Artificial neural network and artificial intelligence model based on bionics

Abstract

The invention relates to the technical field of artificial intelligence, in particular to a bionics-based artificial neuron network and artificial intelligence model, which comprises an artificial neuron, and is composed of a signal processor, a signal transceiver and an energy module which are positioned on the inner layer, and a plurality of magnets and magnet fixing frames which are positioned on the outer layer, the signal processor is in communication connection with the signal receiving and transmitting device through electric signals, the adjacent magnets are connected through the magnet fixing frames, and the energy module is used for providing power support for artificial neurons; the artificial neuron network is composed of a shell layer used for packaging the artificial neurons, a large number of the artificial neurons located in the shell layer and a transparent or semitransparent film used for carrying out region segmentation on the large number of the artificial neurons; according to the invention, nerve impulses transmitted by neurons in a real central nervous system can be simulated, and the space deduction capability and the deduction capability of a real physical process are realized.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular, to an artificial neuron network and an artificial intelligence model based on bionics. Background Art

[0002] Most popular artificial neuron networks are based on chips composed of semiconductors and their data processing, and there is currently little practice in imitating biological neurons in terms of biological and physical mechanisms.

[0003] Since artificial intelligence that has achieved data processing capabilities through data training often requires high energy consumption, semiconductor-based chips, and a large amount of data, and language-based LLMs are generally not considered to have the ability of spatial deduction and the ability to deduce real physical processes; so far, the expectation for process-oriented artificial intelligence is relatively low.

[0004] Currently popular generative artificial intelligence technologies based on neuron networks are based on virtual neurons determined by programs in integrated circuits and are trained by a result-oriented training process based on a large amount of existing data. This method can obtain an artificial intelligence with stable output, but there is still a need for a process-oriented artificial intelligence trained by real physical processes for scientific research and various industries. Summary of the Invention

[0005] (1) Object of the Invention

[0006] In order to solve the problems in the above background art, the present invention provides an artificial neuron network and an artificial intelligence model based on bionics.

[0007] (2) Technical Solution

[0008] To solve the above problems, the first aspect of the present invention provides an artificial neuron network based on bionics, including:

[0009] An artificial neuron, which is composed of a signal processor, a signal transceiver device, an energy module located inside the layer, and a plurality of magnets and magnet fixing frames located outside the layer. The signal processor and the signal transceiver device establish a communication connection through electrical signals. Adjacent magnets are connected through the magnet fixing frames, and the energy module is used to provide power support for the artificial neuron;

[0010] An artificial neuron network, which is composed of a shell layer for encapsulating the artificial neurons, a large number of the artificial neurons located inside the shell layer, and a transparent or semi-transparent membrane for regionally dividing the large number of the artificial neurons; wherein,

[0011] The artificial neurons are classified according to the simulation of the functions of real neurons, and different types of artificial neurons can set different signal transmission protocols;

[0012] The output of the artificial neural network is realized according to the different distributions of the artificial neurons at different positions and different signal transmission mechanisms.

[0013] Preferably, in the signal transmission, a band with a large difference in wavelength from the size of the artificial neuron is selected for signal communication.

[0014] Preferably, an amplifier circuit, a frequency conversion circuit, and an arithmetic circuit are provided in the artificial neuron to implement the signal transmission criterion between artificial neurons.

[0015] Preferably, the signal transceiver device includes a signal transmitter and a signal receiver, where the signal transmitter and the signal receiver can share an antenna or can be separately configured with an antenna.

[0016] Preferably, the antenna is a printed antenna based on an integrated circuit, or

[0017] a thin-film antenna manufactured by chemical and / or physical deposition methods and laser etching methods.

[0018] Preferably, the signal transmission protocol between the artificial neurons is determined by iteration, and the formula is as follows:

[0019] where N represents the number of input signals of the opposite-side antenna mixed by one antenna, M represents the number of neurons of the same protocol, β is a conversion coefficient, and W represents the total power of the artificial neuron network;

[0020] η = (w × υ) / Wmax, where υ represents the number of opposite-side antennas supplied by one antenna. If the received signal strength is greater than the threshold T, then:

[0021] η = ((w × υ) + T × q) / Wmax. Wmax refers to the maximum power allowed by the antenna, T × q represents reflection, q is the occupancy rate, and it conforms to a normal distribution.

[0022] In a second aspect of the present invention, an artificial intelligence model is proposed, which includes the above-mentioned artificial neuron network based on bionics, and further includes an action execution device, a tactile input system, and a frequency conversion device. The input of the action execution device includes an analog-to-digital signal conversion circuit, a rectification circuit, a filtering circuit, an amplifier circuit, or a combination of at least any two of them. The signal is transmitted to the action execution device through an optical fiber and a transmission line;

[0023] The tactile input system generates a tactile signal by a circuit composed of n piezoresistors, and the artificial neuron in the encapsulation shell layer receives the tactile signal;

[0024] The frequency conversion device is disposed within the artificial neuron network and is used to change the frequency at which action potentials are generated by the artificial neurons, so as to achieve information processing and transmission by the artificial neurons.

[0025] Preferably, it further includes an imaging device, which is composed of an imaging system formed by lenses. The opening and closing of the lens of the imaging device, the adjustment of the focal length, the exposure intensity, the polarization characteristics, and the input wavelength range are determined by the model output. The artificial neurons adjacent to the imaging device have different spatial densities and different signal transmission protocols, and their positions are constrained by a membrane that is transparent to electromagnetic waves within the communication protocol range.

[0026] Preferably, the imaging device adopts the optical film and Fresnel lens technology in virtual reality AR glasses, and the signal emission and reception cone angles of a single artificial neuron are defined through an antenna or by adding a reflection device.

[0027] Preferably, it further includes a robot body structure driven by a servo motor, which is the model output.

[0028] The above technical solution of the present invention has the following beneficial technical effects:

[0029] The present invention can simulate the transmission of nerve impulses by neurons in the real central nervous system, has the ability of spatial deduction and the ability to deduce real physical processes, and has relatively lower energy consumption compared to artificial intelligence that obtains data processing capabilities through data training. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a schematic diagram of the structure of the artificial neuron network of the artificial intelligence in the present invention;

[0031] Figure 2 It is a schematic diagram of the structure of the artificial neuron in the present invention;

[0032] Figure 3 It is a schematic diagram of the amplifier circuit in the present invention;

[0033] Figure 4 It is a schematic diagram of the optoelectronic conversion circuit in the present invention;

[0034] Figure 5 It is a schematic diagram of the tactile device composed of a piezoresistor at the fingertip of the robot in an embodiment of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0035] To make the objectives, technical solutions, and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below in conjunction with the specific embodiments and with reference to the accompanying drawings. It should be understood that these descriptions are exemplary and are not intended to limit the scope of the present invention. In addition, in the following description, the descriptions of well-known structures and technologies are omitted to avoid unnecessarily confusing the concepts of the present invention.

[0036] The artificial intelligence involved in the present invention realizes output by using different distributions and different signal transmission mechanisms of artificial neurons 3 near the shell layer. This set of mechanisms mimics the signal transmission mechanism of the cerebral cortex of real vertebrates. This set of mechanisms ensures that the action execution device and other input adjustment mechanisms relying on output signals are based on the emission signals of neurons 3 near the shell layer. The mechanisms include setting the signal transmission characteristics of neurons near the shell layer, the reflection projection and refraction characteristics of the optical film, and the optical characteristics of the medium in which the neurons are immersed.

[0037] Reference Figures 1-4 , To solve the above problems, the present application adopts the following technical solutions:

[0038] An artificial neuron network based on bionics proposed in the first aspect of the present invention has an artificial neuron order of magnitude between one million and ten million. The inner surface of its package is covered with optical signal processing devices to process the optical or radio frequency signal output of the artificial neuron units closest to the package shell layer, and mainly uses the emission signals of the artificial neurons near the package shell layer to output action signals;

[0039] Including:

[0040] Artificial neurons, which are composed of a signal processor located in the inner layer, a signal transceiver device, an energy module, and a plurality of magnets and magnet fixing frames located in the outer layer. The signal processor and the signal transceiver device establish a communication connection through electrical signals. Adjacent magnets are connected through the magnet fixing frames. The energy module is used to provide power support for the artificial neurons; an artificial neuron network, which is composed of a shell layer for encapsulating the artificial neurons, a large number of the artificial neurons located inside the shell layer, and a transparent or semi-transparent film for regionally dividing the large number of the artificial neurons. Among them, the artificial neurons are classified according to the simulation of the functions of real neurons, and different types of artificial neurons can set different signal transmission protocols. The output of the artificial neural network is realized according to the different distributions of the artificial neurons in different positions and different signal transmission mechanisms.

[0041] Specifically, an artificial neuron consists of a magnet 5 and its fixed frame 7, a signal transmitter 8, a signal receiver 9, a signal processor 10, and an energy module (not shown here). Some construction methods may involve devices for transparency adjustment, and the prior art can provide solutions for adjusting the light permeability of neurons. The signal transmitting and receiving devices can share an antenna. Additionally, an optical film can be used to construct a shell layer that is non-penetrating or has a low penetration rate from the inside out to control the input and output of individual neurons. Omnidirectional and unidirectional antennas can also be used as receiving and transmitting antennas to achieve a simpler structure, and intelligent antennas can be set up to reduce the number of artificial neurons. Among them, the signal transmitter 8 and the signal receiver 9 can also refer to field effect transistors, microwave tubes, extremely high frequency transceivers, terahertz transmitters / receivers, infrared transmitters / receivers, etc. It can be understood that in order to prevent magnetic interference with signals, a shell for shielding the magnetic field needs to be designed.

[0042] It should be noted that in the signal transmission, a wavelength band with a large difference from the size of the artificial neuron is selected for signal communication, and an amplifier circuit and a frequency conversion circuit are set in the artificial neuron to achieve the signal transmission criterion between artificial neurons. The signal transceiver device includes a signal transmitter and a signal receiver, where the signal transmitter and the signal receiver can share an antenna or can be separately configured with antennas.

[0043] The antenna is a printed antenna based on an integrated circuit, or a thin film antenna manufactured by chemical and / or physical deposition methods and laser etching methods.

[0044] The signal processing method within a single neuron can use an arithmetic circuit based on threshold and convergence characteristics:

[0045] A typical iterative method is to introduce a weight mechanism in circuit design, including addition weight and frequency conversion weight respectively. A variable resistor controlled by weight controls the input quantity of an addition circuit and other various arithmetic circuits. The total output of the circuit is the weighted average of the outputs of each arithmetic type. The quantity assigned to each circuit by the weight mechanism is determined by antenna power, information carrying capacity, input signal characteristics, etc., and the weight mechanism is also controlled by a convergent feedback transfer function.

[0046] The signal transmission protocol between artificial neurons is determined by iteration, and the formula is as follows: Set the mean value of a quantity η representing the antenna occupancy rate to be within a certain range (determined by experiments).

[0047] Where N represents the number of input signals of the opposite - side antennas mixed by an antenna, M represents the number of neurons of the same protocol, β is a conversion coefficient, and W represents the total power of the artificial neuron network;

[0048] η = (w × υ) / Wmax, where υ represents the number of opposite antennas supplied by one antenna. If the received signal strength is greater than the threshold T, then:

[0049] η = ((w × υ) + T × q) / Wmax. Wmax refers to the maximum power allowed by the antenna, T × q represents reflection, q is the occupancy rate, and it conforms to the normal distribution.

[0050] The above artificial neuron units follow the input-output rules determined by iteration and can replace aging neurons. The connections between the above artificial neuron units are determined by a rule that specifies the emission range, direction, frequency, and intensity.

[0051] The above artificial neural network tactile sense is composed of a piezoresistor and an electro-optical signal conversion device. The input optical signal is directly converted into a signal input to adjacent neurons without digital processing, and the conversion undergoes a designed frequency conversion and direction specification process.

[0052] The visual input is an optical signal slightly wider than the visible spectrum or a radio frequency signal of a specified frequency. The artificial neurons that process visual input are slightly different from the neurons in other parts and are more densely distributed than the neurons in other parts.

[0053] Reference Figure 5 , an artificial intelligence model provided by the second aspect of the present invention includes the above artificial neural network based on bionics, and also includes an action execution device, a tactile input system, and a frequency conversion device. The input of the action execution device includes an analog-digital signal conversion circuit and a rectifier circuit, a filter circuit, an amplifier circuit, or at least any combination of the two. The signal is transmitted to the action execution device through an optical fiber and a transmission line; the tactile input system generates a tactile signal by a circuit composed of n piezoresistors, and the artificial neurons in the encapsulation shell layer receive the tactile signal; the frequency conversion device is arranged in the artificial neural network and is used to change the frequency of the action potential generated by the artificial neurons to realize the information processing and transmission of the artificial neurons.

[0054] In Figure 5 , in the upper right frame of the figure is a cross-section of a fingertip. The flexible shell of the robot is cut open to show the internal piezoresistors, which are installed on flexible rods. When an external force acts, it will cause a change in the resistance, thereby affecting the voltage output by the circuit to the central nervous system.

[0055] The above artificial intelligence model is a trainable artificial intelligence. Its input is an optical signal or a radio frequency signal and a tactile signal converted into an optical signal or a radio frequency signal. Its output can be an electrical signal and an optical signal for controlling an electric motor or other mechanical devices, and it has the ability to control these devices.

[0056] Such asFigure 2 As shown, it presents a half-sectional view of the spherical shell equivalent to the skull encapsulating neurons, and the internal artificial neurons 2, 3, and 4. For clear illustration, only a part of the artificial neurons is drawn. The artificial neurons are roughly evenly distributed within the functional area. Transparent or semi-transparent membranes 11 and 12 separate different regions. The action execution device and the tactile input system are omitted. The frequency conversion device is replaced by an incomplete spherical shell 13 for simplicity. Figure 1 What is shown in [Figure] is the simplified distribution of artificial neuron 2 near the optical signal input device (eye). The distribution is different and the spatial position is restricted within the range close to the eye.

[0057] Specifically, as Figure 3 shown, RL is the load resistance, Vo and io are the output voltage and input current respectively, Vs is the signal source voltage, and Rs is the signal source internal resistance. Using existing technologies, it will not be elaborated here.

[0058] The imaging device consists of an imaging system 1 composed of lenses. The opening and closing of the lens, the adjustment of the focal length, the exposure intensity, the polarization characteristics, the input wavelength range, etc. are determined by the output of artificial intelligence. The neurons near the imaging device (eye) have different densities and signal transmission protocols, and their positions are restricted by membranes transparent to electromagnetic waves within the communication protocol range. Since the membrane may block the passage of energy-providing substances, the artificial neurons around the eye require separate energy supply tubes, and the same applies to artificial neuron 3 near the encapsulation shell layer.

[0059] The visible light band signals received by the neurons near the eye are converted into internal signals by a photoelectric conversion tube, a frequency conversion circuit, and a mixing circuit.

[0060] Since different optical signal input methods require different photoelectric conversion devices, there are also many different types of photodiodes and signal processing circuits with various functions and many possibilities. Here, a typical photoelectric conversion circuit is listed (as Figure 4 ).

[0061] It should be noted that the pixel unit circuit of the 3t-type CMOS image sensor includes a photosensitive diode d1 and a CMOS pixel readout circuit. The CMOS pixel readout circuit is a 3t-type pixel circuit, including a reset transistor m1, an amplifying transistor m2, and a selection transistor m3, all of which are NMOS transistors.

[0062] The integrated circuit equivalent to the photoelectric conversion circuit of a digital camera is installed at the end of the optical path of the artificial intelligence eye. They contain a comparable number of photoelectric tubes to a high-precision digital camera. The difference is that the technology involved in the present invention can omit the digital part of the circuit.

[0063] If the wavelength band of the internal communication signal of the artificial central nervous system includes visible light, this optoelectronic conversion device can be omitted. Instead, a special type of ocular neuron is located in the imaging area.

[0064] The tactile signal is generated by a circuit composed of a rich variety of piezoresistors. There are multiple ways to transmit the signal change generated by the resistor to the artificial neurons within the encapsulation housing, including but not limited to optical signals, electromagnetic pulse signals, high-frequency signals transmitted through cables, and signals transmitted through optical fibers. The output signal of the artificial intelligence is transmitted to the action execution device through optical fibers and transmission lines.

[0065] Furthermore, the distribution of neurons and the signal exchange protocol in different regions can be different, mainly in the form of continuous analog signals. If the signal exchange protocol involves light intensity, a filtering device can be used to obtain a more stable and simpler output and simplify the processing device for the output signal. A reward mechanism for energy supply can also be set, that is, the energy supply obtained by the neurons is positively correlated with obtaining a certain output.

[0066] In addition, the induction and output of signals near the cerebral cortex can be achieved through an analog circuit including a power amplifier, a filter, and a resonator. The induction of the output signal can also be carried out using a popular neural network. The servo motor controlled by the analog signal only needs to be slightly modified based on the existing motor.

[0067] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art of this industry should understand that the present invention is not limited by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed.

Claims

1. An artificial neural network based on bionics, characterized in that: include: The artificial neuron is composed of a signal processor located in the inner layer, a signal transceiver, an energy module, and a plurality of magnets and a magnet fixing frame located in the outer layer. The signal processor and the signal transceiver establish a communication connection through electrical signals, and adjacent magnets are connected through the magnet fixing frame. The energy module is used to provide power support for the artificial neuron, and is powered by a photoelectric effect or a biochemical reaction. The operating wavelength of the signal transceiver device is from microwave to visible light; An artificial neuron network is composed of a shell layer for encapsulating the artificial neurons, a large number of the artificial neurons located inside the shell layer, and a transparent or semi-transparent membrane for dividing the large number of artificial neurons into regions; wherein, Artificial neurons are classified by simulating the functions of real neurons. Different types of artificial neurons can be set with different signal transmission protocols; The output of the artificial neural network is achieved according to the different distributions of the artificial neurons at different positions and different signal transmission mechanisms.

2. The artificial neural network based on bionics according to claim 1, characterized in that: In the signal transmission, a wavelength band with a large difference in wavelength from the size of the artificial neuron is selected for signal communication.

3. The artificial neural network based on bionics according to claim 1, characterized in that: An amplification circuit, a frequency conversion circuit and an operation circuit are arranged in the artificial neuron to realize the signal transmission criterion between the artificial neurons.

4. The artificial neural network based on bionics according to claim 1, characterized in that: The signal transceiver device includes a signal transmitter and a signal receiver, wherein the signal transmitter and the signal receiver may share an antenna or may be configured with separate antennas.

5. The artificial neural network based on bionics according to claim 4, characterized in that: The antenna is a printed antenna based on an integrated circuit, or Thin-film antennas fabricated by chemical and / or physical deposition and laser etching.

6. The artificial neural network based on bionics according to claim 1, characterized in that: The signal transmission protocol between the artificial neurons is determined by iteration, and the formula is as follows: Where N represents the number of opposite antenna input signals mixed by an antenna, M represents the number of neurons of the same protocol, β is the conversion coefficient, and W represents the total power of the artificial neural network; η=(w×υ) / Wmax, where υ represents the number of opposite antennas supplied by one antenna. If the received signal strength is greater than the threshold T, then: η=((w×υ)+T×q) / Wmax Wmax refers to the maximum power allowed by the antenna, T×q represents reflection, q is the occupancy rate, and it conforms to the normal distribution.

7. An artificial intelligence model, characterized in that: The invention comprises an artificial neural network based on bionics as described in claim 1-n, and also comprises an action execution device, a tactile input system and a frequency conversion device, wherein the input of the action execution device comprises an analog-to-digital signal conversion circuit and a rectifier circuit, a filter circuit, an amplifier circuit or at least a combination of any two of them, and transmits the signal to the action execution device via an optical fiber and a transmission line; The unit artificial neurons are suspended in a non-Newtonian fluid with a viscosity and density similar to that of the artificial neurons, and the spacing is maintained by the repulsive force of a spring or by the repulsive force between light pressure and magnets; The tactile input system generates a tactile signal by a circuit composed of n varistors, and the tactile signal is received by an artificial neuron in the package shell; The frequency conversion device is arranged in the artificial neuron network and is used to change the frequency of the action potential generated by the artificial neuron to realize the information processing and transmission of the artificial neuron.

8. An artificial intelligence model according to claim 7, characterized in that: It also includes an imaging device, which is composed of an imaging system composed of lenses. The opening and closing of the lens of the imaging device, the adjustment of the focal length, the exposure intensity, the polarization characteristics, and the input wavelength range are determined by the model output. The artificial neurons adjacent to the imaging device have different densities and different signal transmission protocols in space, and their positions are constrained by a film that is transparent to electromagnetic waves within the range of the communication protocol.

9. An artificial intelligence model according to claim 8, characterized in that: The imaging device adopts the optical film and Fresnel lens technology in virtual reality AR glasses, and formulates the signal transmission and reception cone angle of a single artificial neuron through an antenna or adding a reflective device.

10. An artificial intelligence model according to claim 7, characterized in that: The robot body structure driven by servo motors is also included as model output.

Images