JPH08123885A - Neuron functional element - Google Patents

Abstract

PURPOSE: To process input signals in space and time manner by comparing the total sum of the output signals of chaos vibrators transmitted from the plural columns of input transmission lines with a threshold value and successively transmitting outputted signals to the continuously connected chaos vibrators. CONSTITUTION: This element is provided with the plural columns of the input transmission lines M1 for successively adding and transmitting the output of the plural linearly arrayed chaos vibrators whose vibration frequency is changed by input frequency, an addition comparison part M2 for comparing the total sum of the signals transmitted from the respective input transmission lines M1 with a prescribed threshold value and outputting binary signals and an output transmission line M3 for successively transmitting the signals to the cascade connected plural chaos vibrators. Thus, a position where the input signals are supplied and the time are nonlinearly computed, a function for propagating information received by the synapse of a dendrite by a nerve fiber is accurately approximated and the processing of the input signals in the space and time manner is made possible. Also, the function of a nerve cell is accurately approximated and the function of an axon for propagating the information emitted by the nerve cell is accurately approximated.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a nerve cell functional element, and more particularly to a nerve cell functional element that operates like a nerve cell of a living system. In recent years, in the field of biological neuroscience, signal transmission in nerve cells is not limited to electrical signals,
It has been elucidated that generation of molecular gases (NO, CO) and dynamics of intracellular calcium contribute to signal transduction, and neurons capable of bringing excitability and inhibitory property of synaptic part closer to biological system. A model is desired.

[0002]

2. Description of the Related Art FIG. 7 shows a block diagram of a conventional nerve cell functional element. In the figure, it comprises a weighting circuit 1 simulating synaptic connection and a threshold processing circuit 2 simulating post-synaptic cells. The weighting circuit 1 is composed of n multipliers, and the threshold value processing circuit 2 is composed of a summing operation circuit 2A, a comparator 2B, a threshold value setting circuit 2C and an output resistance R.

The function of each input signal is Xi (i is 1 to n).
Weighted by the multiplier of the weighting circuit 1 with respect to
When each i is multiplied, the sum ΣWiXi is calculated by the sum calculation circuit 2A of the threshold value processing circuit 2. The calculation result signal Y is output to the comparator 2B as the compared signal. On the other hand, the threshold setting circuit 2C outputs the threshold signal Yth to the comparator 2B. As a result, the output signal Z0 = 1 / [1 + e ^{−Y + Yth} ] is output from the comparator 2B, and the output signal Z0 appears across the resistor R.

[0004]

The spatiotemporal pattern of the input signal generated on the dendrites in the nerve cells of the biological system is as follows.
Processing is performed by maintaining or changing the input signal efficiency (synapse transmission efficiency) at a specific position on the dendrite. On the other hand, the conventional nerve cell functional element shown in FIG. 7 is capable of processing the incoming input signals X _{1 to} X _{n at} the same time. However, there is a problem that the signal processing of the spatiotemporal pattern of the input signal that changes in time series cannot be performed.

The present invention has been made in view of the above points,
It is an object of the present invention to provide a nerve cell functional element having a function closer to that of a nerve cell of a biological system and capable of performing spatiotemporal signal processing of an input signal.

[0006]

According to a first aspect of the present invention, as shown in FIG. 1, the outputs of a plurality of chaotic oscillators whose vibration frequencies are one-dimensionally arranged and whose vibration frequencies change depending on the input frequency are sequentially added and transmitted. A plurality of columns of input transmission lines M1, and an addition / comparison unit M2 that compares the sum of signals transmitted from the plurality of columns of input transmission lines with a predetermined threshold value and outputs a binary signal.
And an output transmission line M3 for sequentially transmitting the signals supplied from the addition comparison unit to a plurality of chaos oscillators connected in cascade.

According to a second aspect of the invention, in the plurality of one-dimensionally arrayed chaotic oscillators of the input transmission lines in each of the plurality of columns, the vibration value diffuses to both sides of the array with time.
According to a third aspect of the invention, in the plurality of chaos oscillators connected in cascade in the output transmission line, the vibration value is advected from the input side to the output side with time.

[0008]

According to the first aspect of the present invention, since the output of the chaotic oscillator whose frequency changes according to the input frequency in which the input transmission paths of the plurality of columns are arranged one-dimensionally is sequentially transmitted, the position where the input signal is supplied is The function of non-linearly calculating time and the input signal and propagating the information received at the synapse of dendrites in the nerve fiber can be approximated more accurately, and the spatiotemporal signal processing of the input signal becomes possible. Performs threshold processing on the sum of the signals transmitted from each input transmission path, so that the input signal that has propagated through the nerve fibers can be added to accurately perform the thresholding function of the cell body. Since the chaotic oscillators are connected in cascade, the function of the axon that propagates the information emitted by the cell body can be approximated accurately.

According to the second aspect of the present invention, since the oscillation value of the chaotic oscillator in the input transmission line diffuses to both sides of the array with time, the nerve fiber can bidirectionally propagate the signal to the adjacent portion. Can be approximated exactly. In the invention according to claim 3, since the vibration value of the chaotic oscillator of the output transmission line is advected from the input side to the output side with time, the function that the propagation direction of excitement in the axon is away is accurately described. Can be approximated.

[0010]

FIG. 2 shows a block diagram of an embodiment of the nerve cell functional device of the present invention. In the figure, 10 _{1 to} 10 m are input transmission lines (M1 in FIG. 1). Input transmission line 10 ₁ ~
Each of the 10 m is composed of m compartments, that is, the weighting circuit 12 and the adding circuits 14 _{2 to} 14 _n, and the signals coming from the input terminals 11 _{1 to} 11 _n are respectively input to the weighting circuits 12 _{1 to} 12 _n . After being multiplied by a predetermined weighting coefficient and weighted, the sums are obtained by adding at the adders 14 _{2 to} 14 _n . Weighting circuit 12 for each input transmission line
The weighting factors _{1 to} 12 _n are set by the weighting control circuit 15. The above n is, for example, about 50.

In the weighting control circuit 15, n chaotic oscillators that oscillate in a chaotic manner are one-dimensionally arranged for each input transmission line. This chaotic oscillator uses, for example, a logistic function as shown in equation (1). f (x) = 1-wx ² (1) Here, x is an input to the vibrator, and w varies in the range of 1 to 2.0 depending on the input frequency. In the logistic function, the vibration frequency changes by changing the value of w.
It is known that chaos oscillation occurs at> 1.44. Of course, any function other than this function may be used as long as it is a vibrator that causes chaotic oscillation.

N chaotic oscillators represented by the equation (1)
A two-dimensional array is provided, and a connection relation represented by the equation (2) is given between them. _{xj + 1, i} = (1-ε) f (xj _{, i} ) + 0.5ε {f ( _{xj, i-1} ) + f ( _{Xj, i + 1} ) (2) where X _{j, i} represents the i-th chaotic oscillator in the one-dimensional array at time j, and 0 <ε <1, and the larger this value, the better the propagation efficiency.

This equation (2) is, as shown in FIG.
I-th chaos oscillators x _{j + 1} of _{1, i} is the i-th chaos oscillators x _j of the previous point _j, and _i, chaotic oscillator x _j adjacent _{thereto, i-1} and x _j, It means that it is decided based on the vibration of _{i + 1} . That is, as time passes, the vibration of each chaotic oscillator diffuses to the adjacent chaotic oscillator.

[0014] weighting control circuit 15 is provided in the input transmission line units, input transmission path 10 ₁ to the vibration value of s n number of chaotic oscillator husband to change over time as a weighting factor
The weighting circuits 12 _{1 to} 12 _{n of} 10 m each are set. The weighting factor is set every unit time. FIG. 4 shows that in equations (1) and (2), w = 1.9 and ε =
The following shows how the vibration of each chaotic oscillator diffuses with time j when 0.5, i = 50, x _{0, i} = 0.0, and x _0,25 = 0.5.

According to physiological findings, dendrites are formed from input sites (synapses) that receive information transmission by neurotransmitters from other nerve cells and nerve fibers that propagate the resulting postsynaptic potential. Has been done. This nerve fiber is capable of bidirectional signal propagation to adjacent regions. Also, the dendrites can form multiple synapses on their nerve fibers. When a neurotransmitter arrives, it is performed by slightly increasing (excitatory synapse) or decreasing (inhibitory synapse) the input x _{j, i} at the synapse formation site (i). The input transmission lines 10 _{1 to} 10 m approximately represent the role of the dendrites.

Synaptic plasticity with respect to the input signal can be achieved by increasing or decreasing the value of w. For example, in the case of a logistic function, it is known that the vibration period of the nonlinear oscillator increases by 2 times at the boundary of w = 1.44. That is, for synapses that are used with high frequency, it is possible to introduce learning that increases the value of w and increases the vibration frequency. Further, for an input that is not often used, the vibration frequency can be reduced by reducing the value of w.

In FIG. 2, input transmission lines 10 ₁ to 10 _n
The sum total signals output from the final-stage adders 14n are added by the adder 17 to be the compared signal Y, which is supplied to the comparator 18. The comparator 18 compares the threshold value signal Vth supplied from the threshold value setting circuit 19 with the compared signal Y and outputs a signal having a value of 1 when Y ≧ Vth and a value of 0 when Y <Vth. The output signal of the comparator 18 is supplied to the output transmission line 21.

The physiologically known role of cell bodies is to add input signals propagated through nerve fibers of dendrites,
It is to perform threshold processing and output. In order to perform such an operation, the signals propagating through the dendrites may be added and processed by the Heaviside function to calculate the output. That is, x _{j + 1, S} = Φ (Σx _{j, s−1} ) (3) Here, s is a subscript indicating a cell body, s-1 indicates a dendrite compartment immediately before the cell body, and s-1 = n. Further, Σ represents addition for all dendrites that enter the cell body.
Φ is a Heaviside function, which outputs 1 when the sum of inputs is 0 or more and outputs 0 otherwise. That is, the adder 17, the comparator 18, and the threshold value setting circuit 19 correspond to the addition and comparison unit M2 and approximately represent the role of the cell body.

In FIG. 2, the output transmission line 21 (M3 in FIG. 1) has k weighting circuits 22 ₁ connected in cascade.
It is composed of ~ 22k. This weighting circuit 22 ₁
The weighting factors of .about.22k are set by the weighting control circuit 15. In the weight control circuit 15, the output transmission line 21
The k chaotic oscillators that oscillate in a chaotic manner are arranged in a one-dimensional manner corresponding to. This chaotic oscillator uses the logistic function expressed by the equation (1), and gives the coupling relation expressed by the equation (4) between them.

X _{j + 1, i} = (1−γ) f (x _{j, i} ) + γ f (x _{j, i−1} ) ... (4) where x _{j, i} is one dimension at time j It represents the i-th chaotic oscillator in the array, and 0 <γ <1, and the larger this value, the better the propagation efficiency. As shown in FIG. 5, this equation (4) is the i-th chaotic oscillator x at time j + 1.
The vibration of _{j + 1, i} is the i−1 th chaotic oscillator j, i−1 and the i th chaotic oscillator x _{j, i} at the previous time point j.
It means that it is decided based on the vibration of. That is, as time passes, the vibration is transferred from the upstream side where the value of i is small to the downstream side where the value of i is large in the one-dimensional array. The weighting control circuit 15 sets the vibration value of each of the k chaotic oscillators as a weighting coefficient in each of the weighting circuits 22 _{1 to} 22 _k . The weighting factor is set every unit time.

FIG. 6 shows that in the equations (1) and (4), w = 1.
It is shown that the vibration of each chaotic oscillator advects from the left side to the right side with time j when 53, γ = 0.3, i = 50, x _{0, i} = 1 or 0. The basic function of the axon is to propagate the excitement (1 which is the output value of Φ) generated in the cell body so as not to be attenuated. The transmission direction is also one direction away from the cell body. The output transmission line 21 described above approximately represents the role of the axon.

In the higher animal nervous system, myelin often forms a sheath around the axon. It is known that the excitation potential propagation in such a case propagates like jumping in the sheathed portion. In the case of expressing the propagation in such a sheathed axon by this model, for example, in the equation (4), j is not located on the adjacent grid point, but is, for example, the grid j five points ahead.
It may be rewritten so as to propagate to +5.

In this way, the dendrites are approximately represented by the input transmission lines 10 ₁ to 10 _m , and the adder 17 and the comparator 1 are used.
8. The threshold value setting circuit 19 approximates the cell body, and the output transmission path 21 approximates the axon to qualitatively visualize the information processing of the nerve cell. Further, the weighting circuits 12 _{1 to} 12 _{n of} the input transmission lines 10 ₁ to 10 _{m, respectively.}
, And the input signals of the terminals 11 _{1 to} 11 _n are multiplied by different weighting factors depending on the position and time of the incoming compartment. That is, the input signal is non-linearly calculated according to the incoming position and time. Therefore, it is possible to perform spatiotemporal signal processing of the input signal, such as pattern recognition of the time-series pattern of the input signal.

When the nerve cell function element shown in FIG. 2 is applied as a nerve cell function element of the input layer of a neural network consisting of the nerve cell function elements of the input layer, the intermediate layer, and the output layer, the input layer shown in FIG. In the nerve cell functional element having the configuration shown, the input transmission lines 10 ₁ to 10 _m and the output are set by setting constants w, γ, ε, and values for each weighting circuit from the terminal 15a to the weighting control circuit 15. Each of the weighting circuits 12 _{1 to} 12 _n and 22 ₁ to 2 of the transmission line 21
Change the weighting pattern for each 2 _k .

[0025]

According to the first aspect of the present invention, since the output of the Ocass oscillator whose frequency changes according to the input frequency in which the input transmission lines of a plurality of columns are arranged one-dimensionally is sequentially transmitted, the input signal is supplied. It is possible to more accurately approximate the function of propagating the information received at the synapse of the dendrite in the nerve fiber by performing a non-linear operation on the input signal with respect to the input position and time, and spatiotemporal signal processing of the input signal becomes possible, Since the addition comparison unit performs threshold processing on the sum of the signals transmitted from the respective input transmission paths, it is possible to accurately approximate the function of the cell body that performs threshold processing by adding the input signals propagated through the nerve fibers, and, Since the output transmission path has cascaded chaos oscillators, the function of the axon that propagates the information emitted by the cell body can be approximated accurately.

According to the second aspect of the present invention, since the vibration value of the chaotic oscillator in the input transmission line diffuses to both sides of the array with time, the nerve fiber propagates the signal bidirectionally to the adjacent portion. You can approximate exactly what is possible. Also,
According to the invention described in claim 3, since the vibration value of the chaotic oscillator of the output transmission line is advected from the input side to the output side with time, the function that the propagation direction of excitement in the axon is away is accurate. Is very useful in practice.

[Brief description of drawings]

FIG. 1 is a principle diagram of an element of the present invention.

FIG. 2 is a block diagram of an embodiment of the device of the present invention.

FIG. 3 is a diagram for explaining diffusion of vibration of a chaotic oscillator.

FIG. 4 is a diagram showing a state of vibration according to equations (1) and (2).

FIG. 5 is a diagram for explaining advection of vibration of a chaotic oscillator.

FIG. 6 is a diagram showing a state of vibration according to equations (1) and (4).

FIG. 7 is a block diagram of an example of a conventional element.

[Explanation of symbols]

10 ₁ to 10 _m , M1 input transmission line 11 _{1 to} 11 _n input terminal 12 _{1 to} 12 _n , 22 _{1 to} 22 _k weighting circuit 14 _{1 to} 14 _n , 17 adding circuit 15 weighting control circuit 18 comparator 19 threshold setting circuit 21 , M3 Output transmission line M2 Summing comparison unit

Claims (3)

[Claims] 1. A plurality of rows of input transmission lines for sequentially transmitting outputs of a plurality of chaotic oscillators whose vibration frequencies change in accordance with a one-dimensionally arranged input frequency, and a signal transmitted from each of the plurality of rows of input transmission paths. It has an addition comparison unit that outputs a binary signal by comparing the total sum with a predetermined threshold value, and an output transmission line that sequentially transmits the signal supplied from the addition comparison unit to a plurality of cascade-connected chaotic oscillators. A neuron functional element characterized by the above. 2. The nerve cell function according to claim 1, wherein the one-dimensionally arrayed chaotic oscillators of the input transmission lines in each of the plurality of columns have their oscillation values diffused to both sides of the array with time. element. 3. The nerve cell functional element according to claim 1, wherein the chaos oscillators connected in cascade in the output transmission line have a vibration value that is transferred from the input side to the output side with time.