High-dimensional information cognition system and training method therefor

Abstract

Disclosed in the present invention are a high-dimensional information cognition system and a training method therefor. The method comprises: generating n-dimensional geometric figures; constructing a set of m-dimensional spatial projections of the n-dimensional geometric figures; and using the set of m-dimensional spatial projections to train a neural network. In the technical solution, high-dimensional geometric figures and the projections thereof, and time evolution information are used to train a neural system, an intuitive spatial reasoning capability for an n-dimensional world is constructed, and an intelligent agent is promoted to form an intuitive thinking paradigm for processing high-dimensional figures and corresponding time-evolving events, so as to realize the evolution of a high-dimensional information processing capability of the intelligent agent, thereby solving the problem that the natural evolution of existing intelligent agents is unable to surpass the capabilities of high-dimensional spatial cognition and reasoning.

Description

High-Dimensional Information Cognition System and Its Training Method Technical Field

[0001] The present invention relates to the technical fields of intelligent network systems and high-dimensional information processing, belonging to the field of artificial intelligence, and particularly relates to a high-dimensional information processing system and its training method. Background Art

[0002] In the course of biological evolution and the development of artificial intelligence, three-dimensional space perception has always constituted the basic framework of the cognitive system. The existing biological nervous systems have evolved sensory organs such as vision and touch through natural selection, and the information acquisition dimensions thereof are strictly limited within the three-dimensional Euclidean space. Biological sensing mechanisms such as mammalian retinal imaging and insect compound eye structures all reconstruct three-dimensional space information in a two-dimensional projection manner. This evolutionary path has led to the lack of an analytical framework for four-dimensional and higher geometric forms in the construction of biological neural networks, and its information processing topological structure cannot break through the inherent paradigm of three-dimensional tensor operations.

[0003] Furthermore, there are also significant dimensional bottlenecks in the current artificial intelligence training datasets. The underlying data structures of mainstream image databases, three-dimensional point cloud datasets, and multi-modal training data all originate from the sampling and digitization of the three-dimensional physical world. The artificial intelligence programs obtained through training have similar problems to the evolved biological neural networks, that is, they lack an analytical framework for four-dimensional and higher geometric forms, and their information processing topological structures also cannot break through the inherent paradigm of three-dimensional operations.

[0004] The above-mentioned high-dimensional information processing difficulties have severely restricted the evolution speed and level of intelligent systems: in the biological field, the cerebral cortex cannot evolve an intuitive spatial reasoning ability for n-dimensional high dimensions; in the field of artificial intelligence, the existing models' understanding of high-dimensional geometry stays at the symbolic calculation level and lacks the intuitive thinking paradigm when humans process three-dimensional graphics. Therefore, breaking through the intuitive cognitive ability and reasoning ability of high-dimensional space is the key to the evolution of intelligent entities. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a high-dimensional information cognition system and its training method to solve the problem that the natural evolution of current intelligent entities cannot break through the high-dimensional space cognition and reasoning ability.

[0006] The present application provides a training method for a high-dimensional information cognition system, including the following steps;

[0007] Step I: Generate an n-dimensional geometric figure, where n≥4 and n∈N+;

[0008] Step II: Construct an m-dimensional space projection set of the n-dimensional geometric figure, where 2≤m<n and m∈N+;

[0009] Step III: Using the m-dimensional space projection set to train an intelligent network, thereby obtaining an n-dimensional information cognition system.

[0010] The m-dimensional space projection of the n-dimensional geometric figure described in this patent application should regard any face or any partial face or any two or more combinations of the n-dimensional geometric figure as a type of m-dimensional space projection of the n-dimensional geometric figure.

[0011] In another preferred example, in step I, the n-dimensional geometric figure includes one or more n-dimensional geometric figures that evolve along the time dimension and constitute n-dimensional dynamic events, so that in step II, the m-dimensional space projection set with time evolution characteristics is constructed, and then the m-dimensional space projection set is used to train the intelligent network, thereby realizing the training of the intelligent network by n-dimensional dynamic events; preferably, the evolution along the time dimension refers to the change of one or more of the following characteristics of the n-dimensional geometric figure: position, shape, mechanical strength, mass, color, and temperature.

[0012] In another preferred example, multiple associated n-dimensional geometric figures are used for training to realize the training of n-dimensional dynamic events; the multiple associated n-dimensional geometric figures refer to a group of n-dimensional geometric figures frames formed during the evolution of a group of n-dimensional geometric figures along the time dimension in the multi-dimensional space where they are located; preferably, the evolution along the time dimension refers to the change of one or more of the following characteristics of the n-dimensional geometric figures: position, shape, mechanical strength, mass, color, and temperature.

[0013] In another preferred example, in step II, an m-dimensional space projection set of the n-dimensional geometric figure is constructed, including performing a single-modal or multi-modal projection operation on the n-dimensional geometric figure to generate a projection set containing k m-dimensional projections; preferably, k≥C(n,m).

[0014] In another preferred embodiment, the single-mode or multi-mode projection operation includes one or more of the following projection modes:

[0015] Orthogonal projection: dimensionality folding along the direction of standard basis vectors;

[0016] Rotation projection: non-orthogonal projection after rotation transformation in the selected hyperplane;

[0017] Topology-preserving projection: A continuous mapping that preserves specific topological properties of the original n-dimensional geometry.

[0018] In another preferred example, the m-dimensional space projection is a three-dimensional Euclidean space projection, and the m-dimensional space projection set contains no less than n(n-1)(n-2) / 6 three-dimensional projections, each projection corresponding to a different three-dimensional subspace interception method.

[0019] In another preferred example, one or more of brightness, color information, hardness information, temperature information, density information, force information, and size information are further included in the n-dimensional geometric figure and / or its projection.

[0020] In another preferred example, the intelligent network is a biological neural network.

[0021] In another preferred example, the biological neural network includes a network composed of neuron cells. Preferably, the biological neural network is selected from the following group: brain organoids, biological brains, and brains-on-chips, where the brains-on-chips refer to a chip-neuron hybrid system.

[0022] In another preferred example, step III includes: projecting each projection in the m-dimensional space projection set onto different neurons of the biological neural network respectively, activating the neurons, and prompting the activated neurons to be connected to each other, thereby constructing a neural network that maps all or part of the projections in the m-dimensional space projection set, that is, forming the cognition of the n-dimensional geometric figure; preferably, the association relationship between the projections in the m-dimensional space projection set is converted into the association relationship between the projected neuron groups.

[0023] The present application also provides a training method for a high-dimensional information cognition system, including the following steps: training the intelligent network according to the above method, and obtaining the n-dimensional information cognition system, where n≥4 and n∈N+; using the trained n-dimensional information cognition system to generate an n + 1-dimensional geometric figure and the corresponding projection set and dynamic events, and training to obtain an n + 1-dimensional information cognition system. Preferably, the dynamic event is composed of an n + 1-dimensional geometric figure evolving along the time dimension.

[0024] The present application also provides a high-dimensional information cognition system, including:

[0025] A high-dimensional geometry generation unit configured to generate an n-dimensional geometric figure, where n≥4 and n∈N+;

[0026] A projection construction module for constructing an m-dimensional space projection set of the n-dimensional geometric figure, where 2≤m<n and m∈N+, preferably, the m-dimensional space projection set includes k m-dimensional projections, k≥C(n,m);

[0027] A training module for training the intelligent network with the m-dimensional space projection set to obtain an n-dimensional information cognition system.

[0028] In another preferred example, the projection construction module includes one or more of at least the following three projection operators: orthogonal projection operator, rotation projection operator, and topology-preserving projection operator.

[0029] In another preferred example, the n-dimensional geometric figures generated by the high-dimensional geometry generation unit include one or more n-dimensional geometric figures that constitute n-dimensional dynamic events that evolve along the time dimension, so that the projection construction module constructs the spatial projection set with time evolution characteristics, and then the training module trains the intelligent network with the spatial projection set, thereby realizing the training of the intelligent network by n-dimensional dynamic events.

[0030] In another preferred embodiment, the topology preserving projection operator implementation includes:

[0031] Homology group preserver, which ensures that the geometry before and after projection remains isomorphic in the homology group dimension of order p, where p≤m;

[0032] Fiber bundle mapper, decomposes the n-dimensional principal bundle structure into the direct product projection of the m-dimensional basis space and the (nm)-dimensional fiber space.

[0033] In another preferred example, the intelligent network is a biological neural network; preferably, the biological neural network includes a network composed of neuronal cells; preferably, the biological neural network is selected from the following group: brain-like organs, biological brains, and brain-on-chips, and the brain-on-chip refers to a hybrid system of chips and neurons.

[0034] In another preferred example, the biological neural network is connected to the optic nerve system, and the optic nerve system is used to receive the optical signal of the projection information generated by the projection construction module projected by the pattern projection system, and convert the projection information optical signal into a neural signal and transmit it into the biological neural network; preferably, the projection construction module is a three-dimensional imaging geometry, and a biological neural network is arranged around the three-dimensional imaging geometry; more preferably, the three-dimensional imaging geometry is used for the projection of n-dimensional geometric figures in two-dimensional or three-dimensional projection; more preferably, the three-dimensional imaging geometry and the peripheral biological neural network realize the input of the projection image into the biological neural network through the optic nerve system or optogenetics or neural electrode means.

[0035] In another preferred embodiment, the biological neural network is selected from a brain-on-chip, and the projection information of the projection construction module is converted into chip output information of the brain-on-chip to realize the projection of the biological neural network of the brain-on-chip.

[0036] In another preferred embodiment, the biological neural network utilizes optogenetic manipulation to input the optical signal of the projection information of the projection building module.

[0037] The present application also provides an artificial bio-electronic-mechanical intelligent body, comprising a biological neural network for information processing, an encoding system for receiving external event information and encoding the event information into pattern information, and a pattern projection system for projecting the pattern information onto the biological neural network, for externally outputting result information obtained by the biological neural network processing;

[0038] The high-dimensional information cognition system obtained by the biological neural network through the above-mentioned training method, the biological neural network has the ability to recognize and reason about high-dimensional information; the number of dimensions of the high-dimensional information n≥4 and n∈N+; preferably, it also includes an information output system connected to the biological neural network.

[0039] The present application also provides a device for spatially compressing event information and delivering it to a nervous system, comprising an imaging system that encodes the event information into a plurality of patterns, a plurality of visual nervous systems that use the minimum information units encoded in the patterns to regulate the activity of a single or multiple neurons in the nervous system, and a pattern projection system that projects the plurality of patterns encoded by the imaging system onto each of the visual nervous systems. Preferably, the pattern projection system compresses the spatial size of the pattern through re-imaging, thereby increasing the information density within the space containing the compressed pattern and reducing the size of the image block encoding the minimum information unit within the compressed pattern.

[0040] In another preferred embodiment, the imaging system receives event information, divides each characterization parameter of the event information into different dimensions according to the type of information, and encodes the information of each dimension through the imaging system to form an independent pattern; the event information is encoded into several groups of patterns.

[0041] In another preferred embodiment, the pattern projection system projects a plurality of patterns encoding event information to the visual system.

[0042] In another preferred embodiment, the pattern projection system includes several groups of sub-projection systems, each sub-projection system projects a pattern to a corresponding optic nerve system.

[0043] In another preferred embodiment, the optic nerve system is selected from:

[0044] A first photosensitive element independently provided to the neuron, wherein the first photosensitive element generates a nerve impulse under the action of light and transmits the generated nerve impulse to the neuron; or

[0045] The visual nervous system is selected from neurons with photosensitive points set on the surface or inside. The photosensitive points activate the neurons where they are located under the action of light signals. The neurons are defined as photosensitive neurons, that is, the second photosensitive elements.

[0046] In another preferred example, each pattern formed by the imaging system includes several image blocks, and the brightness and darkness of a single image block respectively represent the smallest information unit. The brightness and darkness of each image block in the pattern are used to encode the event information to be projected in the pattern.

[0047] In another preferred example, the brightness and darkness of the image block encoding event information respectively control the activation and deactivation states of the first photosensitive element at the position where the image block is projected, or the brightness and darkness of the image block encoding event information respectively control the activation and deactivation states of the second photosensitive element at the position where the image block is projected, thereby achieving regulation of a single neuron.

[0048] In another preferred embodiment, the sub-projection system reduces the pattern formed by the imaging system and projects it onto the corresponding optic nerve system;

[0049] Preferably, the sub-projection system further includes an optical path system for adjusting the projection position of the pattern to be projected, so that the same sub-projection system can project different patterns to different areas of a certain brain region, thereby expanding the range of brain regions controlled by the projection system;

[0050] Preferably, the sub-projection system includes an image deflection optical path system for adjusting the projection position, so as to realize field scanning of the projected image in different areas of the same brain area. The same sub-projection system projects different patterns to different areas of a large range of brain areas, thereby expanding the area of ​​the brain area regulated by the projection system.

[0051] In another preferred example, the sub-projection system reduces a single image block encoding information in the pattern formed by the imaging system to a size smaller than the size of the first photosensitive element or the second photosensitive element that receives the information, and independently projects information to each first photosensitive element or the second photosensitive element through a single or multiple image blocks encoding information in the projected pattern; thereby, by regulating a single neuron through a single image block, fine regulation of the activity state of each neuron in a large-scale neuronal cluster can be achieved through projected images.

[0052] In another preferred embodiment, the maximum inner diameter of a single image block in the pattern projected by the projection system is in the range of 20 nm to 10 μm.

[0053] In another preferred embodiment, when the optic nerve system is selected from a first photosensitive element independently of neurons, the optic nerve system includes a photoreceptor cell layer for receiving the pattern projected by the sub-projection system and receiving the pattern light signal; after the photoreceptor cells receive the pattern light signal, they activate the neural interneuron cells, and the activated neural interneuron cells further activate the retinal ganglion cells, which project to the nervous system through axons, convert the pattern information into nerve pulses and project them to the neurons of the nervous system; the optic nerve system is selected from neurons with photosensitive points set on the surface or inside, and the photosensitive points are selected from photosensitive proteins.

[0054] In another preferred embodiment, the nervous system is selected from the biological brain, cerebellum, spinal cord, peripheral nerves or brain organs cultured in vitro.

[0055] In another preferred embodiment, the events include frame events and sequential events; the frame events refer to static events; and the sequential events refer to dynamic events formed by the combination of multiple continuous and related frame events.

[0056] In another preferred example, the dimensions for classifying event information include but are not limited to: mechanical, taste, tactile, olfactory, temperature, and light stimulation dimensions of the event, and each dimension information also includes coordinate information of the dimension information.

[0057] This application also provides a method for fine-tuning large-scale neurons.

[0058] Neurons in the nervous system are provided with photosensory points or connected to the first photosensory element; the neurons providing photosensory points are defined as the second photosensory element;

[0059] After the information to be delivered is encoded into a pattern, the spatial size of the pattern is compressed by re-imaging, thereby increasing the information density within the space where the compressed pattern is located and reducing the size of the image block encoding the minimum information unit within the compressed pattern.

[0060] The compressed pattern is projected onto the first photosensitive element or the second photosensitive element to achieve fine control of each neuron in a large-scale neuronal cluster and high-density information transmission to the nervous system.

[0061] This application also provides a method for fine-tuning large-scale neurons.

[0062] Setting up photosensitive elements on neurons,

[0063] After encoding the information to be delivered into pattern information, the spatial size of the pattern is compressed to reduce the single image block encoding the smallest information unit in the pattern to below the neuron size, thereby obtaining a compressed pattern, and

[0064] Projecting the compressed pattern onto a neural system comprising a collection of neurons equipped with the photosensitive element, controlling a single neuron through a single or multiple image blocks encoding information within the projected pattern, and further regulating a single neuron through a single or multiple image blocks encoding information within the projected pattern; ultimately, finely regulating each neuron within a large-scale neuronal cluster by regulating the information content encoded in each image block within the pattern;

[0065] Preferably, the information includes multiple categories;

[0066] Preferably, the information includes one or more dimensions such as mechanics, taste, touch, smell, temperature, and light stimulation.

[0067] Preferably, the photosensitive element comprises a light-sensitive protein disposed on a neuron.

[0068] This application also provides a method for fine-tuning large-scale neurons.

[0069] Connecting light-sensitive elements to the outside of neurons,

[0070] After encoding the information into pattern information, the spatial size of the pattern is compressed to reduce the single image block encoding the smallest information unit in the pattern to a size smaller than that of the photosensitive element, thereby obtaining a compressed pattern, and

[0071] Projecting the compressed pattern onto the array of photosensitive elements,

[0072] The photosensitive element connected to a single neuron is controlled by one or more image blocks of information encoded in the projected pattern, and the single neuron is then regulated by one or more image blocks of information encoded in the projected pattern. Ultimately, by manipulating the information content encoded in each image block in the pattern, individual neurons within a large-scale neuronal cluster can be finely regulated.

[0073] Preferably, the information includes multiple categories;

[0074] Preferably, the information includes one or more dimensions such as mechanics, taste, touch, smell, temperature, and light stimulation.

[0075] In another preferred embodiment, the photosensitive element includes connecting neurons to retinal ganglion cells provided with photoreceptor cells.

[0076] In another preferred embodiment, the pattern is compressed by a lens imaging system to achieve compression of the pattern space size and improve the information density in the space where the compressed pattern is located.

[0077] In another preferred embodiment, the steps include:

[0078] Step I: Divide event information into several dimensions;

[0079] Step II: Encode the information of each dimension into a corresponding pattern to obtain several patterns corresponding to all event information;

[0080] Step III: Projecting several patterns separately. Different projected patterns activate neurons in different areas of the nervous system. Projecting all event information into the nervous system, a neural network corresponding to the projected event information is formed and activated across multiple brain regions within the nervous system.

[0081] Preferably, in step I, the event information is divided into different dimensions according to the types of characterization parameters;

[0082] Preferably, in the same dimension, similar information is encoded according to intensity and coordinate information to form a pattern corresponding to the information in that dimension;

[0083] Preferably, the process of encoding the event information into a pattern realizes the correspondence between the event information and the pattern; then, through the pattern projection process, the event information is mapped to the first photosensitive element or the second photosensitive element; finally, the neurons are activated, and the event information is finally projected to the corresponding neurons in the nervous system;

[0084] Preferably, in the process of encoding event information to form a pattern, each dimension of event information independently adopts a coding language, coding information density, and coding information sorting scheme.

[0085] The present application also provides an artificial bio-electronic-mechanical intelligent body, including an imaging system that receives external event information and encodes the event information into pattern information, a pattern projection system that projects the pattern information, a photosensitive element that uses the pattern information to activate corresponding neurons in the nervous system, and a nervous system and support system for information processing.

[0086] In another preferred example, the event information comes from the detector's detection of the real physical world; or the event information comes from information generated or output by an electronic computer system; or the event information comes from mixed reality information of the detector's detection of the real physical world and information generated or output by an electronic computer system.

[0087] In another preferred embodiment, the nervous system is also connected to an execution system for outputting the result information obtained by the nervous system processing.

[0088] In another preferred embodiment, the artificial bio-electronic-mechanical intelligent body is trained using the method of human raising babies. Beneficial effects

[0089] The training method of the high-dimensional information cognition system of the present invention is adopted to train the nervous system using high-dimensional geometric figures and their projections and time evolution information, thereby building an intuitive spatial reasoning ability for the n-dimensional world, promoting the formation of an intuitive thinking paradigm for intelligent bodies to process high-dimensional figures and motion states, realizing the evolution of the intelligent body's high-dimensional information processing capabilities, and solving the problem that the natural evolution of current intelligent bodies cannot break through the high-dimensional spatial cognition and reasoning capabilities. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] FIG1 is a schematic diagram of the projection of a high-dimensional geometric figure onto a brain organoid according to the present invention;

[0091] FIG2 is a schematic diagram of a first apparatus for spatially compressing event information and delivering it to the nervous system according to the present invention;

[0092] FIG3 is a second schematic diagram of a device for delivering spatially compressed event information to the nervous system according to the present invention;

[0093] Figure 4 is a schematic diagram A of the projection construction module of the four-dimensional space information cognition system described in Embodiment 4 of the present invention;

[0094] Figure 5 is a schematic diagram B of the projection construction module of the four-dimensional space information cognition system described in Embodiment 5 of the present invention;

[0095] Figure 6 is a schematic diagram of the device of the four-dimensional space information cognition system described in Embodiment 4 of the present invention. Detailed implementation manners

[0096] The present invention will be further described below in conjunction with specific embodiments and Figures 1-6.

[0097] The present invention proposes a high-dimensional information processing system and its training method, which uses high-dimensional geometric figures and their projections and time evolution information to train the nervous system, and constructs an intuitive spatial reasoning ability for the n-dimensional world, and promotes the formation of an intuitive thinking paradigm for intelligent entities to process high-dimensional figures and motion states.

[0098] Embodiment 1: Graphic projection for neurons

[0099] This embodiment mainly provides a scheme for projecting geometric patterns onto a large-scale neuron cluster, especially for projecting graphic information onto a large-scale neuron structure such as a brain organoid or a biological brain, to achieve fine regulation of large-scale neurons, and to jointly use multiple groups of pattern information for neural network construction.

[0100] This embodiment includes a high-dimensional geometry generation unit, a projection construction module, a pattern projection system, and a visual nervous system brain organoid.

[0101] High-dimensional geometry generation unit, the high-dimensional geometry generation unit is configured to generate an n-dimensional geometric figure, where n≥4 and n∈N+;

[0102] Projection construction module, the projection construction module is used to construct an m-dimensional space projection set of the n-dimensional geometric figure, where 2≤m<n and m∈N+;

[0103] Pattern projection system: The m-dimensional space projection image of the n-dimensional geometric figure generated by the projection construction module is projected and compressed through the optical path of the pattern projection system, and the projection image is projected onto the visual nervous system; in other embodiments, the "pattern projection system" can be a functional description, as long as the system can project the m-dimensional space projection image of the n-dimensional geometric figure generated by the projection construction module onto a biological neural network or an external component.

[0104] Visual nervous system: The visual nervous system receives the image projected by the pattern projection system and transmits the image information to the brain organoid connected to the visual nervous system.

[0105] In this embodiment, the high-dimensional geometry generation unit is a computer program that generates high-dimensional geometric figures using mathematical operations;

[0106] In this embodiment, the projection construction module is a computer program that uses mathematical operations to generate an m-dimensional spatial projection image of an n-dimensional geometric figure;

[0107] In this embodiment, the pattern projection system is an optical system that uses an optical path to process the image generated by the projection building module and compresses the image before projecting it into the optic nerve system. In this embodiment, there should be multiple pattern projection systems, each of which projects a pattern formed by the m-dimensional spatial projection of an n-dimensional geometric figure into a optic nerve system. In this embodiment, the pattern projection system utilizes the principle of optical imaging to pass the pattern displayed by the imaging system through an optical device, such as an optical lens group, to reduce the pixel blocks of the encoded image to a size smaller than the size of the photosensitive elements of the optic nerve system that receive the information carried by the pixel blocks. This allows the byte information recorded by each pixel block in the image to be projected onto a single photoreceptor cell. The photoreceptor cell connects with the intermediate metacell, transmitting the byte information to the intermediate metacell. The intermediate metacell further activates the retinal ganglion cell, which transmits the byte information via the retinal ganglion cell axons to specific neurons within the biological brain or brain-like organ. Ultimately, each byte information in the image is transmitted to a specific neuron in the brain-like organ, achieving independent regulation of each neuron in the brain-like organ.

[0108] In this embodiment, when multiple images are projected onto the brain organoid, each image can be projected onto different regions of the brain organoid, or several micro-region groups can be set up in the same region, and different images can be cut and decomposed and projected onto different micro-region groups.

[0109] In this embodiment, the optic nerve system includes a photoreceptor cell layer, wherein the photoreceptor cells of the photoreceptor cell layer are connected to neural intermediary cells, which are connected to retinal ganglion cells. The photoreceptor cells activated by light stimulation transmit light information to the neural intermediary cells. After being activated, the neural intermediary cells transmit the information to the retinal ganglion cells, activating the retinal ganglion cells. The activated retinal ganglion cells transmit the signal through the axons to the neurons inside the connected brain organoid, thereby stimulating specific neurons inside the biological brain or brain organoid by stimulating a certain cell in the photoreceptor layer, and then inputting the image received by the optic nerve system into the brain organoid. It can be known by those skilled in the art of biological science that the optic nerve system mainly converts light signals into signals that activate downstream neurons. Therefore, the various components of the aforementioned first photosensitive element can be increased or simplified as long as the function of converting light signals into signals that activate downstream neurons can be achieved. In this embodiment, the optic nerve system that projects information to the brain organoid adopts the organoid culture scheme. Several independent culture areas are set up on the surface of the brain-like organ, and the supporting substances, nutrients and stem cells required for the development of the visual nervous system are placed in these areas to encourage the stem cells to differentiate into the required visual nervous system and grow nerve axons to selected areas of the biological brain or brain-like organ to project information.

[0110] In this embodiment, the m-dimensional space projection patterns of different n-dimensional geometric figures are respectively projected to different visual nervous systems by the pattern projection system, and then transmitted to different positions of the brain-like organ through different visual nervous systems, thereby realizing the delivery of the m-dimensional space projection patterns of the n-dimensional geometric figures to different brain regions, that is, projecting the m-dimensional space projection set of the n-dimensional geometric figures into the brain-like organ, and constructing a neural network corresponding to the m-dimensional space projection pattern of the n-dimensional geometric figures inside the brain-like organ.

[0111] The brain organoids described in this technical solution refer to neural clusters obtained through in vitro culture techniques. During the growth of the neural clusters, cells that develop into the visual nervous system are mixed in, thereby forming the visual nervous system on the surface of the brain organoids. Alternatively, elements expressing light-sensitive proteins are placed within the cultured neurons to enable them to be regulated by light. Alternatively, the culture device is configured with several areas, some for developing the brain organoids and some for developing the visual nervous system, with a channel established between the two areas. Through the neural development-inducing tissue, the visual nervous system is encouraged to project axons to the brain organoids, thereby transmitting information.

[0112] This embodiment uses the visual nervous system to receive patterned light signals and then projects the pattern information onto individual neurons within the nervous system. Alternatively, light-sensitive proteins or other light-sensitive materials can be placed on neurons to project the pattern directly onto them, thereby regulating them. The technology for receiving light information and activating neurons is a functional description. The light-sensitive proteins, visual nervous system, and other light-sensitive materials described in this embodiment can all achieve the beneficial effects of this technical solution. Therefore, different light-activated neuron technologies should all be considered within the scope of protection of this patent. More specifically, a photosensitive element is set on the neuron to form a photosensitive neuron with photosensitivity. The m-dimensional spatial projection pattern of the n-dimensional geometric figure is compressed by compressing the spatial size of the pattern, and the single image block encoding the minimum information unit in the pattern is reduced to below the size of the photosensitive neuron. The compressed pattern is projected onto a nervous system including a collection of such photosensitive neurons. The control of a single photosensitive neuron is achieved through a single or multiple image blocks in the projected pattern, and then the regulation of a single neuron is achieved through a single or multiple image blocks in the projected pattern. Finally, the pattern is used to achieve fine regulation of each neuron in a large-scale neuronal cluster, and the m-dimensional spatial projection pattern of the n-dimensional geometric figure is projected onto a biological neural network system.

[0113] In this embodiment, "information of each dimension is encoded to form a projection pattern" can be a direct projection of the projection pattern formed by a high-dimensional geometric figure in a low-dimensional space, or it can be a mathematical or optical transformation of the projection pattern formed in the low-dimensional space to form a newly encoded derivative projection pattern.

[0114] Similar to projections in three-dimensional space, each derived projection pattern can also be mixed with other information, such as mechanics, taste, touch, smell, temperature, light stimulation, etc. This hybrid pattern, which adds dimensional information such as mechanics, taste, touch, smell, temperature, and light stimulation to high-dimensional geometric shapes, can construct a high-dimensional static world.

[0115] Since the projection pattern activates the corresponding neurons, it prompts the nervous system to reorganize the m-dimensional space projection set of the n-dimensional geometric figure at the neural network level. At the same time, other dimensional information can also be encoded in the m-dimensional space projection pattern, or the projection area can be divided separately to encode other dimensional information. In this embodiment, the encoding method of other dimensional information does not affect the "understanding or processing" of the dimensional information by the nervous system. It is important to use a stable encoding method for information of a certain dimension, while different encoding methods can be used for information of different dimensions. The other dimensional information includes one or more of brightness, color information, hardness information, temperature information, density information, force information, and size information. For example, the above or other information encoding methods can all be used for information such as mechanics, acceleration, and pressure. The ways of encoding information are endless, and the specific information encoding method should not be regarded as a limitation on the scope of protection of this patent. The above encoding method examples are only used to promote the demonstration of encoding information into patterns.

[0116] For example, when encoding temperature information, considering that the purpose of projecting a pattern is to use the information in the pattern to regulate neurons, the preferred information is encoded into image blocks. The brightness and darkness of a single image block represent the smallest information unit, and the brightness and darkness of the image block correspond to the activation and inactivation of the photosensitive element at the location where the image block is projected. For example, for the temperature information of the target object in the information, an XY coordinate system is set in the pattern. The position information of each point on the object is represented by the (x, y) coordinate value of the central image block of a set of image block matrices (i.e., an array of image blocks, or a set of image blocks). The temperature gradient value at that point on the object's surface is represented by the number of illuminated image blocks in the image block matrix. When a large amount of image block matrix information is projected onto the nervous system through the photosensitive element, the two-dimensional projection shape of the target object and the temperature distribution on the target object's surface are projected onto the nervous system. This embodiment uses an image block matrix to encode information. The brightness or darkness of each image block in the image block matrix will change the temperature gradient value transmitted by the image block matrix. Therefore, each image block is the smallest unit of encoded information in the image, that is, the carrier of the smallest information unit is an image block. Therefore, when the image block is smaller than the diameter of the projected photosensitive element, the image block is only projected on the photosensitive element. The image block can only activate one or two adjacent photosensitive elements and the neurons connected to them. Therefore, the image block matrix regulates the corresponding number of neurons through the image blocks contained therein. The activated neuron group is input with the temperature information recorded by the image block matrix. Furthermore, the image blocks in all matrices in an image project the two-dimensional projection shape of the target object and the temperature distribution of the target surface to the nervous system.

[0117] Therefore, when the shape of the image block is circular or other regular polygons or slightly deformed similar to regular polygons, such an image block is conducive to projecting information to a single neuron with a photosensitive element. Especially when the maximum diameter of the image block is smaller than the diameter of the neuron and the size of the neuron gap, a single image block can achieve and can only achieve the regulation of one neuron, and the entire image can achieve fine regulation of each neuron in a large-scale neural cluster. The principle of projecting information to the neuron connected with the photosensitive element is similar and will not be repeated here, except that the pattern is directly projected onto the neuron provided with the photosensitive element.

[0118] Furthermore, a time evolution module is added to the high-dimensional geometric figure and its projection, and a time dimension parameter is injected into the n-dimensional geometric figure data model to generate a dynamic projection sequence with four-dimensional space-time continuity; and a correlation analysis module is added, configured to analyze the topological correlation formed during the movement of multiple groups of n-dimensional geometric figures in hyperspace, and generate a projection subset with movement trajectory marks. This dynamic projection technology introducing time evolution and projection correlation information of each dimension, combining relevant dimension information such as mechanics, taste, touch, smell, temperature, light stimulation, etc., can project a high-dimensional dynamic world more realistically and train the brain-like organ tissue to form an intuitive cognitive and reasoning ability for the high-dimensional dynamic world.

[0119] In this embodiment, the pattern projection system can be a lens imaging system.

[0120] Embodiment 2: Training method of high-dimensional information cognitive system

[0121] In this embodiment, an n-dimensional geometric figure is first generated, where n≥4 and n∈N+; then an m-dimensional space projection set of the n-dimensional geometric figure is constructed, where 2≤m<n and m∈N+; and then the m-dimensional space projection set is used to train an intelligent network. In this embodiment, the n-dimensional geometric figure evolves along the time dimension to generate a space projection set with time evolution characteristics, realizing the training of n-dimensional dynamic events, especially using multiple groups of correlated n-dimensional geometric figures for training to realize the training of n-dimensional dynamic events; the multiple groups of correlated n-dimensional geometric figures refer to multiple groups of n-dimensional geometric figure frames formed during the movement of a group of n-dimensional geometric figures in the multi-dimensional space where they are located; the use of the m-dimensional space projection set to train the intelligent network includes performing a multi-modal projection operation on the n-dimensional geometric figure to generate a projection set containing k m-dimensional projections, where k≥C(n,m).

[0122] The multi-modal projection operations include: orthogonal projection: performing dimensional folding along the direction of the standard basis vectors; rotational projection: performing non-orthogonal projection after rotation transformation within a selected hyperplane; topological-preserving projection: a continuous mapping that preserves the specific topological properties of the original n-dimensional geometric body. The m-dimensional space projection is a three-dimensional Euclidean space, and the projection set contains no less than n(n - 1)(n - 2) / 6 three-dimensional projections, and each projection corresponds to a different three-dimensional subspace truncation method; one or more of brightness, color information, hardness information, temperature information, density information, force information, and size information are also included in the n-dimensional geometric figure and / or its projection.

[0123] In this embodiment, the intelligent network is selected from biological neural networks. The biological neural network is selected from brain organoids and includes neuron cells inside. Each projection of the n-dimensional geometric figure is projected onto different neurons of the biological neural network, and the neurons are activated, prompting the activated neurons to construct a neural network that maps all projections of the n-dimensional geometric figure to each other, that is, forming a cognition of the n-dimensional geometric figure.

[0124] Furthermore, this embodiment can perform self-iterative upgrade to form an ultra-high-dimensional cognition and reasoning network: train the n-dimensional information processing system according to the content of the above embodiment, where n≥4 and n∈N+; use this n-dimensional information processing system to generate n + 1-dimensional geometric figures and corresponding projections, dynamic events, and thus train to obtain an n + 1-dimensional information processing system.

[0125] Embodiment 3: High-dimensional Information Cognition System

[0126] In this embodiment, the high-dimensional information cognition system includes a high-dimensional geometry generation unit configured to construct an n-dimensional geometric figure data model, where n≥4 and n∈N+; a projection construction module connected to the high-dimensional geometry generation unit for generating a projection set containing k m-dimensional projections, where 2≤m<n and k≥C(n,m); an intelligent network connected to the projection construction module for training using projection information. The projection construction module includes at least three projection operators: an orthogonal projection operator, a rotational projection operator, and a topological-preserving projection operator.

[0127] The projection construction module further includes:

[0128] A time evolution module configured to inject a time dimension parameter into the n-dimensional geometric figure data model to generate a dynamic projection sequence with four-dimensional space-time continuity;

[0129] A correlation analysis module configured to analyze the topological correlations formed during the movement of multiple groups of n-dimensional geometric figures in hyperspace and generate a projection subset with motion trajectory markings.

[0130] The time evolution module includes:

[0131] The curvature calculation submodule calculates the Gaussian curvature change of the n-dimensional manifold on the time axis in real time;

[0132] The singularity warning submodule triggers the compensatory projection generation mechanism when a topological structure mutation is detected in the projection set.

[0133] The topology preserving projection operator implementation includes:

[0134] Homology group maintainer, which ensures that the geometry before and after projection remains isomorphic in the homology group dimension of order p, where p≤m;

[0135] Fiber bundle mapper, decomposes the n-dimensional principal bundle structure into the direct product projection of the m-dimensional basis space and the (nm)-dimensional fiber space.

[0136] In this embodiment, the intelligent network is selected from a biological neural network; the biological neural network is connected to the visual nervous system, which is configured to receive the projection information optical signal from the projection building module and convert the projection signal optical signal into a neural signal, which is then transmitted to the biological neural network. The biological neural network utilizes optogenetic manipulation to input the projection information optical signal from the projection building module.

[0137] Those skilled in the art will appreciate that high-dimensional geometric figures and their projections can be constructed from mathematical algorithms, and the construction of each high-dimensional geometric figure and its projection information are obtained by mathematical algorithms, so no detailed examples will be given in this embodiment.

[0138] An artificial bio-electronic-mechanical intelligent entity comprises a brain organoid and a support system for information processing, a optic system connected to the brain organoid, a pattern projection system for projecting information to the optic system, an encoding system for receiving external event information and encoding the event information into pattern information, and an information output system connected to the brain organoid for externally outputting the information processed by the brain organoid. In this embodiment, the optic system connected to the brain organoid and the pattern projection system for projecting information to the optic system can be considered as the same system, realizing the function of projecting information to the brain organoid.

[0139] In this embodiment, the brain organoid is trained with high-dimensional information and has the ability to recognize and reason about high-dimensional information. It uses this ability to process three-dimensional information or four-dimensional spatiotemporal information to obtain better reasoning results.

[0140] The support system described in this embodiment refers to a system for maintaining brain-like organs, including a nutrition system, a temperature control system, an oxygen supply system, a development system, etc., which are used to maintain the growth, development and information processing of brain-like organs.

[0141] In this embodiment, the external event information comes from the detection of the real physical world by the detector; or the external event information comes from information generated or output by the electronic computer system.

[0142] Example 4: Four-dimensional spatial information recognition system (A)

[0143] In this embodiment, the four-dimensional information recognition system includes: a biological neural network system, a four-dimensional geometry generation unit for constructing a four-dimensional geometric figure data model, and a projection construction module for generating a projection of a four-dimensional geometric figure.

[0144] In this embodiment, a plurality of cube imaging modules (the imaging modules are pattern projection systems) are provided inside, and the six surfaces of each cube imaging module are imaging surfaces.

[0145] In this embodiment, a layer of the optic nerve system is grown on the outer wall of each cube-shaped imaging module, and a biological neural network is grown around the optic nerve system. In this embodiment, the biological neural network is selected from a brain organoid (i.e., a brain organoid that can be developed in vitro from neural stem cells). The optic nerve system receives image information from the imaging surface and projects this image information onto the brain organoid.

[0146] The 4D geometric figure is processed by the projection construction module, generating a set of 3D projections of the 4D geometric figure in 3D space. Each 3D projection is then placed on the six imaging surfaces of a cube imaging module for imaging. Each 3D projection is then projected onto the brain organoid to construct a cognitive network for the 4D geometric figure.

[0147] Example 5: Four-dimensional spatial information recognition system (B)

[0148] In this embodiment, the four-dimensional information recognition system includes: a biological neural network system, a four-dimensional geometry generation unit for constructing a four-dimensional geometric figure data model, and a projection construction module for generating a projection of a four-dimensional geometric figure.

[0149] In this embodiment, several cube-shaped imaging modules (these imaging modules are pattern projection systems) are internally arranged, and each cube-shaped imaging module has six imaging surfaces. In this embodiment, the imaging surface is a neural microelectrode array, and the image information of each surface is presented by activating the microelectrode array.

[0150] In this embodiment, each cube-shaped imaging module is placed inside the brain organoid, with each imaging surface surrounding the organoid's neurons. The image information from each imaging surface is converted into activation information for the neural microelectrode array outside that surface, which is then projected into the brain organoid via the neural microelectrodes.

[0151] The four-dimensional geometric figure is processed by the projection construction module to generate a set of three-dimensional projections of the four-dimensional geometric figure in three-dimensional space. Each three-dimensional projection is placed on the six imaging surfaces of a cube imaging module for microelectrode array imaging. Each three-dimensional projection is then projected onto the brain organoid to construct a cognitive network for the four-dimensional geometric figure.

[0152] Example 6

[0153] In the existing technology, neural electrodes are often used to deliver information to the nervous system, such as Utah electrodes and flexible neural electrodes. However, these neural electrodes have many disadvantages that are difficult to overcome:

[0154] 1. Limited by the channel density of neural electrodes, information cannot be efficiently delivered to the nervous system;

[0155] 2. Limited by the refresh rate of neural electrode signals, the frame rate at which neural electrodes deliver electrical signals to the nervous system is low;

[0156] 3. Neural electrodes need to be implanted in the nervous system to function, which may lead to immune rejection between the neural electrodes and the nervous system, or the neural electrodes may fall off or shift when the animal moves, thereby causing mechanical damage to the nervous system. It also affects the stable use of the neural electrodes and reduces the transmission efficiency and accuracy of neural signals.

[0157] Therefore, the existing technologies for high-density information delivery to the nervous system and achieving fine and stable regulation of large-scale neurons have many difficult-to-overcome difficulties, which restrict the development of related industries in the field of brain science.

[0158] Referring to Figures 2 and 3, the present invention proposes a device for spatially compressing event information and delivering it to the nervous system, as well as a method for fine-tuning large-scale neurons. This method uses a non-direct contact method to project information at a high density to the nervous system, including the brain, spinal cord, brain-like organs, and peripheral nerves, to achieve the beneficial effect of fine-tuning individual neurons in large-scale neuronal clusters. In particular, by projecting information to complex nervous system partitions such as the brain, it achieves the input of multimodal information to the brain, thus avoiding the problems of low efficiency of neural electrode data transmission, mechanical damage to neural tissue by neural electrodes, inaccurate neural electrode information delivery, and the inability of neural electrodes to accurately deliver information to individual neurons.

[0159] Specifically, the device of this embodiment includes an imaging system that encodes event information into several patterns, several visual nervous systems that use the minimum information unit encoded in the pattern to regulate the activity of single or multiple neurons in the nervous system, and a pattern projection system that projects the several patterns encoded by the imaging system to each visual nervous system.

[0160] In this embodiment, event information refers to event information input to the brain. Event information can come from multiple sources. In this embodiment, event information can come from the combined information obtained by several detectors detecting the physical world, or from virtual events generated by a computer system. It can also be a virtual reality hybrid event of physical world events and virtual events generated by a computer system. Describing an event often requires information in multiple dimensions, such as temperature, taste, touch, smell, vision, acceleration, and other information, as well as the coordinate information of this information. In this embodiment, event information is classified, and the overall event information is divided into information of different dimensions. Each item of information in each dimension contains the corresponding coordinate information of the information.

[0161] In this embodiment, information about each dimension of an event is encoded into a specific pattern, and information such as the intensity, location, and attributes of the information is converted into coordinate values, brightness, and color information for the image blocks within the pattern. In this embodiment, the information encoding method and corresponding encoding algorithm for each dimension can be independent, enabling the design of a unique encoding language based on the different information density and attributes of each dimension, achieving highly efficient information input. Details on the encoding method from information to image are provided in the subsequent embodiments of this specification.

[0162] The visual nervous system is selected from: a first photosensitive element set independently of neurons, which generates nerve pulses under the action of light and transmits the generated nerve pulses to neurons; or the visual nervous system is selected from neurons with photosensitive points set on the surface or inside, which activate the neurons where they are located under the action of light signals. The neurons are defined as photosensitive neurons, that is, the second photosensitive element.

[0163] In this embodiment, a first photosensitive element is used that is independent of the neuron setting: specifically, each visual nervous system includes a photoreceptor cell layer, the photoreceptor cells of the photoreceptor cell layer are connected to the neural interneuron cells, and the neural interneuron cells are connected to the retinal ganglion cells. The photoreceptor cells activated by light stimulation transmit the light information to the neural interneuron cells. After the neural interneuron cells are activated, they transmit the information to the retinal ganglion cells, activating the retinal ganglion cells. The activated retinal ganglion cells transmit the signal through the axons to the neurons connected to them in the brain, thereby stimulating specific neurons in the brain by stimulating a certain cell in the photoreceptor layer. In this embodiment, the visual nervous system that projects information to the brain adopts an organoid culture scheme. Several independent culture areas are set on the surface of the brain, and the supporting substances, nutrients, and stem cells required for the development of the visual nervous system are placed in this area to promote the differentiation of stem cells into the required visual nervous system and grow nerve axons to selected areas of the brain to project information.

[0164] In this embodiment, there are multiple pattern projection systems, each projecting a pattern encoded with event information of a specific dimension to a specific visual system. In this embodiment, the pattern projection system utilizes the principle of optical imaging, passing the pattern displayed by the imaging system through an optical device, such as an optical lens assembly, to reduce the image's smallest information unit, or image block, to a size smaller than the size of the first or second photosensitive element that receives the information carried by the image block. For example, the smallest information unit in the image, that is, the image block, is reduced to a size below the first photosensitive element that receives the information carried by the image block, so that each byte of information in the image is projected to a single photoreceptor cell. The photoreceptor cell is connected to the intermediate meta-cell to transmit the byte information to the intermediate meta-cell. The intermediate meta-cell further activates the retinal ganglion cells, thereby transmitting the byte information to specific neurons in the brain through the retinal ganglion cell axons, and finally achieving the transmission of each byte of information in the pattern to specific neurons in the brain, and realizing independent regulation of each neuron in the brain; for example, the smallest information unit in the image, that is, the image block, is reduced to a size below the second photosensitive element that receives the information carried by the image block, thereby projecting the information carried by each image block in the image to a single photoreceptor neuron with light sensitivity, thereby transmitting the byte information to specific neurons in the brain through the retinal ganglion cell axons, and finally achieving the transmission of each byte of information in the pattern to specific neurons in the brain, and realizing independent regulation of each neuron in the brain.

[0165] The encoding of information into a pattern as described in this embodiment should be regarded as a functional statement, that is, any technology that encodes information into a pattern should be regarded as a protection scheme of this patent. Taking into account that the purpose of projecting a pattern is to apply the information in the pattern to the regulation of neurons, the preferred information is encoded into an image block, and the brightness and darkness of a single image block represent the smallest information unit, respectively. The brightness and darkness of the image block correspond to the activation and inactivation of the photosensitive element at the position where the image block is projected. For example, the temperature information of the target object in the event information, an XY coordinate system is set in the pattern, and the position information of each object is represented by the (x, y) coordinate value of the central image block of a set of image block matrices (i.e., an array of image blocks, or a set of image blocks), and the temperature gradient value of the point on the surface of the object is represented by the number of illuminated image blocks in the image block matrix. When a large amount of image block matrix information is projected onto the nervous system through the photosensitive element, the event information of the two-dimensional projection shape of the target object and the temperature distribution on the surface of the target object is projected onto the nervous system, that is, the frame event of "the temperature of the target object surface at this moment"; when the surface temperature information of the target object is projected dynamically and in real time, the time series event of "the temperature change process of the target object surface" is projected onto the nervous system, that is, multiple frame events related to each other are continuously projected, thus completing the projection of the time series event.

[0166] In addition, since the projected event information prompts the nervous system to recombines event information of different dimensions at the neural network level, in this embodiment, the way the information is encoded does not affect the nervous system's "understanding or processing" of the event. It is important to use a stable encoding method for information of a certain dimension, while different encoding methods can be used for information of different dimensions. The above or other information encoding methods can all be used for information such as mechanics, acceleration, and pressure. The ways of encoding information are endless, and the specific information encoding method should not be regarded as a limitation on the scope of protection of this patent. The above encoding method examples are only used to facilitate the demonstration of encoding information into patterns.

[0167] Furthermore, the above embodiment uses a symmetric matrix of image blocks to encode information. The brightness or darkness of each image block in the image block matrix will change the temperature gradient value transmitted by the image block matrix. Therefore, each image block is the smallest unit of encoded information in the image, that is, the carrier of the smallest information unit is an image block. Therefore, when the image block is smaller than the diameter of the projected photosensitive element, the image block is only projected on the photosensitive element. The image block can only activate one or two adjacent photosensitive elements and the neurons connected to them. Therefore, the image block matrix controls the corresponding number of neurons through the image blocks contained therein. The activated neuron group is input with the temperature and coordinate information recorded by the image block matrix. Furthermore, the image blocks in all matrices in an image project the event information of the two-dimensional projection shape of the target object and the temperature distribution on the surface of the target object to the nervous system.

[0168] Therefore, when the image block is a circle or other regular polygon, or a slightly deformed regular polygon, it is advantageous for projecting information to a single neuron with a photosensitive element. In particular, when the maximum diameter of the image block is smaller than the neuron diameter and the interneuronal gap, a single image block can control only one neuron, and the entire image can thus achieve fine-grained control of every neuron in a large neural cluster. The principle of projecting information to neurons connected to photosensitive elements is similar and will not be repeated here, except that the pattern is projected directly onto the neurons equipped with photosensitive elements.

[0169] In this embodiment, the imaging system includes a computer system that encodes event information into pattern information, and also includes a display device that displays the encoded pattern. The display device in this embodiment is based on existing technology and offers a variety of options, broadly classified into three types: self-luminous, transmissive, and reflective. Self-luminous devices primarily include CRTs, LCDs, and OLEDs. Transmissive devices, such as LCDs that simply retain the liquid crystal optical rotation display structure and project patterns by blocking light, also include similar devices such as photolithography masks in equipment such as lithography machines. Backlit devices include DLP-based digital micromirror chips (DMDs) (mostly used in projectors) and LCOS pattern projection devices.

[0170] Depending on the display device of the above-mentioned imaging system, there will be different pattern projection systems. For example, the optical path design of a lithography machine similar to chip processing and manufacturing projects the pattern of the mask or translucent LCD screen to the target position; for example, there will be an optical path design similar to a projector, which projects the image of the DMD or LCIOS chip to the target position, such as the optical path design of a common projector; for example, through the lens imaging principle, the pattern of the self-luminous display device is directly projected to the target position. In short, the display and projection technical effects of the image can be achieved through the optical path design of the existing technology, and the specific optical path will not be listed or designed in this embodiment. As long as the optical path can be reduced and projected after the lens combination and adjustment, it should be sufficient to disclose the technical solution of this application.

[0171] In this embodiment, the projection system further includes an optical path system for adjusting the position of the projected pattern, enabling the same sub-projection system to project different patterns to different regions of a given brain area, thereby expanding the range of brain areas controlled by the projection system. Specifically, through the image deflection optical path system, the pattern encoding information is automatically associated with the projected pattern position, enabling field scanning of the projected image across different regions of the same brain area. During a single full-field scan, the same sub-projection system can project different patterns to different regions across a wide range of brain areas. During multiple full-field scans, the same sub-projection system can project different patterns to different regions across a wide range of brain areas, as well as to the same region at different times, thereby expanding the area of ​​brain areas controlled by the projection system.

[0172] Those skilled in the art will appreciate that other solutions capable of projecting such images to the visual system can also achieve the beneficial effects of this technical solution and should also be considered within the scope of protection of this patent.

[0173] In this embodiment, if the pattern is projected directly onto a neuron with photosensitivity, considering that the diameter of a neuron axon is generally not less than 12 microns, the maximum inner diameter of the image block in the projected pattern in this embodiment is controlled to be in the range of 20nm to 10μm, which better balances the imaging difficulty and the fineness of control, and realizes fine control of each neuron in the nervous system.

[0174] In this embodiment, event information is classified and segmented into information of different dimensions, which are then encoded into different patterns. These patterns are then projected by the pattern projection system to different optic nerve systems, which in turn transmit them to different brain regions. This allows event information segmented by different dimensions to be delivered to different brain regions, projecting multimodal information into the brain and building a multimodal neural network corresponding to the events.

[0175] In this embodiment, this system can be used to input dynamic events, namely sequential events, into the brain. Sequential events are dynamic events formed by combining multiple consecutive, interrelated frame events. This device projects the entire sequential event by sequentially and continuously projecting each frame event, forming a network group describing the sequential event.

[0176] This technical solution is used to project information to brain-like organs. The brain-like organ refers to a neural cluster tissue obtained by in vitro culture technology. During the growth of the neural cluster tissue, cells that develop into the visual nervous system are mixed in, thereby forming a visual nervous system on the surface of the brain-like organ; or elements that express light-sensitive proteins are set inside the cultured neurons to enable the neurons to be regulated by light; or several areas are set in the culture device, some areas are used to develop brain-like organs and some areas are used to develop the visual nervous system, and a channel is set between the two, so that the visual nervous system is prompted to project nerve axons to the brain-like organ through the neural development induction tissue to transmit information.

[0177] This embodiment uses the visual nervous system to receive patterned light signals and then projects the pattern information onto individual neurons within the nervous system. Alternatively, light-sensitive proteins or other light-sensitive materials can be placed on neurons to project the pattern directly onto them, thereby regulating them. The technology for receiving light information and activating neurons is a functional description. The light-sensitive proteins, visual nervous system, and other light-sensitive materials described in this embodiment can all achieve the beneficial effects of this technical solution. Therefore, different light-activated neuron technologies should all be considered within the scope of protection of this patent.

[0178] This embodiment provides a method for fine-tuning large-scale neurons: neurons within the nervous system are provided with or connected to photosensitive elements, event information is encoded into pattern information, the pattern is compressed in spatial dimensions, the information density within the space where the compressed pattern is located is increased, and the compressed pattern is projected onto the neuronal photosensitive elements to achieve high-density information transmission to the nervous system.

[0179] More specifically, photosensitive elements are provided on neurons to form photosensitive neurons with photosensitivity. After encoding the event information to be delivered into pattern information, the spatial size of the pattern is compressed to reduce the single image block encoding the smallest information unit in the pattern to less than the size of the photosensitive neuron. The compressed pattern is then projected onto a neural system including a collection of such photosensitive neurons. The single photosensitive neuron is controlled by one or more image blocks encoding information in the projected pattern, and the single neuron is further regulated by one or more image blocks encoding information in the projected pattern. Ultimately, by regulating the information content encoded in the pattern, fine regulation of individual neurons within a large-scale neuronal cluster is achieved.

[0180] Alternatively, after encoding event information into pattern information by connecting a photosensitive element externally to a neuron, the pattern's spatial size is compressed, reducing individual image blocks of the encoded information in the pattern to a size smaller than the size of the connected photosensitive element. The compressed pattern is then projected onto the photosensitive element, and the photosensitive element corresponding to a single neuron is controlled by one or more image blocks of the information encoded in the projected pattern. Furthermore, the individual neuron is regulated by one or more image blocks of the information encoded in the projected pattern. Ultimately, by regulating the content of the pattern's encoded information, fine regulation of individual neurons within a large-scale neuronal cluster is achieved. The external connection to the photosensitive element includes connecting the neuron to a retinal ganglion cell equipped with a photoreceptor.

[0181] In this embodiment, the information includes multiple categories: one or more dimensions such as mechanics, taste, touch, smell, temperature, and light stimulation; in the same dimension, similar information is encoded according to intensity and coordinate information to form a pattern corresponding to the dimensional information.

[0182] In this embodiment, the pattern is compressed by the lens imaging system to achieve compression of the pattern space size, thereby improving the information density in the space where the compressed pattern is located.

[0183] An artificial bio-electronic-mechanical intelligent entity comprises a brain organoid and a support system for information processing, a visual system connected to the brain organoid, a pattern projection system for projecting information to the visual system, an imaging system for receiving external event information and encoding the event information into pattern information, and an information output system connected to the brain organoid for externally outputting the results of the brain organoid processing. The support system in this embodiment refers to a system for maintaining the brain organoid, including a nutrition system, a temperature control system, an oxygen supply system, a development system, etc., to maintain the growth and development of the brain organoid and information processing.

[0184] In this embodiment, the external event information may be derived from a detector detecting the real physical world; or it may be derived from information generated or output by an electronic computer system. In this embodiment, the selection of external event information should adhere to the principles of human civilization and maintain the basic respect humans give to living beings for artificial bio-electronic-mechanical intelligent entities.

[0185] All documents mentioned in the present invention are considered to be included in their entirety in the disclosure of the present invention so that they can be used as a basis for modification when necessary. In addition, it should be understood that after reading the above disclosure of the present invention, those skilled in the art may make various changes or modifications to this application, and these equivalent forms also fall within the scope of protection claimed in the present invention.

Claims

1. A training method for a high-dimensional information cognition system, characterized in that: It includes the following steps; Step I: Generate an n-dimensional geometric figure, where n≥4 and n∈N+; Step II: Construct a set of m-dimensional spatial projections of the n-dimensional geometric figure, where 2≤m<n and m∈N+; Step III: Use the set of m-dimensional spatial projections to train an intelligent network, thereby obtaining an n-dimensional information cognition system.

2. The training method of the high-dimensional information recognition system according to claim 1, characterized in that: In Step I, the n-dimensional geometric figure includes one or more n-dimensional geometric figures that evolve along the time dimension and constitute an n-dimensional dynamic event. Thus, in Step II, a set of m-dimensional spatial projections with time-evolution characteristics is constructed. Furthermore, the set of m-dimensional spatial projections is used to train the intelligent network, thereby realizing the training of the intelligent network by the n-dimensional dynamic event. Preferably, the evolution along the time dimension refers to the change of one or more of the following characteristics of the n-dimensional geometric figure: position, shape, mechanical strength, mass, color, and temperature.

3. The training method of the high-dimensional information recognition system according to claim 1, characterized in that: Use multiple associated n-dimensional geometric figures for training to realize the training of the n-dimensional dynamic event. The multiple associated n-dimensional geometric figures refer to a set of n-dimensional geometric figure frames formed during the evolution of a set of n-dimensional geometric figures along the time dimension in the multi-dimensional space where they are located. Preferably, the evolution along the time dimension refers to the change of one or more of the following characteristics of the n-dimensional geometric figure: position, shape, mechanical strength, mass, color, and temperature.

4. The training method of the high-dimensional information recognition system according to claim 1, characterized in that: Constructing the set of m-dimensional spatial projections of the n-dimensional geometric figure in Step II includes performing a single-modal or multi-modal projection operation on the n-dimensional geometric figure to generate a projection set containing k m-dimensional projections. Preferably, k≥C(n,m).

5. The training method of the high-dimensional information recognition system according to claim 4, characterized in that: The single-modal or multi-modal projection operation includes one or more of the following projection methods: Orthogonal projection: Perform dimensional folding along the direction of the standard basis vector; Rotational projection: Non-orthogonal projection after performing a rotation transformation within a selected hyperplane; Topology-preserving projection: Continuous mapping that preserves the specific topological properties of the original n-dimensional geometric body.

6. The training method of the high-dimensional information recognition system according to claim 4, characterized in that: The m-dimensional spatial projection is a three-dimensional Euclidean space projection. The set of m-dimensional spatial projections contains no less than n(n - 1)(n - 2) / 6 three-dimensional projections, and each projection corresponds to a different three-dimensional subspace intercepting method.

7. The training method of a high-dimensional information recognition system according to claim 1, characterized in that: One or more of brightness, color information, hardness information, temperature information, density information, force information, and size information are also included in the n-dimensional geometric figure and / or its projection.

8. The training method of a high-dimensional information recognition system according to claim 1, wherein: The intelligent network is a biological neural network.

9. The training method of the high-dimensional information recognition system according to claim 8, characterized in that: The biological neural network includes a network composed of neuron cells. Preferably, the biological neural network is selected from the following group: brain organoids, biological brains, and brains-on-chips, where the brains-on-chips refer to a chip-neuron hybrid system.

10. The training method of a high-dimensional information recognition system according to claim 8, characterized in that: Step III includes: projecting each projection in the m-dimensional space projection set onto different neurons of the biological neural network respectively, activating the neurons, and prompting the activated neurons to be connected to each other, so as to construct a neural network that maps each projection of all or part of the m-dimensional space projection set, that is, to form the cognition of the n-dimensional geometric figure; preferably, the association relationship between each projection in the m-dimensional space projection set is converted into the association relationship between the projected neuron groups.

11. A training method for a high-dimensional information cognition system, characterized in that: It includes the following steps: training the intelligent network according to the method of any one of claims 1 to 10, and obtaining the n-dimensional information cognition system, where n≥4 and n∈N+; using the trained n-dimensional information cognition system to generate n+1-dimensional geometric figures and corresponding projection sets and dynamic events, and training to obtain the n+1-dimensional information cognition system. Preferably, the dynamic event consists of n+1-dimensional geometric figures evolving along the time dimension.

12. A high-dimensional information recognition system, characterized in that: It includes: A high-dimensional geometry generation unit configured to generate an n-dimensional geometric figure, where n≥4 and n∈N+; A projection construction module for constructing an m-dimensional space projection set of the n-dimensional geometric figure, where 2≤m<n and m∈N+, preferably, the m-dimensional space projection set includes k m-dimensional projections, k≥C(n,m); A training module for training the intelligent network with the m-dimensional space projection set to obtain an n-dimensional information cognition system.

13. The high-dimensional information recognition system according to claim 12, characterized in that: The projection construction module includes at least one or more of the following three projection operators: an orthogonal projection operator, a rotation projection operator, and a topology-preserving projection operator.

14. The high-dimensional information recognition system according to claim 12, characterized in that: The n-dimensional geometric figure generated by the high-dimensional geometry generation unit includes one or more n-dimensional geometric figures that constitute an n-dimensional dynamic event evolving along the time dimension, so that the projection construction module constructs the space projection set with time evolution characteristics, and then the training module trains the intelligent network with the space projection set, so as to realize the training of the intelligent network by the n-dimensional dynamic event.

15. The high-dimensional information recognition system according to claim 13, characterized in that: The implementation of the topology-preserving projection operator includes: A homology group retainer to ensure that the geometric body before and after projection is isomorphic in the p-th homology group dimension, where p≤m; A fiber bundle mapper to decompose the n-dimensional principal bundle structure into the direct product projection of the m-dimensional base space and the (n-m)-dimensional fiber space.

16. The high-dimensional information recognition system according to claim 12, characterized in that: The intelligent network is a biological neural network; preferably, the biological neural network includes a network composed of neuron cells; preferably, the biological neural network is selected from the following group: brain organoids, biological brains, and brains-on-chip, and the brains-on-chip refer to a chip and neuron hybrid system.

17. The high-dimensional information recognition system according to claim 16, characterized in that: The biological neural network is connected to the optic nerve system, and the optic nerve system is used to receive the optical signal of the projection information generated by the projection construction module projected by the pattern projection system, and convert the projection information optical signal into a neural signal to transmit and input into the biological neural network; preferably, the projection construction module is a three-dimensional imaging geometry, and a biological neural network is arranged around the three-dimensional imaging geometry; more preferably, the three-dimensional imaging geometry is used for the projection of n-dimensional geometric figures in two-dimensional or three-dimensional projection; more preferably, the three-dimensional imaging geometry and the peripheral biological neural network realize the input of the projection image into the biological neural network through the optic nerve system or optogenetics or neural electrode means.

18. The high-dimensional information recognition system according to claim 16, characterized in that: The biological neural network is selected from the brain-on-chip, and the projection information of the projection construction module is converted into chip output information of the brain-on-chip, thereby realizing the projection of the biological neural network of the brain-on-chip.

19. The high-dimensional information recognition system according to claim 16, characterized in that: The biological neural network utilizes optogenetic manipulation to input a light signal of projection information projected by a pattern projection system and generated by a projection building module.

20. An artificial bio-electronic-mechanical intelligent agent comprising a biological neural network for information processing, an encoding system for receiving external event information and encoding the event information into pattern information, and a pattern projection system for projecting the pattern information onto the biological neural network for externally outputting the result information obtained by the biological neural network processing; The high-dimensional information cognition system obtained by the biological neural network through any training method of claims 1-10, the biological neural network has the ability to recognize and reason about high-dimensional information; the number of dimensions of the high-dimensional information n≥4 and n∈N+; preferably, it also includes an information output system connected to the biological neural network.

Images