals. FIG. 37 is a response waveform diagram of the output neuron model of the circuit E when a potential stimulus is applied at 15000 step intervals.

In the circuit E manufactured as described above, 1000
In response to an input signal with a step interval from 0 to 13000, an output signal tuned to it was shown. 14000 and 1
In response to an input signal with a 5000-step interval, a response was made in a long cycle and a short cycle, indicating asynchronous. It has tuned and untuned selectivity for signal spacing and changes to one of two phases. Further, in the case of the circuit E, when the input signal was interrupted, the circuit E returned to the original cycle autonomously except at intervals of 10,000 steps. In addition to having selectivity for the interval between input signals, it also functioned as an active filter that autonomously returned to the original state when the input signal was interrupted. 10,000
At the step interval, the oscillation at half the cycle of the original oscillation was maintained and a flip-flop response was shown. Note that, in this specification, the flip-flop response refers to an operation that can shift from one phase state to another phase state in two phases and maintain the destination phase state.

FIG. 32 shows the result when an input signal is applied at an interval of 10000 steps using the circuit E and then an input signal is applied again at an interval of 15000 steps. The vibration changed to the same as the 15000 step interval shown above.
In this circuit E, the flip-flop response could be stopped.

FIG. 38 shows the result when the potential stimulus to the two input neuron models, the olfactory cell models 97 and 98, is shifted by 5000 steps when a signal is input at an interval of 15000 steps. When a signal is being input, the output neuron model showed oscillations at regular intervals without fluctuations, but this was different from the case where the input signal was given at the same time at 15000 steps as described above. It has been found that the response result can be changed depending on the way of inputting the signal to the input neuron model even when the input is given at the same step interval. Also in this case, the oscillation of the original cycle is restored due to the interruption of the input signal, and the two-phase change is realized by the way of applying the input signal.

As described above, even in a circular circuit in which the coupling of excitatory synapses is added to the coupling of inhibitory synapses, the same autonomous oscillation and active filter function as those of a circular circuit formed by coupling of inhibitory synapses can be confirmed. It has been found that the degree of freedom can be increased. It has been found that in the circuit E, by changing the interval and manner of the input, a change to three phases is realized, whereby the input signal can be identified.

[0060]

According to the above-described neuron model circuit of the present invention, the neuron model circuit enables autonomous oscillation using a neuron model according to the Hodgkin-Huxley equation, and the neuron model circuit network and flip-flop circuit And information processing methods such as an active filter. If the circuit of the neuron model of the present invention is used, a difference between input signals can be identified by a change in a cycle of vibration after information processing. For this reason, by combining with sensory sensors such as sight, hearing, touch, taste, and smell, it is possible to discriminate the difference in information sensed by the sensor by the difference in the vibration cycle of the sensor output.

The circuit of the neuron model of the present invention can be easily designed by combining the units of the neuron model, and the size can be easily and easily increased. If the input signal is incompatible with the circuit, the circuit of the neuron model of the present invention does not process the signal, but processes only the signal that is compatible with the circuit. Is robust and can process only necessary information.

Further, in the circuit of the neuron model of the present invention, since the difference between the period of the oscillation of the input signal and the period of the autonomous oscillation is detected, the change of the input signal can be processed with high sensitivity. In addition, since the input signal itself triggers information processing, a separate control signal is not required, and a circuit therefor is not required. Therefore, cost reduction and size reduction can be realized, and automatic operation is also possible. If the input signal is interrupted, the circuit can be designed so as to automatically return to the original autonomous vibration, or it can be made not to return, and the degree of freedom in circuit design is large. Further, since the basic structure and algorithm of the circuit of the neuron model of the present invention are simple, the cost of the device can be reduced.

[Brief description of the drawings]

FIG. 1 is a circuit diagram showing an embodiment of a circuit of a neuron model of the present invention.

FIG. 2 is a circuit diagram showing an embodiment of a neuron model circuit network using the neuron model circuit of the present invention.

FIG. 3 is a circuit diagram showing a circuit A according to the embodiment of the present invention.

FIG. 4 is a circuit diagram showing a circuit B according to the embodiment of the present invention.

FIG. 5 is a circuit diagram showing a circuit C according to the embodiment of the present invention.

FIG. 6 is a circuit diagram showing a circuit D according to the embodiment of the present invention.

FIG. 7 is a circuit diagram showing a circuit E according to the embodiment of the present invention.

FIG. 8 is a response waveform diagram of an output neuron model of the circuit A when a potential stimulus is applied ten times at 3000 step intervals.

FIG. 9 is a response waveform diagram of an output neuron model of the circuit A when a potential stimulus is applied ten times at 5000 step intervals.

FIG. 10 is a response waveform diagram of an output neuron model of the circuit A when a potential stimulus is applied ten times at 8000 step intervals.

FIG. 11 is a response waveform diagram of an output neuron model of the circuit A when a potential stimulus is applied ten times at an interval of 10,000 steps.

FIG. 12 is a response waveform diagram of an output neuron model of the circuit A when a potential stimulus is applied 10 times at 12000 step intervals.

FIG. 13 is a response waveform diagram of an output neuron model of the circuit A when a potential stimulus is applied ten times at an interval of 14000 steps.

FIG. 14 is a response waveform diagram of an output neuron model of the circuit A when a potential stimulus is applied ten times at an interval of 20,000 steps.

FIG. 15 is a response waveform diagram of an output neuron model of the circuit B when a potential stimulus is applied ten times at an interval of 10,000 steps.

FIG. 16 is a response waveform diagram of an output neuron model of the circuit B when a potential stimulus is applied 10 times at 12000 step intervals.

FIG. 17 is a response waveform diagram of an output neuron model of the circuit B when a potential stimulus is applied ten times at an interval of 20,000 steps.

FIG. 18 is a response waveform diagram of an output neuron model of the circuit C when a potential stimulus is applied ten times at an interval of 10,000 steps.

FIG. 19 is a response waveform diagram of an output neuron model of the circuit C when a potential stimulus is applied ten times at 11000 step intervals.

FIG. 20 is a response waveform diagram of an output neuron model of the circuit C when a potential stimulus is applied ten times at 12000 step intervals.

FIG. 21 is a response waveform diagram of an output neuron model of the circuit C when a potential stimulus is applied ten times at 13000 step intervals.

FIG. 22 is a response waveform diagram of an output neuron model of the circuit C when a potential stimulus is applied ten times at 14000 step intervals.

FIG. 23 is a response waveform diagram of an output neuron model of the circuit C when a potential stimulus is applied ten times at 15000 step intervals.

FIG. 24 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus is applied ten times at 10,000 step intervals.

FIG. 25 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus is applied ten times at 11000 step intervals.

FIG. 26 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus is applied ten times at 12000 step intervals.

FIG. 27 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus is applied 10 times at an interval of 13000 steps.

FIG. 28 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus is applied ten times at an interval of 14000 steps.

FIG. 29 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus is applied ten times at 15000 step intervals.

FIG. 30 is a response waveform diagram of an output neuron model of the circuit D when a potential stimulus to an olfactory cell model as three input synapses is shifted by 5000 steps when a signal is input at an interval of 15000 steps.

FIG. 31 is a response waveform diagram of an output neuron model of the circuit E when a potential stimulus is applied ten times at 10,000 step intervals.

FIG. 32 is a response waveform diagram of an output neuron model of the circuit E in a case where an input signal is supplied at an interval of 15000 steps and an input signal is supplied again at an interval of 15000 steps.

FIG. 33 is a response waveform diagram of an output neuron model of the circuit E when a potential stimulus is applied ten times at 11000 step intervals.

FIG. 34 is a response waveform diagram of an output neuron model of the circuit E when a potential stimulus is applied 10 times at an interval of 12000 steps.

FIG. 35 is a response waveform diagram of an output neuron model of the circuit E when a potential stimulus is applied 10 times at an interval of 13000 steps.

FIG. 36 is a response waveform diagram of an output neuron model of the circuit E when a potential stimulus is applied ten times at an interval of 14000 steps.

FIG. 37 is a response waveform diagram of an output neuron model of the circuit E when a potential stimulus is applied ten times at 15000 step intervals.

FIG. 38 is a response waveform diagram of an output neuron model of the circuit E when a potential stimulus to the olfactory cell model as two input neurons is shifted by 5000 steps when a signal is input at an interval of 15000 steps.

FIG. 39 is a circuit diagram of an oscillation circuit including a pair of a pair of inhibitory neuron models.

FIG. 40 is a circuit diagram of an oscillation circuit that arranges five inhibitory neuron models in a ring shape and sequentially transmits potential fluctuations.

FIG. 41 is a block diagram showing an example of a circuit when the inhibitory neuron model is configured by an amplifier and a buffer.

[Explanation of symbols]

1 Closed loop 2 Unit of neuron model 11, 12, 20-27, 41, 42, 53, 54, 6
5, 66, 77, 78, 79, 91, 92 Inhibitory neuron model 13, 31, 32, 33, 45, 57, 67, 83, 9
3 Excitatory neuron model 14, 15, 46, 47, 60, 61, 72, 73, 8
8,89,99,100 Olfactory cell model

Claims (30)

[Claims] 1. An input unit for receiving an input signal, and a closed loop connected to the input unit and having input / output characteristics consisting of a plurality of neuron models according to the Hodgkin-Huxley equation and transmitting a jump after suppression. Neuron model circuit. 2. The neuron model circuit according to claim 1, wherein said closed loop includes a plurality of inhibitory neuron models. 3. The neuron model circuit according to claim 2, wherein said inhibitory neuron model is connected via inhibitory synapses. 4. The neuron model circuit according to claim 1, wherein said closed loop is formed by combining a combination of an inhibitory neuron model and an excitatory neuron model. 5. The circuit of claim 4, wherein said excitatory neuron model is coupled to said inhibitory neuron model by an excitatory synapse. 6. The neuron model circuit according to claim 1, wherein a predetermined potential stimulus is applied to said input unit. 7. The neuron model circuit according to claim 6, wherein said potential stimulus is periodic. 8. The required electric potential stimulus includes olfactory, visual, auditory,
2. The neuron model circuit according to claim 1, wherein the signal is a signal corresponding to any one of taste, touch, memory, cognition, and movement, and a combination thereof. 9. The neuron model circuit according to claim 1, wherein the circuit of the neuron model is configured using an arithmetic processing configuration by computer software or a semiconductor device. 10. A plurality of neuron model circuits each having an input unit for receiving an input signal and a closed loop connected to the input unit and transmitting a jump after suppression, and connecting these neuron model circuits. A network of neuron model circuits, characterized in that the information is transmitted through the network. 11. The network according to claim 10, wherein the closed loop comprises a plurality of neuron models whose input / output characteristics follow the Hodgkin-Huxley equation. 12. An input unit for receiving an input signal, and a closed loop connected to the input unit and configured to transmit a jump after suppression, comprising a plurality of neuron models having input / output characteristics according to Hodgkin-Huxley equation. A flip-flop response to the flip-flop circuit. 13. The closed loop comprises a plurality of inhibitory neuron models.
A flip-flop circuit as described. 14. The method according to claim 13, wherein the inhibitory neuron model is connected via inhibitory synapses.
Flip-flop circuit as described 15. The flip-flop circuit according to claim 12, wherein said closed loop is formed by combining a combination of an inhibitory neuron model and an excitatory neuron model. 16. The flip-flop circuit according to claim 15, wherein said excitatory neuron model is coupled to said inhibitory neuron model by an excitatory synapse. 17. The flip-flop circuit according to claim 12, wherein a predetermined potential stimulus is applied to said input section. 18. The flip-flop circuit according to claim 17, wherein said potential stimulus is periodic. 19. The flip-flop according to claim 17, wherein the required potential stimulus is a signal corresponding to any one of smell, sight, hearing, taste, touch, memory, cognition, and movement, and a combination thereof. Circuit. 20. An input unit for receiving an input signal, comprising a plurality of neuron models whose input / output characteristics conform to the Hodgkin-Huxley equation, the input unit having a closed loop connected to the input unit and transmitting a jump after suppression. An active filter characterized in that different output signals can be obtained according to the following. 21. The closed loop comprising a plurality of inhibitory neuron models.
Active filter as described. 22. The method according to claim 21, wherein the inhibitory neuron models are connected via inhibitory synapses.
Active filter as described. 23. The active filter according to claim 20, wherein said closed loop comprises a combination of an inhibitory neuron model and an excitatory neuron model. 24. The active filter according to claim 23, wherein the excitatory neuron model is connected to the inhibitory neuron model by an excitatory synapse. 25. The active filter according to claim 20, wherein a predetermined potential stimulus is applied to said input section. 26. The active filter according to claim 25, wherein said potential stimulus is periodic. 27. The active device according to claim 25, wherein the required potential stimulus is a signal corresponding to any one of smell, sight, hearing, taste, touch, memory, cognition, and movement, and a combination thereof. filter. 28. A required input signal is input to an input unit, and a signal including a post-suppression jump as a closed-loop signal transmission composed of a plurality of neuron models whose input / output characteristics follow the Hodgkin-Huxley equation and connected to the input unit. Let me communicate
An information processing method comprising extracting an output signal from a part of the closed loop. 29. The closed loop comprising a plurality of inhibitory neuron models.
The information processing method described. 30. The information processing method according to claim 28, wherein said input signal is a signal corresponding to any one of smell, sight, hearing, taste, touch, memory, cognition, and movement, and a combination thereof. .