A brain-based element sharing type data processing architecture and an implementation method thereof

Abstract

This invention discloses a human brain-based element-sharing data processing architecture and its implementation method. Based on the perception, memory, and neuronal connectivity characteristics of the human brain's biological neural network, this architecture decomposes the target processing object into indivisible and indispensable basic core elements according to their inherent attributes. These elements are then combined according to a preset hierarchical relationship to construct a multi-level sub-element structure. The basic core elements carry real attribute information and are fixedly stored in the bottom-level module. Sub-elements introduce an attention mechanism to filter and adapt element combinations, recording only the temporal, spatial, and logical relationships with the basic core elements / other sub-elements, without storing specific attribute information. All event elements exist in a distributed manner based on the hierarchical structure and are processed and expressed through relationships. This invention constructs a global sharing, multi-level bidirectional traceability and association mechanism between basic core elements and sub-elements. Information import naturally maps the inherent relationships of things, abandoning the rigid dependence on large-scale sample training and ultra-large-scale computing power. It adopts a distributed element-sharing and hierarchical parallel computing architecture, enabling efficient operation in conventional computing environments.

Description

Technical Field

[0002] This invention relates to the field of computer technology, mainly to the design of software and hardware architectures for information expression, storage and data processing, specifically to a human brain-based element-sharing data processing architecture and its implementation method, applicable to computer architecture design, artificial intelligence model construction and the construction of various data processing architectures. Background Technology

[0004] Artificial information processing systems are mainly divided into two categories: the first category is traditional computer systems and various information acquisition and processing systems based on the von Neumann architecture, whose core is the digitization of information expression, the abstraction and modularization of information processing; the second category is machine learning models represented by artificial neural networks, which are essentially mathematical mapping and fitting models from input to output, achieving task objectives through large-scale data training, weight iteration and global fitting.

[0005] Compared to the human brain, while artificial systems such as computers have significant advantages in data storage and computation speed, the differences in tasks such as fuzzy information processing, common sense understanding, and adaptive learning are not merely a matter of speed or efficiency, but rather stem from fundamentally different underlying operating mechanisms. The following section will illustrate this with examples of traditional computer systems and artificial neural networks.

[0006] 1. Computer systems: Information digitization, abstraction, and modular processing.

[0007] Figure 1 This is a schematic diagram of a typical computer data representation and storage method. Assume the system has three independent inputs. , , From the perspective of source-destination communication transmission, each input signal's original source possesses a dual core identifier of address and property, giving the signal independent and distinguishable unique attributes. The common transmission line enables parallel transmission of signals from different sources and multiplexing of signals from the same source. The three independent input terminals allow for precise separation of various signals transmitted through the common line at the receiving end. Regardless of whether the signals are signals with different addresses and properties, signals with the same address and properties, signals with different properties but the same address and different properties, or multiple signals with the same address and properties, they can all be independently expressed and stored within the computer system through these three independent input terminals, ensuring independent processing and no interference between signals.

[0008] Data converters 1, 2, and 3 correspond to three input channels, used to convert the corresponding channel signals into binary data. Each input signal can be a physical quantity of different magnitudes, or represent the structure, index, or attribute of an object, or a symbolic representation or various abstract expressions of an object. Each input can form a series of data. The three data channels are independently stored in binary data memories 1, 2, and 3, respectively. The data converters are connected to the memories by straight lines with arrows; this structure is only for illustration and is used to indicate that the corresponding channel data is independently stored in a designated location. The three memories can belong to different physical memories or be different storage segments within the same physical memory. The memory access method typically uses a von Neumann bus structure.

[0009] Therefore, the typical characteristics of information processing in computer systems are datafication, abstraction, and modular processing. All types of external information are uniformly converted into binary data representation. The meaning and ownership of data are defined through agreed-upon data structures, storage formats, and abstraction rules. Storage, transmission, and execution are completed in a modular form, using blocks, segments, and address-based addressing. These modules include both local data or program instructions, as well as complete data files or programs. This type of data processing essentially covers all existing data acquisition and information transmission systems.

[0010] 2. Artificial Neural Networks: Mathematical models for implementing input-output mapping.

[0011] Artificial neural network models initially drew on the basic characteristics of biological neural networks. On the one hand, they simulated the interconnection structure between neurons and relied on the network topology to achieve information processing and transmission. On the other hand, they simulated the strength and transmission efficiency of synaptic connections and completed learning and optimization by adjusting synaptic weights. Based on this, artificial neural network models were constructed.

[0012] Figure 2 The diagram shows the basic structure of the most primitive and fundamental neural network, the perceptron, which uses a three-input, single-output configuration. This structure includes three input signals. , , and their corresponding connection weights , , The middle section includes a weighted summation unit, used to perform the product summation operation of the input signal and the corresponding weight, and to add a bias term b. An activation function unit is then placed after this unit. The weighted summation result is then subjected to a nonlinear mapping process. The mathematical expression for this perceptron is: Its working principle is to first multiply each input signal by its corresponding weight and then sum them, then add a bias term, and finally use an activation function to achieve a nonlinear transformation and output the result. .

[0013] During network training, the weights are continuously adjusted and optimized through iterative learning. , , Ultimately, a set of optimal weight parameters is determined, which defines the mapping relationship between the input signal and the output result. Subsequent developments of various improved neural networks, including typical models such as Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), and Deep Neural Networks (DNNs), while featuring significant optimizations and improvements in network structure and implementation algorithms, have not fundamentally altered their core underlying computational logic. They all use linear weighted summation and nonlinear mapping as their basic computational units, continuing the fundamental working principle of neural networks established by the perceptron.

[0014] 3. Common defects of existing information processing systems

[0015] The deficiencies referred to here are the shortcomings of existing information processing systems compared to the working mechanisms of the human brain, especially in areas such as fuzzy information processing, common-sense reasoning, adaptive learning, and generalized understanding. Specifically:

[0016] (1) The information processing process takes independent things as its core task, and the common characteristics contained in different things do not have a natural sharing and association mechanism. For example Figure 1 In this system, the three input signals are processed and stored separately. Whether there is identical data or other similar characteristics in each signal is not reflected in the data processing structure. To retrieve similarities or duplicates among the information from each signal, additional dedicated processing programs are required. This is evident in various aspects; for example, different applications perform their functions independently, with no inherent connection between them; computer storage is in the form of independent files, making it impossible to know if two documents share common content; and there is no inherent correlation in the storage structure whether two or more movies feature the same actor.

[0017] (2) Artificial neural networks are merely formal biomimicry and do not truly construct the necessary relationships and knowledge sharing mechanisms between information. Although artificial neural networks borrow from the topological structure of the biological brain and the characteristic of information transmission between neurons through synaptic connections, and formally achieve distributed storage similar to biological neural networks, this distributed storage does not truly establish semantic relationships and knowledge sharing between information.

[0018] The root cause lies in the fact that, during the preprocessing stage of vectorizing and digitizing real-world information, neural networks only complete formal encoding, failing to structurally express the inherent meaning, hierarchical relationships, and entity attributes of the information itself. Network learning is essentially still a fitting of data distributions, rather than an understanding, summarization, and reuse of knowledge. Therefore, it lags fundamentally behind the human brain in areas such as few-shot learning, cross-task generalization, and interpretable reasoning. Furthermore, the random initialization of model weights and the gradient update process based on the loss function are entirely data-driven rather than oriented towards the connections and causal relationships of things in the real world. This means that what the model learns is essentially statistical regularities and geometric divisions in the data space, rather than an understanding and association of real-world phenomena.

[0019] Even with improved architectures like the Transformer incorporating attention mechanisms, they can only capture local and relative relationships within sequences or feature spaces, failing to achieve global, cross-modal knowledge association and reasoning similar to human cognition. Furthermore, current mainstream neural networks lack complete, explicitly reproducible memory capabilities, unable to accurately extract, fully reproduce, and consciously recall learned information like the human brain. Their storage is implicit, distributed weight distribution, rather than directly readable, interpretable structured memory.

[0020] (3) Artificial information processing systems, including various artificial neural networks, are essentially data-driven and computation-dominated systems, using mathematical models, algorithms, and numerical calculations as their basic means and core tools. This computation-centric model is significantly inadequate in dealing with ambiguous events, qualitative judgments, and experience-based decisions in the real world, and is far less efficient than the human brain.

[0021] When judging ambiguous events such as distance, strength, and closeness, humans rely on intuition, relative changes, and historical experience, without needing precise values ​​or rigorous calculations. In contrast, artificial systems must rely on precise values, which not only incurs huge computational costs but also require pre-set numerical conditions to make decisions, resulting in extremely poor flexibility.

[0022] Looking further, this data-driven mechanism leads to a huge waste of computational power. Take artificial neural networks as an example: they treat a set of input vectors as a whole, forcibly fitting and associating each vector combination, assigning weights to each component. Even if many components have no real connection to the final outcome, the model will still forcibly establish connections through training, essentially doing a lot of useless work. This is similar to determining the closeness of a person's relationships with others; humans would search layer by layer based on natural logical relationships such as surname, clan, and lineage. However, data-driven neural networks do not rely on such prior structures, but instead input everyone indiscriminately, forcibly statistically associating them. This is not only inefficient but also violates the natural hierarchy and interdependence between things in the real world.

[0023] (4) Currently, mainstream artificial intelligence models generally rely on massive amounts of data and large-scale sample training, which places extremely high demands on hardware computing power, computing speed, data transmission rate and storage capacity. During the model operation, a large amount of data needs to be called repeatedly, which not only consumes huge computing power and power resources, but is also limited by physical manufacturing bottlenecks such as chip process and circuit transmission, resulting in high overall resource consumption and cost.

[0024] Such systems typically rely on large cloud computing centers and server clusters to operate, making it difficult to deploy and run independently on ordinary standalone machines or smart terminals without a network. All types of terminal devices require a continuous network connection to the cloud to perform their functions, hindering flexible configuration and independent local operation in a standalone environment. Furthermore, because the model structure, training methods, and operational logic are all centrally determined by the cloud center, the terminal merely acts as an execution unit, lacking the ability for autonomous learning, training, and self-optimization. This prevents the development of personalized, differentiated local intelligence based on individual needs, severely limiting the space for user customization and intelligent evolution. Summary of the Invention

[0026] Purpose of the invention:

[0027] In view of the inherent defects and shortcomings of existing traditional computer systems and various artificial neural networks in information processing mechanisms, the purpose of this invention is to overcome the problems of modularization, isolation, lack of natural connection, reliance on data-driven and numerical fitting to achieve learning, and inability to recognize based on the inherent connections of things themselves in the existing technology, and to establish an information processing architecture and implementation method that is closer to the cognitive mechanism of the human brain.

[0028] The present invention aims to enable information processing systems to express and store external things in an elemental, hierarchical, and structured manner, so that the common features and inherent relationships between different things form a natural sharing and association mechanism, rather than relying on additional programs to achieve association retrieval.

[0029] Meanwhile, this invention abandons the learning method that only drives weight adjustment based on output error, enabling the system to perform targeted learning and association construction based on the objective attributes, internal structure and logical connections of things themselves, thereby breaking through the limitations of existing artificial information processing systems in fuzzy information processing, qualitative judgment, common sense reasoning, few-sample learning and cross-task generalization.

[0030] Furthermore, this invention aims to reduce ineffective calculations and computational power consumption, enabling the system to efficiently process and make decisions based on the natural hierarchical and interdependent relationships of things, without having to forcibly fit irrelevant information, and ultimately achieving more efficient, brain-like, and more realistic intelligent information processing.

[0031] II. Theoretical Model of Brain Neural Networks

[0032] To achieve the above-mentioned objectives, this invention is based on in-depth research into the structure and working mechanism of the brain's biological neural network. It fully follows the working mechanism and operating characteristics of the human brain, imitates its essential working mode, and combines the characteristics of computer systems to apply it to the artificial data processing process, so that the artificial system has a working mode that is closer to that of the biological brain.

[0033] Current research on the working mechanisms of the brain in fields such as biology, medicine, and neuroscience remains fragmented and localized. Existing research largely focuses on local aspects such as neuronal firing mechanisms and single information encoding methods, failing to develop a complete, unified, and systematic theoretical framework for brain information processing from a systems engineering perspective.

[0034] Taking the processing of sensory information such as vision and hearing as an example, the biological brain receives external information from sensory organs, converts it into neural impulse signals, transmits them to the brain center through neural pathways, and then the brain completes signal analysis, image formation, information recognition, memory storage and retrieval. Essentially, this constitutes a complete information input, transmission, processing, storage and output chain.

[0035] However, existing research has not yet established a complete theoretical framework for the brain's information transmission and processing system, and lacks a systematic understanding of the neural mechanisms underlying the actual operation of the brain. For example, key questions such as how information is transmitted within the brain, the specific meanings carried by neural impulses, how the source location is distinguished after information arrives in the brain, and the form in which information is expressed and stored, have not been given unified and complete theoretical explanations, making it difficult to reconstruct the brain's true working mode as a whole.

[0036] The brain neural network theoretical model described in this invention is the result of the inventor's long-term research on the relationship between consciousness and human body regulation. Utilizing systems theory and incorporating modern communication principles, the basic structure of communication systems, and information processing and transmission methods, the inventor systematically integrated and refined existing findings in brain science. Based on existing research revealing neuronal working modes, firing characteristics, neural network connection structures, and information transmission features, the inventor gradually constructed a complete information processing mechanism covering the entire process of brain information transmission, reception, expression, memory, and recall through systematic analysis and theoretical induction. The complete content of this brain neural network theoretical model has been elaborated in detail in the inventor's forthcoming book, *Consciousness: The Core Driving Force of Material and Life Evolution and Self-Regulation*, as well as the traditional Chinese electronic version, *The Core Driving Force of Consciousness Life Evolution and Regulation*.

[0037] The theoretical framework, hierarchical model, and core algorithms upon which this invention is based are all original achievements of the inventors and can serve as the theoretical foundation of this invention and a publicly available reference for distinguishing it from existing technologies. This invention is a complete technical solution further developed based on the aforementioned original theoretical framework.

[0038] To facilitate understanding of the technical solution of this invention, only the core content of the brain neural network model regarding information reception, expression, memory and recall will be briefly summarized below.

[0039] (I) Main characteristics and core concept definitions of the brain's neural network

[0040] 1. Information expression methods in the brain: the proposal of spacer neurons and quantification neurons

[0041] All the information (i.e., stimuli) received by the brain includes: visual, auditory, olfactory, and gustatory stimuli received through the four major sensory organs: eyes, ears, nose, and tongue; and internal bodily stimuli transmitted through receptors and peripheral nerves throughout the skin and body, responsible for sensing pressure, temperature, pain, and reflecting various physiological states. All of these internal and external stimuli constitute the original source of information for the brain.

[0042] Therefore, the brain must differentiate the received stimuli: first, it must differentiate the source of the stimulus, i.e., which sensory organ or part of the body the stimulus comes from; second, it must differentiate the type of stimulus, i.e., whether it belongs to the visual, auditory, temperature, pain, or other signal categories; and third, it must differentiate the intensity of the stimulus, i.e., the strength of the signal such as visual brightness, sound volume, or temperature level. We call the stimulus location, stimulus type, and stimulus intensity the three elements of information reception in the brain, and these three elements can uniquely identify a stimulus.

[0043] Neuroscience research shows that for all types of stimuli, the neural pathway from the receptor to the brain is transmitted using the same form of nerve impulses. The type and source of the stimulus cannot be distinguished solely from the nerve impulses themselves. Therefore, it is reasonable to infer that the brain relies on specialized neurons to receive specific nerve impulses from specific locations in the body and specific types of stimuli. Receptors at specific locations transmit signals to corresponding specialized neurons through dedicated neural pathways, thus uniquely determining the source and type of stimulus based on the activated neurons. This mechanism is consistent with the laws of neural transmission revealed by the specific neural energy law. The intensity of the stimulus is determined by the firing frequency of the nerve impulses.

[0044] These neurons, specifically designed to correspond to a particular stimulus location and type, are called "space-occupying neurons." Their significance lies not only in determining the stimulus location and type, but also in representing the projection of body parts onto the cerebral cortex. The brain determines the location of the stimulus through this one-to-one projection relationship (see Penfield's "cortical man"). This mechanism is analogous to the correspondence between actual geographical locations and maps.

[0045] The locating neuron identifies the source and type of stimulus. The remaining key question is how to express stimulus intensity in the brain—that is, how the brain converts and expresses neural impulses of different frequencies into corresponding stimulus intensities. In artificial systems, this problem is relatively easy to solve; the impulses can be counted, integrated, or otherwise converted into numerical values ​​of different magnitudes for display and storage. However, from the structural characteristics of the brain's neural networks, there is no functional mechanism corresponding to this artificial processing method; at least, no structural evidence of such functionality has been found so far. Therefore, we further propose the concept of "quantitative neurons."

[0046] like Figure 3 As shown, , , This represents three spacer neurons, each responsible for receiving three different nerve impulses. These neurons can be receptors from different parts of the body with the same or different types of stimulation, or they can be receptors from the same part of the body with different types of stimulation. For example, they could correspond to sensory cells from different locations on the retina, or receptors from the same location on the body that correspond to temperature, pressure, and pain, respectively.

[0047] Assuming the three stimulus signals are as follows Figure 1 Thus, they each have intensity levels of L, M, and N. In artificial information processing systems such as computers, these levels are stored as data in corresponding storage units; while in the brain's nervous system, after each locating neuron, L, M, and N neurons are branched off, each "representing" a different stimulus level, hence they are called "quantitative neurons," i.e. Figure 3 shown ( , , ..., ), ( , , ..., ), ( , , ..., ).

[0048] The rationale and basis for designing this type of quantified neuron in this study are mainly reflected in the following two points:

[0049] First, from the perspective of brain neural network structure and biological evolution, no structure similar to that of a computer capable of data conversion, storage, and numerical computation has yet been discovered in the brain's neural network. Furthermore, from the perspective of biological evolution and survival, all information processing methods employed by organisms are based on the principles of efficiency and energy conservation. Energy conservation requires simple structures and high operational efficiency; complex structures inevitably lead to higher energy consumption, and energy supply is crucial for survival for all organisms. In addition, biological reactions must be rapid, as rapid reactions are directly related to survival; therefore, the brain does not have the time for complex calculations and conversions. From a structural and evolutionary perspective, the brain is more suited to extremely simple and direct processing methods than complex numerical computation and storage models.

[0050] Second, from the physiological mechanism of neuronal impulse release and transmission, neurons release and transmit neurotransmitters through the presynaptic and postsynaptic membranes. The intensity of stimulation corresponds to a corresponding amount of neurotransmitter. These neurotransmitters are packaged in multiple vesicles and released sequentially in units of vesicles, thereby generating nerve impulses. This process essentially discretizes the intensity of stimulation: the total amount of neurotransmitter is determined by the intensity of stimulation, and the number of vesicles released is directly related to the number of nerve impulses, thus establishing a correspondence between nerve impulses and stimulation intensity. This is an objective phenomenon that has been confirmed by physiological research.

[0051] When a nerve impulse is transmitted to the occupant neuron responsible for receiving it, the signal is received according to a mechanism that is the opposite of but consistent with the release mechanism: a single nerve impulse triggers a corresponding number of neurotransmitters, and the more nerve impulses there are per unit time, the greater the total amount of neurotransmitters released; then, based on the total amount of neurotransmitters, the occupant neuron selects and opens different neural pathways, thereby activating neurons at different quantification levels.

[0052] Through the above mechanism, the frequency of a neural pulse can be mapped to a specific neuron, thus achieving a one-to-one correspondence between neural pulse frequency, quantified neuron, and stimulus intensity. This model can also be implemented in artificial systems: for example, the total charge is stored in a large capacitor, then distributed to multiple small capacitors through charging, and then the small capacitors discharge sequentially to form a sequence of pulses. This is also the simplest and most energy-efficient implementation method in circuit systems.

[0053] 2. The brain recognizes stimuli through sensory perception.

[0054] Within artificial systems, information is primarily expressed and transmitted using optical and electrical signals. However, these signals cannot be directly received and distinguished by humans. Therefore, artificial systems must convert internal electrical signals and other low-level signals into visual and auditory forms that humans can recognize through devices such as displays and speakers. These displays and speakers act as interfaces between the artificial system and humans, allowing humans to understand the system's operational status and processing results.

[0055] Similarly, in the brain, nerve impulses transmitted through nerve fibers carry information such as the type and intensity of stimulation. However, the conscious mind cannot directly recognize and distinguish the nerve impulses themselves, so they need to be converted into a form that the brain can recognize. This form is the subjective sensation generated by the brain, which includes, but is not limited to, various forms of perception such as vision, hearing, smell, heat, and pressure.

[0056] The formation of subjective feelings is not achieved through numerical calculations, but is directly triggered by the fixed correspondence between quantified neurons and the sensory formation area. Figure 3 Space-occupying neurons and its corresponding quantized neurons ~ For example, a sensory formation zone is set as the functional unit that ultimately produces a sensation from the stimulus in that pathway.

[0057] The sensory formation area has L input terminals, each corresponding to a specific sensory intensity level, essentially a pre-set multi-level intensity channel. The input signals at each terminal are identical; the generation of sensation depends solely on the activated input terminal. Therefore, by connecting quantified neurons representing different stimulus intensities to different input terminals in the sensory formation area, each input terminal will generate a sensation of corresponding intensity when it receives a neural impulse.

[0058] Therefore, from the moment a stimulus is received by a receptor to the moment the brain generates a corresponding sensation, the entire process does not require complex numerical calculations, logical operations, or other forms of complex data processing. The sensation is generated solely through a fixed mapping relationship between the stimulus → nerve impulse frequency → occupant neuron → quantized neuron → corresponding input terminal of the sensory formation area. The mechanism is achieved by pre-setting the structure, similar to multi-level volume control: when a certain level is activated, a corresponding fixed volume is output, without the need for real-time calculation of the volume value.

[0059] Each type of sensation in the brain corresponds to a sensory formation area, employing the structure described above. For example, if If it is a spacer neuron that is a sensory cell at a certain location in the retina, then ~ It can correspond to the 256 gray levels used in similar artificial systems when processing video signals; the corresponding sensory formation area corresponds to a pixel unit in the visual formation area of ​​the brain.

[0060] 3. The essence of information: the various manifestations of things and the relationships between things.

[0061] The essence of information lies in reflecting the external manifestations, internal structure, various characteristics and attributes of objective things, as well as the interrelationships between different things. The same thing can be expressed through multi-dimensional and multi-level external phenomena and internal characteristics. These various characteristics complement and explain each other, collectively forming a complete description of the thing. Information not only embodies the thing itself but also the connections, interactions, and correspondences between things. The memory and storage of information by biological neural networks is essentially the recording and solidification of various manifestations of things and their interrelationships, enabling living organisms to perceive environmental changes, identify natural phenomena, and grasp the laws governing survival-related things through this information, thereby achieving understanding and adaptive responses to the external world.

[0062] No single attribute or feature can constitute complete and meaningful information; isolated information units have no practical informational value. For example, recording only a person's name without establishing connections to their physical characteristics, identity information, social relationships, or other relevant features makes that name merely an isolated symbol, lacking complete informational meaning because the object, identity, and related characteristics it refers to are unknown. Whether it's the natural memory of biological neural networks or the data recording of artificial systems, the same principle applies: completely isolated and unconnected data units do not exist. Only when a single piece of data establishes a correspondence with other information units, forming a mutually reinforcing and interpretive relational structure, can it form complete information with practical meaning, thereby realizing cognitive, comprehension, and application value.

[0063] 4. The core of the brain's working mode: conscious choice and neuronal plasticity

[0064] The core of the brain's working model lies in the synergistic effect of two major mechanisms: conscious selection and neuronal plasticity. The formation of all information as memory depends on conscious active selection. Only when consciousness pays attention to, focuses on, or shows interest in information can that information be received by the brain and form a memory. Without conscious participation, even if the senses receive visual or auditory signals, effective memory cannot be formed. For example, the habitual action of locking a car door, if not consciously focused on and associated with, will not leave a corresponding memory in the brain, leading to a lack of confirmation that the action was completed. This clearly demonstrates that conscious selection is the prerequisite and key for information to enter the memory system.

[0065] At the same time, the vast number of neurons in the brain are plastic, capable of generating excitation or inhibition under the brain's control, and achieving dynamic connections. Information selected by consciousness and its associated objects can establish connections between corresponding neurons under the brain's control. This connection and remodeling between neurons is the physical basis for the brain to form memories and store information.

[0066] Under the combined influence of conscious selection and neuronal plasticity, the content memorized in the brain must be interconnected; conversely, things that are not interconnected cannot form stable and effective memories in the brain.

[0067] 5. Definition of the elements of a thing

[0068] To accurately describe the brain's mechanisms for perceiving, recognizing, and remembering information, and to distinguish it from artificial signal processing systems that rely solely on threshold responses, this invention proposes relevant definitions of elements. The brain's processing of external stimuli does not solely depend on receptor thresholds generating neural impulses. After the stimulus signal is transmitted to the brain, it only forms a basic physiological response. It requires conscious selection, attention, and processing by the brain to form effective perception and memory. Therefore, this invention defines an element as a perceptual unit that can be clearly recognized by the brain's consciousness, possesses a certain stimulus intensity, has corresponding meaning, and can be recognized by the brain.

[0069] Among them, the smallest necessary unit that can fully support the execution of a specific target task at the current descriptive level is called the basic core element. Here, "smallest" refers to logical necessity: some scenarios are physically indivisible, while others do not require further subdivision. The basic core element is the basic unit that makes up or describes things. All things are formed by combining basic core elements according to different relationships, thus constituting sub-elements at different levels and forming a hierarchical cognitive structure.

[0070] Taking the Chinese cognitive system as an example: strokes are the indivisible basic core elements that constitute Chinese characters. Strokes combine to form sub-elements of Chinese characters; Chinese characters further combine to form higher-level sub-elements such as words, phrases, and sentences. Each level of element progressively constitutes a complete cognitive object. When the system's processing target is the composition of Chinese characters, strokes can serve as the lowest-level basic core element. At this point, the basic core element is indivisible, and further subdivision would be meaningless. When the system's processing target is language comprehension and semantic expression, there is no need to disassemble the internal structure of a single character. In this case, a single Chinese character can serve as the basic core element, and words, phrases, sentences, etc., successively constitute higher-level sub-elements.

[0071] By adopting this hierarchical division method that adapts to the task objectives, the system can be broken down to the smallest unit required to complete the current task, without unnecessary excessive splitting, thereby simplifying the structure, improving processing efficiency, and giving the entire element system good adaptability and flexibility.

[0072] The difference between the basic core elements and the sub-elements lies not only in their size and hierarchy, but more importantly, in the fact that the basic core elements consist of a single thing, such as a single visual stimulus or a single auditory stimulus, without including other forms of stimulation; while the sub-elements embody the relationships between things, which can include different types of things or stimuli, for example, formed by the interrelationship between visual and auditory stimuli. This distinction is also an important characteristic of the brain's neural network, providing a fundamental definition for the subsequent construction of the brain cognitive network model in this invention.

[0073] (II) Memory Formation in the Brain's Neural Networks

[0074] When the brain processes information, its functional logic inevitably follows a hierarchical structure, progressing from basic core elements to complex concepts layer by layer. Although neurons in the brain are distributed in a lattice pattern, rather than being strictly hierarchical like those in artificial systems, there is a certain order in the selection of blank neurons during the construction of neural networks: regions closer to quantified neurons are used more frequently and naturally form a structure with hierarchical characteristics. However, this hierarchy is not absolutely strict (for example, deep neurons can be directly connected to basic core element neurons to achieve flexible associations of complex concepts).

[0075] To facilitate the analysis and construction of memory networks, this approach employs a hierarchical method for explicit and structured description, dividing the neural network into multi-layered structures. This method not only aligns with the core characteristics of brain functional levels but also makes the model logic clearer and the analysis more convenient, meeting the intuitive requirements of manual modeling.

[0076] 1. Formation of the first layer of memory network

[0077] Figure 4 This is a schematic diagram of the brain's memory network structure, where layer 0 is the foundational core layer, composed of neurons. - These are referred to as basic core element neurons, each corresponding to a different basic core element that the brain can recognize. For example, , They can represent two different graphics respectively, and It can represent a short, recognizable sound, etc. Based on the hierarchical structure of the brain's cognitive processes from simple to complex, a memory network structure organized hierarchically according to things is constructed. The basic memory process is as follows:

[0078] (1) Stimulus triggering and sensory generation: External stimuli cause multiple basic core element neurons to be excited simultaneously or in a certain temporal sequence (e.g. and (This process generates corresponding sensations in the sensory generation area, which are then recognized by the brain's consciousness.)

[0079] (2) Consciousness focus and neuronal activation: The brain’s consciousness focuses its attention on the basic core element neurons that are excited, and clearly perceives that these excited neurons are related in time, space or logic, that is, recognizes that they represent different attributes or aspects of things; the basic core element neurons that are noticed and selected by consciousness are kept in an excited state, while other neurons that are not noticed are inhibited or ignored.

[0080] (3) Matching existing memory connections: The brain's consciousness examines the neurons of the basic core elements that are currently excited and determines whether they have established a connection with a neuron in the first layer of storage. If the connection exists, it means that this combination of stimuli has been stored, and the brain's consciousness will perform subsequent operations (such as strengthening the connection or recalling the memory) according to actual needs; if not, proceed to the next step;

[0081] (4) Formation of new memory connections: Select a blank neuron that has not yet established a connection in the first layer of storage area (e.g., The pattern neuron that is currently excited is connected to the blank neuron, and the blank neuron represents the combination of stimuli perceived by the brain consciousness.

[0082] (5) Repetition and network formation: When a new stimulus is received, repeat steps (1)-(4); by repeatedly performing the above process, the first layer of the memory network is gradually formed.

[0083] Except for the basic core element neurons in layer 0, which represent a single, recognizable stimulus of a specific type, the neurons in other layers represent combinations or connections of any stimulus type, collectively referred to as sub-element neurons; these neurons are represented by letters. The subscript is a two-digit number, where the first digit represents the floor number and the second digit represents the sequential number on that floor (e.g., ...). Refers to the first neuron in the first layer. (Referring to the third neuron in the second layer). The first-layer network established according to the aforementioned connection rules and steps contains... - There are a total of 5 sub-element neurons. Among them, With the 0th layer P3 connection indicates , Constituting the same thing, by Representative, or conversely, things. Include and These are the two fundamental core elements. Similarly, and , connect, and , connect, and , , connect, and , Connection. If things are represented as a set of elements, then... , , , , .

[0084] 2. Formation of the second and higher layers of memory networks

[0085] The formation of the second-layer network also follows the connection rules of the first layer: starting with the nodes of the first layer ( Based on this, when two or more nodes are excited at the same time, it is first checked whether these excited neurons have already established connections with neurons in the second layer. If so, it means that the memory has been formed; otherwise, blank neurons in the second layer are selected to connect with them to form nodes representing more complex things. Figure 4 In the middle, the second layer contains three neurons. , , , From the first layer , , Simultaneous excitation connects, that is Include , and Represented by sets: Similarly, , The formation of higher-level networks follows the same rules, building upon the nodes of the previous layer and progressing layer by layer according to the same connection rules to construct a more complex overall network structure representing things.

[0086] (II) The brain's neural network recall process

[0087] As the foregoing has shown, the memory process is essentially the establishment of neural connections between all the fundamental core elements that constitute a thing, arranged hierarchically. The occurrence of a thing corresponds to the neurons of the fundamental core elements that constitute that thing being excited sequentially according to a certain pattern; and the memory process is precisely the process of fixing and storing the excitation patterns of these fundamental core element neurons in the form of neural connections.

[0088] Therefore, the memory of things is essentially a reproduction of the scene of excitation of the basic core element neurons. When the basic core element neurons are reactivated and generate excitement, corresponding perceptions and sensations are formed in the brain, which are then perceived and experienced by our consciousness.

[0089] In the brain neural network model upon which this invention is based, according to the signal transmission characteristics of neurons, although the nerve impulses of a single neuron are transmitted unidirectionally from dendrites to axons, bidirectional transmission of nerve impulses can be achieved through feedback connections between two neurons, thus endowing the defined neuron model with bidirectional signal transmission characteristics. Based on this characteristic, Figure 4 The memory network shown can then possess the function of recall.

[0090] The process of recall in the brain is complex and influenced by a variety of factors. In the brain's memory network, the excitation of any neuron can trigger a chain reaction in its connected neurons. Therefore, external stimuli, spontaneous neuronal excitation, and conscious selection by the brain can all trigger recall behavior, and the recall process is controlled by conscious thought. The following explanation uses external stimuli as an example to illustrate the recall process in the memory network.

[0091] exist Figure 5 In this context, we assume that the fundamental core element is the neuron. Stimulated (i.e., received again) The things they represent, and the feelings they evoke, are also selected by attention. In a state of excitement. The excitation will be transmitted to all neurons connected to it. The first pathway, following the dotted arrow, shows the sequence of excited neurons: The second path, following the direction of the solid arrow, shows the order of neuronal excitation as follows: The third route, also along the direction of the solid arrow: Therefore, the core neurons responsible for the excitation induced by P3 include P1, P2, P4, and P6. These neurons, through reverse stimulation of the quantified neurons they connect to, generate corresponding sensations through the sensory generation area. Of these sensations, those selected by the brain's consciousness appear in awareness, and the relevant elements are recalled, while the remaining neurons and associated neurons are temporarily inhibited.

[0092] (III) Typical characteristics of the brain's neural network

[0093] 1. Mesh structure, distributed information storage, and parallel information processing

[0094] The brain's neural network is a mesh structure, not a tree structure, exhibiting typical characteristics of distributed information storage and parallel information processing. This mesh structure differs from a closed, hierarchical tree structure. For example, a family tree may form a single, clearly branched tree structure based solely on blood relations, but individual nodes within the tree are not isolated; they are inevitably connected to a wide range of social information. Social elements such as name, workplace, and contact information create multi-faceted connections between the node and other external nodes, transcending a simple tree structure. Therefore, the complex and real-world information connections inevitably form an open and interconnected mesh structure. Distributed storage refers to representing a complete entity as a collection of basic core elements. The neural network encodes and interconnects the basic core elements contained within the entity and the temporal, spatial, and logical relationships between these elements through neurons at various levels of sub-elements. A complete entity is collectively represented by the relationships distributed across different neurons. The neurons at each level do not carry complete, real information themselves; they are only used to represent the relational characteristics between elements, thus achieving distributed information expression. Based on this mesh structure and distributed storage, parallel information processing is achieved. When a neuron is activated, all related neurons can be activated synchronously, enabling information to be transmitted and responded to quickly and in parallel within the network.

[0095] 2. Global element sharing and association

[0096] The second major characteristic of this brain neural network is the interconnection and sharing of elements, which is also the core of the logic behind this invention's element-sharing data processing architecture based on the human brain. As can be seen from the aforementioned structure and working characteristics, the brain first analyzes and decomposes various things and perceptual information (such as sound, light, and other stimuli) from the external world, extracting the corresponding basic core elements. These basic core elements are then used as the basic units to construct and represent all complex things; that is, all complex things are essentially composed of combinations of basic core elements. During the process of memorizing and representing things, similar or related elements corresponding to different things can freely establish associations. As long as there is a logical or actual connection between elements, they can form mutual connections in the neural network. Simultaneously, various elements already existing and activated in the brain are shared by the network as public elements. The same element can be referenced and reused by multiple different things, without repeated memorization and redundant storage of the same single information. All elements achieve efficient sharing and reuse through association relationships, thereby greatly improving data processing and information storage efficiency.

[0097] 3. Unique information filtering mechanism

[0098] The third major characteristic of the brain's neural network is its unique information filtering mechanism, which mainly comprises two levels: First, a filtering mechanism based on conscious attention. While the human body's sensory organs (eyes, ears, nose, tongue, body, etc.) continuously receive massive amounts of external information in a conscious state, only information selected and deemed useful or interesting by conscious attention can enter the neural network to form effective memories and be further processed. Attention permeates the entire process of memory, processing, and recall, enabling the targeted filtering and extraction of effective information from a vast amount of data. Second, a dynamic filtering mechanism based on the plasticity of neuronal synapses. Neurons form special connections through synapses. These synaptic connections are not fixed and rigid but possess adjustable conduction efficiency and connection strength. When nerve impulses are repeatedly transmitted between synapses, the higher the frequency of use and the longer the duration, the stronger the synaptic connection becomes, and the higher the conduction efficiency. Conversely, synaptic connections that are not used for a long time will gradually weaken, and the conduction efficiency will decrease. The brain relies on this synaptic plasticity to dynamically filter and strengthen information. Frequently used information, which is more important or sensitive to the brain, has its corresponding synaptic connections continuously strengthened, making it easier to activate and recall. Conversely, the synaptic connections corresponding to information that has not been used for a long time naturally weaken, becoming increasingly difficult to retrieve or even forgotten. This mechanism makes frequently used and valuable information in the neural network easier to activate and retrieve, while information that has not been used for a long time gradually weakens, thus enabling the priority retrieval of important information from a massive amount of memory.

[0099] However, in existing artificial neural networks, synaptic connection efficiency is usually directly equated to network weights, and the magnitude of the weights is used to characterize their contribution to information processing. This approach differs fundamentally from the working mechanism of the brain's neural networks: in artificial neural networks, changes in weights directly alter the mapping relationship between input and output, thus fundamentally affecting information representation and processing results; while in the biological brain, synaptic connection efficiency only reflects the ease of information retrieval and activation intensity, and changes in its intensity do not change the information nature and memory content carried by the neuronal connection structure.

[0100] This further leads to a fundamental difference in efficiency between the brain's neural network and conventional artificial neural networks. The information storage in the brain's neural network is essentially the establishment of connections between neurons. This process can be completed instantly, achieving information recording and representation without extensive iterative training. The brain's learning process involves repeatedly stimulating established neuronal connections to continuously strengthen corresponding synaptic connections, thereby achieving solid memory of knowledge. The thinking process, on the other hand, involves comprehensively processing existing knowledge and elements, forming new neuronal connections between previously unconnected information, generating new knowledge through the construction of new connections, and completing the understanding, reasoning, and creation of information. Artificial neural networks, however, misunderstand the essential meaning of the brain's synaptic connection efficiency, reducing the learning process to a training process of adjusting and fitting network weights. This requires massive amounts of data and extensive iterative calculations to continuously optimize weights in order to complete information learning and storage. This is not only extremely time-consuming and inefficient, but also unable to achieve the brain's instant connection, rapid memory, and autonomous innovative thinking. The two differ fundamentally in their working mechanisms and efficiency.

[0101] More importantly, existing artificial neural networks are essentially mathematical mapping models, establishing only a mapping relationship between input and output. During the learning process, they only form a fitting relationship for the corresponding output results, unable to fully store and represent the original information itself, nor do they possess the ability to completely restore and reproduce information. This is fundamentally different from the brain's ability to completely remember and accurately recall information. Artificial intelligence models based on such neural networks lack true content memory and experiential memory, cannot actively retrieve and recover historical memories and past experiences, and cannot rely on historical experience to achieve continuous learning and reasoning like the human brain. They have fundamental defects in memory mechanisms and recall abilities.

[0102] III. Technical Solution

[0103] This invention discloses a human brain-based element-sharing data processing architecture and its implementation method, which is designed based on the theoretical model of the human brain's biological neural network. The human brain's biological neural network processes information based on the physical connection structure between neurons. Neurons can dynamically and plastically connect according to information processing needs, completing the perception and quantitative expression of information. This architecture and method aim to adapt the aforementioned core working characteristics and mechanisms of the human brain's biological neural network to information processing systems composed of computers and various artificial data processing hardware. Considering the inherent characteristics of computers and various artificial data processing hardware, their hardware circuits are pre-built and lack the real-time, dynamic, and plastic connection capabilities of human brain neurons, making it impossible to completely imitate the physical connections of the human brain's neural network. Therefore, this invention, targeting the characteristics of computer hardware architecture, transforms the physical connection relationships of human brain neurons into data structures and relationships that can be described in computer language, realizing the practical adaptation of the biomimetic brain neural network on the computer.

[0104] The specific mapping and adaptation scheme is as follows: A hierarchical storage architecture is constructed based on the computer's memory hardware. Following the hierarchical characteristics of the human brain's neural network, each layer of neurons is mapped to a corresponding memory. This memory can be configured as an independent memory list, and each memory can serve as an independent computing unit, configured with a dedicated computing core responsible for reading, accessing, and other data operations within its own memory list. Each individual neuron in each layer is mapped one-to-one to a single storage unit in the corresponding memory. The storage unit represents the neuron, and the content stored within it describes the connection relationships between the corresponding neuron and other neurons; that is, the content of the storage unit indicates the connection points between the neuron corresponding to that unit and other units. From the perspective of hierarchical association of information input and output, the connection relationship between the neuron corresponding to any layer of storage unit and the neuron corresponding to the lower-level storage unit represents which lower-level neurons constitute that neuron, i.e., recording its association with related lower-level neurons; its connection relationship with the neuron corresponding to the upper-level storage unit represents that neuron as a component of the upper-level structure, i.e., recording its hierarchical affiliation within the upper-level structure.

[0105] Based on the core mapping relationship between the human brain's neural network and computer storage architecture, this architecture and method decomposes the target processing object into indivisible and indispensable basic core elements according to its inherent attributes. There is no unified or fixed pattern for the division of these basic core elements; it can be flexibly implemented according to the characteristics of the specific processing object and the desired outcome. Regardless of the division method used, the target processing object can be decomposed into basic core elements, and these elements can be combined in the order of decomposition to form a complete description of the target processing object. Each of these basic core elements corresponds to a specific neuron in the human brain's biological neural network and also to a specific storage unit in the underlying memory of the hierarchical storage architecture. The target processing object is constructed from these basic core elements according to a preset hierarchical relationship, encompassing data carriers, abstract concepts, and logical relationships. The absence or alteration of any basic core element will result in an incomplete description of the target processing object or directly change the inherent attributes of the target processing object. In an information processing system, fundamental core elements, carrying real attribute information, reside in the underlying storage module at fixed storage addresses. They serve as the unified information source for the entire system's data processing workflow, allowing direct or indirect access by various functional components. The system constructs a multi-level element structure based on these fundamental core elements, collectively referring to elements at each level above them as sub-elements. The construction of sub-elements can incorporate an attention mechanism, selecting and combining suitable fundamental core elements according to actual processing needs. Each sub-element does not store specific attribute information of the fundamental core elements; it only records its own relationships with fundamental core elements or other sub-elements. These relationships include temporal, spatial, and logical relationships between the fundamental core elements constituting the sub-element and other sub-elements, as well as other definable or describable actual or abstract relationships. The target processing object comprises several independent events. Each independent event does not independently possess its own dedicated fundamental core elements and sub-elements at each level. All elements corresponding to events exist in a distributed manner within the multi-level element structure. Event processing and expression are completed by calling the underlying, uniformly stored fundamental core elements and the relationships between these elements.

[0106] Furthermore, the element sharing and hierarchical structure described in this technical solution are not artificially set local rules, but rather a global collaborative mechanism based on the essential attributes of objective things and the laws of life perception.

[0107] First, the element sharing mentioned is a global sharing across the entire domain, system, category, and event, rather than a local sharing limited to a single event, task, or type of thing. From the basic core elements to the sub-elements at all levels, regardless of whether they correspond to the same or different things, or the same or different attributes—such as visual features, auditory sounds, temperature perception, odor signals, food attributes, danger signals, and other completely different things—all such attribute information exists uniformly within the system, is uniformly invoked, and is shared globally, without belonging to a particular local scene or independent task. From the perspective of life evolution and survival perception, organisms do not perceive a certain type of thing or attribute in isolation, but rather form effective cognition by establishing relationships between different things and attributes. For example, associating the attribute element "temperature change" with "food acquisition," or associating heterogeneous attribute elements such as "the sound and smell of a specific animal" with "survival safety and danger warnings," these are all key relationships between different things and attributes. Therefore, the system must support global element sharing across things, attributes, and events to achieve a complete expression and simulation of real-world relationships.

[0108] Secondly, there is a necessary, inherent, and inseparable correspondence between element sharing and hierarchical structure. From the perspective of objective things in nature, everything possesses diversity and a hierarchical structure; nothing has only a single attribute. The composition, structure, attributes, and scale of things naturally exhibit hierarchical characteristics from bottom to top and from simple to complex. For example, an individual animal contains multiple levels of attributes such as species, morphology, sound, smell, and behavior; a natural event contains multiple levels of structure such as environment, time, object, and result. From the perspective of information description and expression, the cognition and description of things is essentially a complete depiction of their various attributes, components, and relationships; therefore, a matching hierarchical structure must be used for expression. Because things themselves possess a natural hierarchical structure and diverse attributes, the perception, analysis, and modeling of things by organisms inevitably follow hierarchical rules. The hierarchical structure in this technical solution is precisely the technical realization of this objective law.

[0109] Therefore, by expressing things hierarchically, the basic core elements and each layer of sub-elements can form relatively independent information units, which can correspond to independent neuron nodes in a neural network structure. This is why elements can be shared by other different things. For example, sub-element C in a system is composed of sub-element A and sub-element B. When describing and expressing C, it needs to be indirectly achieved through the relationship between A and B; this is the core way to define and express C. This is similar to explaining the phenomenon of lightning, which usually involves the mutual corroboration of thunder and lightning. For example, thunder is accompanied by lightning, and seeing lightning often corresponds to thunder. It can also be associated with weather phenomena such as wind and rain, where lightning, wind, and rain are all related characteristics and attributes of the lightning phenomenon. At the same time, it is precisely because element A exists as an independent element in the system network that it can be reused as a shared unit by other sub-elements. For example, if sub-element D also contains element A, then element A can be a common component of both C and D, and can be shared globally, uniquely exist, and reused in the system without having to create separate independent elements A for C and D. In this case, C and D form a natural connection through sharing element A, enabling different things, events, and structures to achieve cross-object and cross-scenario feature association and information exchange through sharing underlying elements. This is the core manifestation of the element sharing mechanism in this solution.

[0110] Under this mechanism of global sharing and natural hierarchy, the basic core elements, as the smallest identifying units that distinguish the attributes of things, are shared globally within the system. Sub-elements at each level do not store independent substantive attribute information, but only record the association and combination relationships between elements. The system constructs a hierarchical structure of sub-elements from the basic core element layer upwards, with the complexity of each level increasing progressively, forming a complete system from simple to complex, from bottom-level attributes to upper-level complex information. At the same time, all elements have bidirectional traceability across all levels. Any sub-element can be traced upwards to its upper-level structure and downwards to its corresponding bottom-level basic core element, thereby achieving cross-thing, cross-attribute, and cross-level association fusion and associative activation. This enables the entire system to simulate the associative memory and complete information recall based on association relationships in the biological brain.

[0111] Furthermore, the relationships between the elements formed by the memory network described in this technical solution originate from the inherent constituent attributes of things themselves, and belong to a broad sense of association. These relationships not only include strong associations based on physical laws, chemical principles, structural relationships, logical relationships, and causal relationships, but are also not limited to the traditional forms of association mentioned above. Instead, they cover a wider range of associations that are closer to real-world cognition. Such associations are often established simply because several related factors belong to the same life scenario or spatiotemporal scenario, that is, they appear together at the same time, in the same environment, or during the same event, thus naturally forming an association and being remembered. For example, a person can naturally associate and remember information such as the weather conditions, travel destination, people encountered, and events that occurred on a particular day. These elements may not have strict logical or causal meaning, but because they appear together in the same life scenario in chronological order, a stable association is formed in memory. The fundamental reason why the brain can effectively remember things is that the elements contained within things are naturally related. In the brain's memory mechanism, there is no isolated element that exists independently and has no connection with other elements. If an element is completely independent and not associated with any other information, it has no meaning at the cognitive and memory level and cannot be effectively remembered and retained. Therefore, the formation of memory based on association is an inevitable law of biological cognition and memory mechanisms.

[0112] In contrast, the training methods of existing artificial neural networks and large language models rely on input corpora, documents, and texts that already possess objectively existing connections. Characters, words, sentences, and contexts all follow established grammatical rules, language structures, and usage habits, inherently possessing complete and stable relationships. Under this premise, existing models still ignore these naturally occurring connections, employing random weight allocation, probability statistics, and fitting calculations to re-find and construct connections between elements. This is essentially redundant and unnecessary, significantly increasing computational overhead and training costs. Furthermore, it prevents the model from truly reproducing the inherent connections between things, resulting in a lack of interpretability in the memory and reasoning processes. This fundamentally differs from the efficient, natural, and direct memory mechanisms of the biological brain. The memory network of this invention directly relies on and retains the extensive connections inherent in the elements themselves, eliminating the need for additional algorithms to re-find, calculate, or fit connections. Structurally, it achieves a relational memory mechanism consistent with the essence of biological memory, possessing higher operational efficiency, cognitive rationality, and practicality.

[0113] Furthermore, the technical solutions and methods described in this invention can be applied to computers, data processing equipment, and various data processing systems. Its core idea lies in breaking down all operations, machine instructions, and computational processes involving various hardware components in a computer system into basic core elements, and forming sub-elements at various levels based on these. Regardless of the operating system or application program, as long as a corresponding operation is involved, these common basic core elements and sub-elements at various levels are uniformly invoked, achieving hierarchical sharing and association. Simultaneously, different types of information entities processed by the system, such as text, data, audio, and video, are broken down into basic core elements and sub-elements at various levels according to their characteristics and event types, and are uniformly organized, associated, invoked, and shared according to the hierarchical relationship of this invention.

[0114] Based on the above architecture, this invention reconstructs the data storage and management structure of existing computer systems. All applications, program instructions, various files, documents, and audio / video data are uniformly stored in the same memory network, similar to the memory mechanism of the biological brain. The brain does not have a scattered, independent folder-like storage structure, but rather organizes information uniformly through associative memory. Under this invention's architecture, there are no longer scattered, independent files and data entities in the system; only the memory network file needs to be loaded onto the disk to achieve unified management of all content. During retrieval and retrieval, users only need to enter keywords to extract target content from the memory network.

[0115] Because all information within the system is organized based on fundamental core elements and sub-elements at various levels, identical elements can be globally shared and reused without being stored repeatedly. Different files, different data, and different media resources can automatically form associations by referencing common fundamental core elements or sub-elements at various levels. For example, in existing storage architectures, two movies are stored as independent files, and even if they contain the same actors, the same scenes, or similar storylines, there is no connection between them. However, under the memory network architecture of this invention, two movies can be automatically linked together because they share the same fundamental core elements such as actors and scenes. When a user searches for one movie, other movies starring the same actor can be directly associated with it; when searching for a specific actor, all film and television works in which that actor has participated can be immediately retrieved, truly realizing brain-like associative memory, associative retrieval, and intelligent data management, fundamentally overcoming the shortcomings of existing computer systems such as independent files, fragmented data, and inability to be associated and reused.

[0116] Furthermore, the technical solution described in this invention can also be applied to the Internet, cloud computing, blockchain, and distributed / decentralized data processing systems. The multi-level memory network architecture of this invention simulates the connection mechanism of neurons in the biological brain. Its core feature is that the various element units in the network only exhibit associative connections, unrestricted by spatial location or regional scope. Regardless of physical distance, as long as an association exists, stable connection and indexing can be achieved. Based on this feature, the multi-level network of this invention can be seamlessly extended to distributed networks covering large areas or even globally. The basic core elements required by the system can be flexibly deployed according to actual needs. They can be distributed and stored on various distributed nodes to improve computing and response speed, or centrally stored on the same node for unified management. Regardless of the deployment method, the number, address, and identifier of the basic core elements remain unique and fixed throughout the entire network, ensuring that all nodes in the network can accurately locate and access them. For each level of sub-elements and their corresponding hierarchical units in the network, highly flexible deployment characteristics are also present: sub-elements at the same level can be centrally stored on one node, or they can be distributed across multiple different nodes according to their unit numbers. For example, some units of the second level can be deployed on one node, and other units of the second level can be deployed on other nodes. They only need to be distinguished and managed through a unified hierarchical identifier and unit number. The third, fourth, and higher levels can also be flexibly allocated and deployed in this way. Since the address and identifier of each level of storage and unit are uniquely determined in the entire network, the indexing and association relationships between elements are completely unaffected by physical location and regional distance. Therefore, the association and sharing mechanism of this invention can perfectly adapt to distributed, large-scale, and cross-regional data processing scenarios such as the Internet, cloud computing, and blockchain, realizing efficient association computing and data sharing without spatial limitations.

[0117] Furthermore, the technical solution described in this invention can also serve as an encryption tool for data or program encryption protection. In application, basic core elements can be customized based on the characteristics of the data or program to be protected. Corresponding hierarchical relationships and mutual indexing rules between elements can be formulated based on the structural characteristics and application scenarios of the data or program. A unique memory network structure is constructed through this method and stored as an independent file in a specific format. Since this file only stores elements and index relationships, rather than the plaintext structure of the original data or program, unauthorized users cannot directly identify and parse its content. To recover and read the original data or program, a dedicated loading program is required to restore the basic core elements, sub-elements at all levels, and the index relationships between elements, thereby completely reproducing the original data or program content. This method achieves high-strength data and program encryption protection, offering higher security and resistance to cracking compared to traditional encryption methods.

[0118] Furthermore, for all the application scenarios described above, encryption and verification measures can be adopted at every stage of system use to further enhance data confidentiality and overall operational security. For example, the index information of each element unit can be encrypted, and a dedicated verification mechanism can be set for the unit number; at the same time, encryption verification and access control can be set for the access process of the basic core element layer and each level of sub-element layer, and only authorized access objects can call and read the corresponding level or unit. Especially in distributed data processing systems such as the Internet, cloud computing, and blockchain, independent access permissions and encryption verification rules can be set for each distributed node; even if a node is illegally cracked, since that node only stores index information, it needs to point to and access the element content of other nodes to obtain the complete data. If access permissions to other nodes are not obtained, cracking only a single node cannot obtain effective information, thereby significantly improving the confidentiality, anti-attack capability, and data security strength of the system as a whole. Attached Figure Description

[0120] This invention comprises seven accompanying drawings, all of which are schematic diagrams, used to illustrate the relevant technical background principles, core architecture, and specific implementation process of the invention. The detailed descriptions of each drawing are as follows:

[0121] 1. Figure 1 This diagram illustrates a typical computer data representation and storage method, and is related to the background technology. The diagram shows three independent input ports X1, X2, and X3, each connected to a data converter. Each data converter is connected to a dedicated binary data storage device 1, 2, and 3. The data 1 to data L, data 1 to data M, and data 1 to data N generated after conversion from the three inputs are stored independently in their respective binary data storage devices, intuitively demonstrating the modular and isolated storage characteristics of traditional computer information processing.

[0122] 2. Figure 2 This is a schematic diagram of the basic structure of a perceptron, which is related to the background technology. The diagram shows three input signals X1, X2, and X3, each of which is connected to weights W1, W2, and W3 respectively. All the connections between the inputs and weights are aggregated to a weighted summation unit. The output of the weighted summation unit is connected to an activation function unit f(·). The activation function unit finally outputs the result y. The diagram also shows the core operation expression of the perceptron, y=f(w1x1+w2x2+w3x3+b), which fully demonstrates the basic operation logic of a traditional artificial neural network.

[0123] 3. Figure 3This diagram illustrates the connection between locating neurons and quantifying neurons, and the formation of sensation. It also represents the brain neural network theory upon which the invention is based. The diagram shows three locating neurons X1, X2, and X3, each connected to several quantizing neurons (e.g., X1 corresponds to a1, a2, etc., X3 corresponds to b1 to b...). N (etc.), the output of all quantized neurons is connected to the sensory formation area, clearly demonstrating the hierarchical connection mechanism by which the brain distinguishes the location and type of stimulus through spacer neurons, maps the intensity of stimulus through quantized neurons, and finally generates the corresponding sensation in the sensory formation area.

[0124] 4. Figure 4 This is a schematic diagram of the brain's memory network structure, illustrating the core theory of the brain-like architecture of this invention. The diagram clearly defines a three-layer network structure: Layer 0 is the basic core element layer, with basic core element neurons P0, P1, P2, P3, P4, P5, and P7; Layer 1 is the sub-element layer, with sub-element neurons F... 11 F 12 F 13 F 14 F 15 Each neuron establishes a connection with the corresponding basic core element neuron in layer 0; layer 2 is a higher-level sub-element layer, denoted as Fsub-element neuron. 21 F 22 F 23 Each neuron establishes a connection with the corresponding sub-element neurons in the first layer, fully demonstrating the principle of memory network construction that combines basic core elements to form sub-elements at each level.

[0125] 5. Figure 5 This is a schematic diagram of the brain's memory network recall process, illustrating the brain's memory recall mechanism upon which this invention is based; the diagram retains... Figure 4 The three-layer network structure (layer 0, basic core elements such as P0 and P1 neurons, layer 1, F...) 11 F 12 Sub-element neurons, layer 2 F 21 F 22 F 23 The sub-element neurons, with the stimulation of the basic core element neurons in layer 0 as the trigger point, demonstrate the transmission path of nerve impulses between neurons in each layer, the chain excitation response, and the connection and activation relationship between neurons, clearly reflecting the brain's memory triggering and signal transmission mechanism.

[0126] 6. Figure 6The diagram illustrates the hierarchical storage structure of the color image elements in Example 1, and serves as a schematic representation of the implementation of the technical solution of this invention in an image processing scenario. The diagram shows that the invention decomposes the color image into basic core elements, and then uses these basic core elements as units to construct first-level sub-elements and second-level sub-elements, ultimately combining them into a hierarchical decomposition and storage architecture for a complete image. This is a specific application of the hierarchical element processing solution of this invention in image data.

[0127] 7. Figure 7 This diagram illustrates the layered list structure of image storage in Example 1, representing the core implementation details of the present invention. The diagram shows four storage lists: W0, W1, W2, and W3. W0 is the basic core element layer, with each unit labeled with its basic core element and output index (including the connection list name, unit number, and weight). W1 and W2 are sub-element layers, with each unit labeled with its input index (including the connection list name and unit number) and output index. W3 is the highest level, labeled with its input index and image name (pic). Precise connections are established between list units through indexes, fully presenting the unit composition, input / output index relationship, weight configuration, and inter-level association logic of the hierarchical storage of the present invention. Detailed Implementation

[0129] The present invention will be further described in detail below with reference to specific accompanying drawings and embodiments. The embodiments focus on how to determine the basic core elements based on the processing object, how to divide the sub-elements at each level according to a preset hierarchical structure, and what structures and methods are used to achieve the association and sharing between the basic elements and the sub-elements at each level. The embodiments highlight the simulation of neuron plasticity in artificial systems such as computers, that is, establishing connections between elements in real time according to actual needs, and demonstrating the efficiency and dynamic adjustment mechanism of synaptic connections between neurons.

[0130] The scope of protection of this invention is not limited to the following embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the scope of the concept and principles of this invention are included within the scope of protection of this invention.

[0131] I. Example 1

[0132] This embodiment uses the storage, extraction, and analysis of a single color image as an example to illustrate the specific implementation method of the present invention.

[0133] After storing images using the hierarchical element structure described in this invention, complete extraction and local analysis of image content can be achieved: including identifying shared elements in the image and reflecting the relationships between elements; it can analyze which areas and which graphic elements in the image contain common elements, thereby achieving structured parsing and processing of image data.

[0134] 1. Image structure analysis and element analysis

[0135] For ease of description, this embodiment uses the appendix. Figure 6 (1) The color image shown is composed of 8×4 small squares of a single color. Whether it is a biological vision system or an artificial vision / video system, the processing of color images is usually decomposed into grayscale signals of three basic tones. For the sake of simplicity, this embodiment directly uses each small colored square as the basic core element. Since each small color block is a single color and is a regular square, and the overall image is also a regular rectangle, the entire image information can be completely expressed by using a single small color block as the basic core element, so there is no need for further subdivision.

[0136] Appendix Figure 6 (2) Schematic diagram of element hierarchy. Each small colored block in the diagram represents a basic core element; every 4 basic core elements form a first-level sub-element, and there are a total of 8 first-level sub-elements, which are identified by the letters a to h respectively; every 4 first-level sub-elements form a second-level sub-element, of which first-level sub-elements a, b, c, and d constitute second-level sub-element A, and first-level sub-elements e, f, g, and h constitute second-level sub-element B; finally, the second-level sub-elements A and B together form a complete picture, which is represented by pic.

[0137] Appendix Figure 6 (2) The small square with 4 numbers in the upper right corner points to the first-level sub-element f through the arrow, which is used to indicate the numbering order of the 4 basic core elements inside each first-level sub-element: among the 4 basic core elements, the two at the top are numbered 1 and 2 from left to right, and the two at the bottom are numbered 3 and 4 from left to right.

[0138] The labels and symbols of each level of sub-elements in the diagram are used to determine the relative position of the elements in space, and must be clearly expressed during image storage and retrieval.

[0139] 2. Data storage structure

[0140] Appendix Figure 6(3) is a schematic diagram of the data storage structure. The data storage structure corresponds to the four-level hierarchical division of the image, with four lists, W0, W1, W2, and W3, set up as storage. Among them, W0 is used to store basic core elements, with each unit storing a small color block. The number next to the unit is the unit address, and this address rule applies to all lists; W1 and W2 are used to store first-level sub-elements and second-level sub-elements, respectively. The letters in each unit of W1 represent the corresponding first-level sub-elements; W3 is used to store the entire image. The structure is organized in a way similar to neural network connections. Multiple units in W0 converge to a unit in W1 through straight lines, indicating that the first-level sub-element is composed of the basic core elements in the corresponding W0 unit. The numbers on the connecting lines represent the position of the basic core elements in the first-level sub-elements. If cell 0 of W1 is labeled with a and g, it indicates that the first-level sub-element a and the first-level sub-element g have the same structure, and there is no need to create a new cell for g. Similarly, if cell 3 of W1 is labeled with d and e, it indicates that the first-level sub-element d and the first-level sub-e have the same structure, and no new cell will be generated again. The same sub-element is stored using cell reuse. The composition relationship from W1 to W2 and from W2 to W3 follows the hierarchical combination rule from W0 to W1.

[0141] The above content is for illustrative purposes only, illustrating the overall data structure formed after the hierarchical division of the image, and is not a true representation of the actual operating state of the system. For example, the lines connecting the lists in the diagram are not actual physical connections or signal lines, but rather logical pointers and hierarchical calls between data achieved through index association. This section is only used to illustrate the overall organization of the data structure, thereby completing the explanation of the overall architecture. The next section will describe the specific data storage process in detail.

[0142] 3. Image storage

[0143] The image storage process is actually about creating an attachment using computer languages. Figure 6 (3) illustrates the relationships between the core elements and their sub-elements, and between the sub-elements themselves. Specifically, this means that the index indicates which elements are connected to which elements, and the order in which they are connected. Therefore, this paper will not go into detail about the specific process of establishing this relationship, but will directly give the results and explain the key points and essentials in relation to the results.

[0144] The data structure construction method and data processing and storage process adopted in this invention completely simulate the human brain's thinking process of analyzing, memorizing and learning objective things. Therefore, in the process of constructing the basic core elements and each layer of sub-elements, the habits of human brain cognition can be followed, allowing different analytical perspectives, different construction sequences and different processing methods for the same thing, rather than being limited to a single fixed process.

[0145] In this embodiment, the first step is to construct the basic core elements related to the image, namely the smallest units such as the basic color blocks that make up the image. Two flexible implementation methods can be used in the specific construction process:

[0146] The first approach is to process the entire image as a whole, extract all the basic core elements from the image, store them in a basic element list beforehand, and then build the subsequent hierarchical structure based on this list.

[0147] The second approach starts with a hierarchical structure. First, establish the first-level sub-elements in the first-level hierarchy W1. For example, first establish the first-level sub-element 'a', and then analyze the basic core elements contained in this sub-element, such as the smallest unit like red. During the construction process, first query the existing list of basic core elements. If the corresponding element exists, it is directly referenced without being stored repeatedly. If it does not exist, the element is added to the list of basic core elements and then referenced. In this way, the construction of the entire structure is gradually completed.

[0148] The two methods described above are similar to the process by which humans learn Chinese characters: the first method is equivalent to first memorizing and mastering the basic unit of a single Chinese character, and then combining and applying it; the second method is equivalent to gradually recognizing and accumulating Chinese characters through learning words and sentences in actual use. Both methods can achieve the hierarchical and element-based storage goals of this invention. This invention does not limit the construction process to a uniform and fixed method, but rather follows the cognitive habits of the human brain, adopting a flexible and adaptive method of element extraction and index construction.

[0149] In a hierarchical structure, the connection between any sub-element at a lower level and its sub-element at the next lower level can be viewed as the lower-level sub-element as input, representing its source of composition; the connection with a higher level can be viewed as the sub-element's output, representing its contribution to the structure above. Therefore, each sub-element contains both input and output components to reflect its relationship with the levels above and below, corresponding to input and output indices respectively.

[0150] Appendix Figure 7 This results in the final list structure. List W0 stores the basic core elements, and each unit consists of two items: the basic core element and the output index. The output index indicates which higher-level sub-elements constitute the basic core element, and each output index contains three elements: the linked list name, the specific unit number, and the weight.

[0151] For example, cell 0 of W0 has an output index (W1,0,2), indicating that this feature is a component of the sub-feature represented by cell 0 in W1. The weights here are analogous to the efficiency of synaptic connections in brain neurons; the weight automatically increases by one each time the feature is used, so a weight of 2 indicates that the feature has been used twice. (See attached comparison) Figure 6(3) It can be seen that cell 0 of W1 stores two identical first-level sub-elements a and g. This basic core element is used twice, so its weight is 2. Similarly, cell 1 of W0 corresponds to three output indices, indicating that the basic core element stored in this cell is a component of three sub-elements, and is related to the attached... Figure 6 (3) There are 3 output lines in this unit that correspond completely.

[0152] Each cell in W1 contains two items: an input index and an output index, separated by square brackets. Since the input and the output of the previous layer belong to the same connection line, each layer expresses its weight in the output index; the input index only contains the list name and cell number it connects to. Taking cell 0 as an example, it has three input indices, pointing to cells 0, 1, 2, and 3 of W0, corresponding to the four connections derived from these four cells.

[0153] It is particularly important to emphasize that the input indices have a fixed order. This order represents the spatial relationship between the previous layer's features and the current sub-features. An incorrect order will result in positional deviations in the final generated image. For example, the input index for cell 2 in W1 must be [(W0,1),(W0,4),(W0,3),(W0,6)], and cannot be arbitrarily written in cell number order as [(W0,1),(W0,3),(W0,4),(W0,6)]; using the latter will change the spatial position of the color blocks within cells 3 and 4 within the sub-features, causing image errors.

[0154] The output index of W1 is in triplet form, including the list name of the output connection, the cell number, and the weight. This cell has two output indices, both pointing to cell 0 of W2, with a weight of 1, indicating that it is referenced only once.

[0155] The structure of W2 is the same as that of W1, and will not be described again here. For the image described in this embodiment, W3 is the highest level and is not a component of other higher levels, therefore no output index is set. Unit 0 of W3 includes the input index and the image name pic.

[0156] 4. Image extraction and analysis

[0157] See attached document Figure 6 and attached Figure 7All sub-elements at each level, and all elements within the entire image, are arranged and numbered according to a unified set of rules. Each first-level and second-level sub-elements contains four units. Specifically, each of the first-level sub-elements a through h contains four basic color blocks, numbered as follows: the top two are 1 and 2, and the bottom two are 3 and 4 from left to right. Second-level sub-elements include A and B. A contains four first-level sub-elements a, b, c, and d, numbered in the same way: top 1, 2; bottom 3, 4. B contains four first-level sub-elements e, f, g, and h, also numbered 1, 2, 3, and 4. The complete image is composed of second-level sub-elements A and B, arranged from left to right, with A on the left and B on the right.

[0158] When storing data, the relevant indexes are recorded in the above positional order; when extracting the complete image, the same arrangement order and positional relationship must be strictly followed to restore the image, ensuring that each level of sub-elements is restored to its original arrangement position and the image structure is completely reconstructed.

[0159] The process of extracting the complete image (corresponding to the brain's recall process): Based on the image name "pic", the input index (W3, 0) can be retrieved; based on the input index within this cell, the input indices representing secondary sub-elements A and B (W2, 0) and (W2, 1) can be found; then, based on the input indices, the primary sub-elements and their input indices are found level by level, and finally, the complete image is restored and reconstructed according to the basic core elements contained in each primary sub-element.

[0160] Combined with appendix Figure 6 and appendix Figure 7 It can also be seen that, starting from any node, one can trace downwards to the constituent elements corresponding to that node, and upwards to determine the superior constituent elements to which that node belongs. Based on this, images can be analyzed using keywords or other features. For example, the output index of (W0,1) shows that the basic color block stored in this unit is a component of three first-level sub-elements, meaning that all three first-level sub-elements pointed to by this unit contain this basic core element. The input index of (W1,0) shows that the first-level sub-elements corresponding to this unit are composed of the basic color blocks stored in units 0-3 of W0. Its output index shows that this first-level sub-element is also a constituent unit of second-level sub-elements A and B. Therefore, when multiple images are stored in this system, it is possible to analyze which images a node shares with.

[0161] II. Example 2

[0162] This embodiment discloses a language model based on a hierarchical structure. Its hierarchical design is consistent with that of Embodiment 1. Both adopt a hierarchical list structure with W0 as the core element layer. This embodiment sets up a total of eight levels from W0 to W7. The organization, storage and retrieval of language elements are realized through unified hierarchical association rules.

[0163] The basic functions of the language model are as follows: by reading a large amount of text data, the text is divided into different levels of elements such as single characters, words, sentences, paragraphs, and chapters according to the granularity, and the text is stored in a structured manner according to the W0 to W7 hierarchical structure to achieve the complete expression and storage of the entire document; the elements of each level are associated and shared through indexes, and the elements of the same level can be reused by multiple upper-level elements, thereby achieving efficient organization and expression of language information.

[0164] 1. Data structure of each list

[0165] (1) W0 (basic single-word layer)

[0166] As the lowest-level core element layer of the model, the storage unit is a single Chinese character (including punctuation). Each unit's content is configured in two ways: If the character in the unit is the first character of a word, phrase, or sentence, an output index is set within the unit. The index includes the name of the higher-level list it points to, the target unit number, and the association weight. The weight represents the frequency of the character's association with subsequent words or sentences. The initial weight is set to 1 when the association is first established, and automatically increments by 1 each time it is reused. If the character is not the first character of a word, phrase, or sentence, and is only used as a connecting character, no output index is set for this unit. The association between this character and other language elements is reflected and associated by the input index of its parent element. This design simplifies index storage redundancy, allows for more focused core initial associations, significantly improves the efficiency of tracing and retrieving from the starting character to higher-level words and sentences, and facilitates quick location of language element association chains starting with a specific character.

[0167] (2) W1 (word level)

[0168] The storage unit is an independent word, without basic core element storage. It only achieves association with upper and lower levels through dual indexes, including two core components: an input index and an output index. The input index is a binary structure, containing only the list name pointing to the lower-level W0 layer and the corresponding unit number, without association weight. The association frequency between this level and the lower-level word is reflected by the weight of the W0 layer output index. The configuration rules of the output index are consistent with the W0 layer. If the word is the first word of a phrase or sentence, an output index is set in the unit. The index contains the name of the higher-level list it points to, the target unit number, and the association weight. The weight represents the frequency of the word's association with subsequent higher-level elements. The initial value is set to 1 when the association is first established, and it automatically increments by 1 each time it is reused. If the word is not the first word and is only used as a connector, no output index is set in this unit, and its association with higher-level elements is reflected by the input index of the corresponding upper-level element.

[0169] (3) W2 (Short Sentence Level)

[0170] The storage unit is an independent phrase, without basic core element storage. The unit only has a dual-index structure with input and output indices, and the index configuration rules are consistent with the W1 layer. The input index is a binary structure, containing the list name pointing to the lower-level W0 and W1 layers and the corresponding unit number. It adapts to the mixed composition logic of phrases containing single characters, punctuation, and words, and has no associated weights. The frequency of association with lower-level single characters and words is reflected by the weight of the corresponding lower-level output index. If the phrase is the opening phrase of a sentence, an output index is set within the unit. The index contains the name of the higher-level list it points to, the target unit number, and the association weight. The weight represents the frequency of the phrase's association with subsequent sentence-type elements. The initial value is set to 1 when the association is first established, and automatically increments by 1 each time it is reused. If the phrase is not the opening phrase and is only used as a connecting element within a sentence, no output index is set for this unit, and its association with higher-level elements is reflected by the input index of the corresponding upper-level element.

[0171] (4) W3 (Complete Sentence Level)

[0172] The storage unit is an independent complete sentence with complete sentence-ending punctuation (。?!;), without basic core element storage. The unit only has a dual-index structure with input index and output index, and the index configuration rules are consistent with the lower layer.

[0173] (5) W4 (paragraph level)

[0174] The storage unit is an independent paragraph after text segmentation, without basic core element storage. The unit only has a dual index structure of input index and output index, and the index configuration rules are consistent with the lower layer.

[0175] (6) W5 (segment)

[0176] The storage unit is a section in the document with an independent identifier (such as section XX, subsection, etc.), without basic core element storage. The unit only has a dual index structure of input index and output index, and the index configuration rules are consistent with the lower layer.

[0177] (7) W6 (Chapter Level)

[0178] This is the highest level of the model. The storage unit is a chapter in the document with an independent identifier (such as Chapter X, Chapter, etc.). There is no storage of basic core elements. The unit has a dual index structure with input index and output index. The index configuration rules are consistent with those of the lower level.

[0179] (8) W7 (Complete Document Layer)

[0180] This layer is the highest level of the model, with no higher-level elements to connect to, therefore no output index is set. An input index and document name are provided. The input index points to W6, summarizing all chapters contained in the document, directly achieving structured encapsulation of the entire document.

[0181] While this model pre-defines an eight-level structural framework from W0 to W7, it does not require all document content to extend to the highest level. Instead, it flexibly adapts and builds content as needed based on the document's length and structural characteristics. For short articles and scattered notes without standardized chapter or section divisions, containing only basic elements such as words, sentences, and paragraphs, their content will naturally terminate after being structured and stored at the corresponding W3 (complete sentence level), W4 (paragraph level), and W5 (section level), without needing to build meaningless section or chapter levels upwards. Only documents with complete chapter and section structures (such as books, standardized papers, and long manuscripts) will be adapted and built up to the W6 (chapter level) and W7 (complete document level). This design ensures the complete hierarchical expression of long structured documents while avoiding hierarchical redundancy in short documents, balancing storage integrity and lightweight design.

[0182] 2. Document splitting rules

[0183] This model uses the jieba word segmentation tool to achieve hierarchical text segmentation. Combined with manually set granularity rules, it segments the original document from fine to coarse into single characters, words, phrases, complete sentences, paragraphs, sections, chapters, and complete documents, adapting to W0 to W7 hierarchical structures. The core segmentation rules are as follows:

[0184] (1) Basic element segmentation: First, the text is broken down into individual Chinese characters / punctuation marks (corresponding to W0 layer), and then 2-5 characters of independent words are extracted by word segmentation tool (corresponding to W1 layer). Single characters that do not form words are retained as basic units of W0 layer;

[0185] (2) Phrase segmentation: Based on the single characters of W0 layer and the words of W1 layer, the phrases are spliced ​​together to form semantically complete independent phrases (corresponding to W2 layer), which are adapted to the mixed composition logic of words, single characters and punctuation in the text;

[0186] (3) Sentence segmentation: The sentence is segmented into clauses by punctuation marks (,;), and then the clauses are integrated and segmented into independent complete sentences with complete punctuation marks (.?!) by the end of the sentence;

[0187] (4) Chapter segmentation: Divide the text into independent paragraphs by line breaks (corresponding to W4 level); then divide it into sections (corresponding to W5 level) and chapters (corresponding to W6 level) by chapter / section identifiers (such as Chapter X, Section XX) in sequence; finally integrate all chapters to form a complete document (corresponding to W7 level).

[0188] The segmentation process follows the principle of starting with fine details and moving to coarse details, and segments according to semantic integrity and language usage habits. For text fragments that do not have complete semantics but are often used as fixed units in actual use and can be combined with other units, they are also extracted as sub-elements to form corresponding hierarchical units. Each level performs a deduplication operation before storing elements to ensure that the same elements are not stored repeatedly.

[0189] 3. Index creation

[0190] This model supports two indexing methods, which can be flexibly selected according to the actual application scenario. The first method is to first completely segment the document and then build the index uniformly. First, according to the aforementioned segmentation rules, the entire document is segmented level by level from single characters, words, phrases to sentences and paragraphs. The basic core elements and sub-elements of each level obtained from the segmentation are stored in the corresponding level units. After the full text segmentation and storage are completed, the index association is uniformly built according to the original structure of the document and the order and combination relationship between elements. This method is simple to implement, and the segmentation process and the indexing process are independent of each other and do not require additional marking. However, when processing large-scale documents, the text content needs to be traversed twice, which involves some redundant processing overhead.

[0191] The second approach, which is the preferred method adopted in this model, involves segmenting and indexing simultaneously. While segmenting the document paragraph by paragraph and sentence by sentence, basic and sub-elements at each level are extracted in real time and stored in their corresponding storage units. Based on the segmentation order, text connections, and hierarchical relationships, the index is built and associated synchronously. The index building process is integrated with the segmentation and storage processes, eliminating the need for a second traversal of the entire text, resulting in higher processing efficiency, especially suitable for real-time processing of large-scale documents.

[0192] For example, when the word "we" appears in the document, first check for duplicates of the single characters "wo" and "men" in W0. If they do not exist, allocate storage units for them; if they already exist, directly reuse them. Subsequently, check for duplicates of the word "we" in the word layer W1. If it does not exist, allocate a unit for it and store it; if it already exists, directly reuse it. Subsequently, set an output index for "wo" in W0, pointing to the unit where "we" is located in W1. When establishing the association for the first time, the weight is reset to 1. As the last character of the word, "men" does not set an output index. At the same time, establish input indexes within the unit of "we" in W1, respectively pointing to the units where "wo" and "men" are located in W0. The order of the input indexes strictly corresponds to the language writing order of the word and cannot be reversed. The input index only records the composition relationship and does not set a weight. When "we" appears again later, "wo", "men", and "we" already exist and there is no need to repeat storage. Only add 1 to the weight of the output index corresponding to "wo" and the weight of the subsequent relationship corresponding to "we". The storage, duplicate checking, index establishment, and weight update of the remaining hierarchical elements are all performed according to the above rules.

[0193] To further improve the efficiency of element duplicate checking and retrieval, this model uses hash values as the index markers for each hierarchical element. By generating a unique hash value for each element, when performing duplicate checking, there is no need to traverse and match all units within the layer one by one. Just quickly compare the hash values to determine whether the element already exists, significantly reducing the retrieval overhead and time complexity. In the subsequent element retrieval and index call process, it also achieves fast positioning and access based on the hash value, thus significantly improving the processing speed and operation efficiency of the entire system.

[0194] 4. Functional Features and Technical Effects of this Model

[0195] This model can implement functions such as automatic text segmentation, element storage, duplicate removal, index construction, and weight statistics. It has the characteristics of clear structure, rigorous logic, high processing efficiency, and low resource occupancy. By adopting hierarchical storage and hash fast retrieval, it can effectively avoid duplicate storage and redundant traversal, while ensuring the integrity of the language order and semantic accuracy, significantly improving the text processing and retrieval efficiency, and having strong practicality and scalability.

[0196] This model is a simple language model, specifically a text processing model. Its design goal is not to build a complete large language model, nor to compare it with existing large language models, but rather to simulate and implement the hierarchical element association and sharing mechanism proposed in this model within the existing data architecture and hardware environment. This is the core design goal of this model. The above structure and process demonstrate that the language text processing model proposed in this solution is simple in structure, logically clear, and practically feasible. The model was tested on a simple laptop with 8GB of memory, programmed in a Python environment; it loaded documents of approximately 900,000 words, identifying about 4,500 individual characters and approximately 35,000 words with fewer than 5 characters. Loading a 300,000-word book took approximately 30 seconds. The model can extract complete documents based on document names, search for sentences starting with or containing keywords, and perform strict or fuzzy matching of documents containing keywords. Therefore, with slight refinement based on the word base, it can be developed into a document management system to retrieve and extract large amounts of stored text according to individual needs. For example, if the document is data-driven, such as a massive protein molecule library used for drug screening, the parameters of protein molecules and a large number of candidate small molecules can be normalized and directly compared as basic core elements. In this case, the association and matching are established at the same time as the document is stored according to this model, and there is no need to compare them one by one in vector form.

[0197] In the current implementation of this model, text reading is the core data loading method, supporting the parsing and import of Word and TXT text formats. Functionality can be expanded upon this foundation, adding human-computer interaction interfaces such as typing input and voice input to simulate the process of children learning language, gradually guiding the computer to accumulate language materials according to the language structure of characters, words, and sentences.

[0198] The element relationships established in this model can be compared to the language regions in the human brain's neural network. A large number of language elements are interconnected, with closely related and frequently used connections corresponding to higher connection efficiency and easier retrieval. The weight parameters in this model quantify this feature. When expressing a certain thing, it can automatically associate various words related to that thing, generating multiple candidate words or sentences after a single word. The content stored in each level of the model functions similarly to the tokens obtained by existing large language models through training on massive amounts of text. However, unlike large language models where the associations between tokens depend on vector distance and probability statistics rather than directly on language semantics and structure, this model uses fixed language collocation structures and weight parameters to jointly reflect semantic relationships and usage patterns, more closely resembling the original structure and actual usage habits of natural language.

[0199] This model does not provide semantic descriptions of words and sentences, does not have a dedicated word explanation database, and does not employ measures to enable computers to understand the meaning of words and sentences. It can only achieve simple text retrieval and element extraction functions, which can be directly implemented based on the text structure. To enable the model to have stronger language output capabilities, word and sentence explanations need to be added on top of this, and existing machine learning technologies in related fields such as Chinese need to be introduced. The core of the data processing in this model is not to find the relationships between elements through massive computations, but to directly utilize the existing relationships between things and store them in a structured way, which is equivalent to memorizing a large amount of text content in the system. If only existing connection weights are used for hard matching, the generated language will not be flexible and standardized enough. Therefore, to further achieve machine intelligence based on this model, the focus is not on increasing the amount of computation, but on transplanting human thinking patterns into computers, describing them through computer language, and achieving the goal of generating new knowledge from existing knowledge.

[0200] III. Summary and Conclusion

[0201] As can be seen from the above embodiments, the information processing system constructed according to the present invention, in the entire process of processing various target objects, from the decomposition of basic core elements to the construction of sub-elements at each level and the establishment of their relationships, reflects the thinking mode of the human brain, and can be regarded as a complete transplantation of the human brain's cognitive process into an artificial system.

[0202] Therefore, the core technology of various simple or complex systems constructed using this invention is no longer about competing on hardware storage capacity and computing speed, but rather on how to accurately describe and implement human thought patterns using computer language. The more perfect this implementation, the closer the artificial intelligence system is to human brain intelligence. Combined with the high-speed data processing capabilities of the artificial system itself, it is expected to surpass the processing efficiency and capabilities of the human brain in many application scenarios. For this reason, various simple or complex systems constructed using this invention will be influenced by the thought patterns and application scenarios of the system builders and actual users, thus resembling human intelligence and possessing typical personalized characteristics.

[0203] The data processing architecture of this invention is entirely based on the physical connections of neurons in the brain, rather than relying on traditional data-driven methods. In the image storage and recall process described in Example 1, this model employs a hierarchical, mutually indexed data structure. The relationships between elements can be derived directly through logical deduction. The entire process is driven directly by physical connections, without relying on massive amounts of data or large-scale computation. In contrast, existing data-driven models such as neural networks, even with simplified structures, require iterative training to obtain internal connections and weights, making it impossible to directly derive a stable and interpretable physical connection structure. Therefore, this model, in principle, achieves low computational cost, high interpretability, and strong structural stability, fundamentally different from existing data-driven methods.

[0204] In Example 2, the W0-W7 list is stored on the disk as a .pkl file. Each time a document is loaded, the existing data is retrieved from the disk, new knowledge is added on top of the existing data, and the entire data is written back to the disk after processing. This achieves continuous accumulation and consolidation of knowledge. Another file on the disk is used for the construction and parsing of the memory network. The memory network formed in this way has extremely high flexibility. Memory networks generated independently by different terminals or nodes can be easily merged: merging is based on basic core elements as units, identical elements are merged and shared, and different elements are directly appended. After the basic core elements of the merged network are incorporated into the main network, only the original index relationships need to be migrated synchronously to complete the overall merging. There is no need to reconstruct the network, and the merging process only adds new knowledge without destroying the original knowledge. This network also supports multi-node distributed collaborative creation. Each node adopts a unified list name and hierarchical structure, and unit numbers and storage space are allocated according to segments. The address of each layer of unit is globally unique, which can realize mutual referencing and sharing among multiple nodes to form a unified distributed memory network. It is significantly superior to the human brain's neural network in terms of knowledge fusion and distributed collaborative memory because human brain memory cannot be directly and conveniently merged and it is difficult to form similar collaborative memory.

[0205] The technical solution of this invention has extremely high implementation flexibility: it can be simulated on existing general-purpose computer architectures or implemented at the hardware level, so that each storage list is an independent computing unit that can independently process its own input and output indices, thereby achieving true parallel processing and possessing wide applicability and feasibility.

Claims

1. A data processing architecture based on human brain element sharing and its implementation method, characterized in that, The target processing object is decomposed into indivisible and indispensable basic core elements based on its inherent attributes. Each of these basic core elements corresponds to a specific neuron in the human brain's biological neural network. The target processing object is constructed by combining the basic core elements according to a preset hierarchical relationship. The target processing object covers data carriers, abstract concepts and logical relationships. The absence or change of any basic core element will result in an incomplete description of the target processing object or directly change the attributes of the target processing object. In the information processing system, the basic core elements carry real attribute information and reside in the underlying storage module in the form of a fixed storage address. They serve as a unified information source for the entire data processing process of the system, which can be directly or indirectly called by various functional links of the system. The system constructs a multi-level element structure based on the basic core elements. The elements at each level above the basic core elements are collectively referred to as sub-elements. The construction of sub-elements introduces an attention mechanism, which selects and combines suitable basic core elements according to the attribute characteristics of the actual processing needs. Each level of sub-element does not store specific attribute information of the basic core element, but only records the relationship between itself and the basic core element or other sub-element. The relationship includes the time relationship, spatial relationship, logical relationship and other definable or describable actual or abstract relationship between the basic core element that constitutes the sub-element and other sub-element. The target processing object contains several independent events. Each independent event does not independently enjoy its own basic core elements and sub-elements at all levels. All elements corresponding to the events exist in a distributed manner based on the multi-level element structure. The processing and expression of events are completed by calling the underlying unified storage of basic core elements and the relationship between elements.

2. The brain-based element-sharing data processing architecture and its implementation method according to claim 1, characterized in that, The global sharing and multi-level interconnection of the basic core elements and sub-elements constitute an inseparable collaborative mechanism. Cross-level and cross-event sharing of elements is achieved through the relationships between elements, and the various relationships between elements are constructed based on the globally shared unified basic core elements. Based on the underlying requirement of sharing elements across the entire system, each level of sub-element does not carry substantial attribute information, but only records the relationship between it and the basic core elements and other sub-elements. The information expression of each level of sub-element depends entirely on the retrieval of the underlying basic core elements. If the basic core element layer is shielded, no substantial and effective information can be obtained from the sub-elements themselves. The multi-level element structure under this collaborative mechanism follows a hierarchical construction rule from simple to complex. From the basic core element layer upwards, the complexity of each level of sub-elements increases progressively. The correlation between elements is reflected in full-level traceability. Any sub-element in each layer of the system can be traced downwards to locate the underlying sub-elements and basic core elements that constitute it, and can be traced upwards to clarify the upper-level sub-elements to which it belongs as a core component, thus realizing bidirectional traceability between upper and lower levels of elements. The basic core elements are unique identifiers that distinguish the attributes of things. All kinds of relationships between basic core elements, between basic core elements and sub-elements at all levels, and between sub-elements at all levels do not require distinguishing the attributes of things, and can realize cross-attribute relationship integration. Based on the core features of global sharing and bidirectional traceability, this architecture and implementation method construct a human brain-like associative memory and information activation mechanism through the association links between elements. When any basic core element or sub-element is retrieved and activated, all related basic core elements and sub-elements at all levels are reached and activated through the pre-constructed element association links, realizing the full-domain activation from single information to related information.

3. The brain-based element-sharing data processing architecture and its implementation method according to claim 1, characterized in that, This architecture and implementation method decomposes the various inherent attributes of objective things into corresponding basic core elements through the physical construction of basic core elements and the global interconnection and sharing mechanism of basic core elements and sub-elements at each level. Then, based on the natural relationship between things themselves, the sub-elements at each level are combined and constructed to achieve a complete representation of things. In the information import stage, the inherent attributes of elements and the natural relationships between elements are used to directly map and memorize the objective relationships that already exist between things and the attribute characteristics of different aspects of things. There is no need to mine and find the relationships between things through large-scale sample data training, algorithm fitting or logical deduction. This eliminates the hard dependence on large-scale sample training datasets, ultra-large-scale computing power and high power resources from the bottom up. The system adopts a distributed element sharing and hierarchical parallel computing architecture, which reduces the computing power load requirement of a single node and can run efficiently in low-resource hardware environments such as personal computers with conventional computing power.

4. The brain-based element-sharing data processing architecture and its implementation method according to claim 1, characterized in that, The hierarchical data processing architecture for element sharing can be directly applied to the underlying data construction and full-process processing of computers, artificial intelligence modeling, and various intelligent processing terminals and operating systems. Hardware-level machine language operation instructions are broken down into basic core elements according to their function and attribute characteristics. Each level of functional module is generated by sharing these basic core elements and forms system sub-elements. The intelligent processing terminal and the operating system use the shared basic core elements at the local level as the sole data source for each functional module. They achieve collaborative processing between modules through the relationships between the elements, and each functional module does not independently store the specific attribute information of the basic core elements. All numerical processing-related software, operating instructions, and original files are broken down into elements according to functional units and stored in a distributed manner in a hierarchical feature network. The storage medium only contains the hierarchical feature network and does not contain independent software, data, or instruction files of various forms. Only the feature directory-style reference index is retained. From a hardware structure perspective, hierarchical processing units, basic core elements, and system sub-elements can all be used as independent computing units. A single sub-element can run independently, and several sub-elements can also share and cooperate based on their relationships. The configuration of computing units can be flexibly set according to actual processing needs. The hierarchical parallel computing of the architecture is achieved through the parallel operation of independent computing units. Independent computing units can store their own associated hardware operation instructions and independently process data input and output indices.

5. The brain-based element-sharing data processing architecture and its implementation method according to claim 1, characterized in that, The aforementioned element sharing and multi-level interconnected processing mode can be adapted for deployment in wide-area distributed networks such as blockchain networks and cloud computing networks. Each network node connects to the underlying unified basic core element storage module level by level according to the set hierarchical element structure. The deployment location of this basic core element storage module is not restricted; it can be centrally stored at the network's central node or distributed across several designated nodes or all network nodes. All basic core elements and sub-elements are configured with a unique identity identifier that is unified across the entire network. This identity identifier is associated with the input and output indexes of this architecture and implementation method, so that nodes can index and call each other. Each network node only stores element fragments of the corresponding level and the relationships between elements, and does not directly store the complete access link of the underlying basic core element library. The element information stored therein needs to be spliced ​​together through multi-level index association to form valid data. Individually parsing or cracking any node can only obtain fragmented element information, and cannot obtain the complete basic core elements and the element processing logic of this architecture and implementation method.

6. The brain-based element-sharing data processing architecture and its implementation method according to claim 1, characterized in that, When the architecture and method are applied to the encryption and decryption of programs / data, the smallest functional unit with independent functional meaning that cannot be further divided is first extracted from the program / data as the basic core element. The basic core element is divided into attributes and types according to the original executable logic, execution order and structural relationship of the program / data. Each basic core element is configured with a unique system identifier containing element attributes and logical order. Based on the original structure and execution flow of the program / data, the basic core elements are combined according to their relationships to form sub-elements at each level. Each sub-element unit does not contain any substantive executable information, but only indicates which basic core elements / next-level sub-elements it is composed of in a specified logical order, as well as its hierarchical relationship as a component of the next-level sub-element. By using the unique system identifier of the basic core elements and the associated index of each level of sub-elements, the functions and execution logic of the original program / data are fully mapped.

7. The brain-based element-sharing data processing architecture and its implementation method according to claim 6, characterized in that, It also includes steps such as encrypted storage of elements, decoding by a dedicated parsing program, and full-process index format verification, specifically as follows. By using a pre-defined encrypted storage format for feature units, the hierarchical indexing relationships and association rules between features are embedded in a dedicated parsing program. The parsing logic of this dedicated parsing program uniquely corresponds to the hierarchical indexing relationships and association rules of the features, serving as the unique key for parsing and restoring the complete program / data. During parsing, the dedicated parsing program analyzes the index relationships contained in each sub-element according to preset rules, tracing back to the source to restore the complete program / data information. When fragmented element units are obtained individually or when the dedicated parsing program is not available, it is impossible to identify the element relationships and restore the complete program / data information. Standardized format rules and verification mechanisms are established for the mutual indexing relationships between element units at different levels in the element sharing architecture. The rules clarify the hierarchical association logic, field definitions, and verification identifiers of the element index. When an external program / data is to be included in the aforementioned element-sharing architecture, it is first decomposed into elements and its index format compliance is verified to determine whether it can generate a valid hierarchical index and establish a legitimate relationship between elements in accordance with the aforementioned standardized format rules. If it cannot be completed, its inclusion is automatically rejected and the external program / data is blocked and discarded. During the parsing phase, the retrieved feature units undergo secondary index format verification through the dedicated parsing program. Invalid data that fails the verification is directly filtered out to prevent it from participating in the integration and operation of the complete program / data.

Images