Neuron fragment and neuron fragment dynamic networking calculation method and system

Abstract

The application discloses a minimum function unit with field coupling capability, namely a neuron fragment, deploys the neuron fragment in an initial stable state information field to form a ground state field, after external input information coding is projected to the ground state field, each fragment perceives field density in parallel through field coupling, and the activated fragment projects field density preference to the outside to update the field density distribution of the information field according to the self field density preference and the activation threshold value, and the continuously activated fragment naturally forms a stable resonance cluster after field coupling evolution, and finally, the field density distribution is decoded and output to calculate the result. The application does not need large-scale pre-training, does not depend on central scheduling, and has high system stability; has field original growth period memory capability, and has high computing efficiency for repeated tasks; naturally supports cross-domain expansion, can be deployed in a distributed computing node, an edge end and a cloud end heterogeneous environment, realizes cross-domain resource cooperative networking and unified computing, and can realize cross-domain seamless deployment, and has strong versatility and expansibility.

Description

Technical Field

[0001] This invention relates to the fields of artificial intelligence, brain-like computing, and distributed collaborative computing, specifically to a method and system for dynamic networking of neuronal fragments and neuronal fragments driven by field coupling, applicable to various scenarios such as general artificial intelligence, edge computing, robot control, and brain-like intelligence. Background Technology

[0002] Existing artificial intelligence neural networks (such as Transformer and deep learning models) adopt a fixed topology and fixed connection design, and realize information processing through sequential computation layer by layer. Their core relies on weight parameters to store knowledge, which has the defects of rigid structure, weak adaptability, low reasoning efficiency, and difficulty in achieving true brain-like intelligence.

[0003] Specifically, existing neural networks have fixed structures for neurons and modules, which cannot be dynamically split, generated, or evolved; modules rely on pre-defined hard connections to transmit information, lacking global collaborative capabilities; knowledge is stored in fixed weights, making it difficult to achieve incremental learning and dynamic adaptation to complex environments; and the computation process requires waiting layer by layer, making it impossible to achieve full-field parallel collaboration. The improvement of computing power depends on the accumulation of parameters, resulting in diminishing marginal benefits, high power consumption, weak edge adaptability, and local failures that can easily lead to overall collapse. They cannot be dynamically expanded, making it difficult to simulate the dynamic reconstruction and low-energy consumption characteristics of the biological brain, and thus unable to achieve true self-awareness, self-growth, and self-networking. Summary of the Invention

[0004] In view of the aforementioned deficiencies in the prior art, this invention is based on the prior application No. 2026109020501, a method for constructing a self-evolving information field (hereinafter referred to as the parent patent). This patent discloses a method for constructing a self-evolving information field, an initialization method, and field evolution rules, providing a unified information interaction and computing carrier for this invention. Based on the aforementioned information field, this invention further proposes a method for generating neuronal fragments, activating them in the information field, and dynamically networking them to achieve adaptive computing for specific tasks. Driven by the field coupling mechanism, the method enables the free growth, autonomous connection, and dynamic networking of neuronal fragments. Through the collaborative computing of neuronal fragments driven by field coupling, the limitations of existing fixed-structure neural networks are overcome, thereby constructing a low-power, highly stable, scalable, and brain-like intelligent artificial intelligence underlying architecture.

[0005] The information field serves as the sole medium for information interaction and collaborative computing among all neuron fragments. It carries the transmission of external input signals, the evolution of internal reasoning processes, and knowledge storage throughout the entire process. It provides the necessary support for indirect collaboration between neuron fragments, which are not directly physically connected but only through field coupling. It is the core medium for information interaction and collaborative computing among neuron fragments and the fundamental premise for the "no fixed topology, self-organizing network" core design of this invention.

[0006] Based on the state differences of the information field, the information field in this invention is specifically divided into three types: initial equilibrium state, ground state field, and excitation field. The three types work in sequence and cooperate to complete the information transmission and evolution of the entire calculation process.

[0007] The initial equilibrium state is the initial equilibrium field space state generated when the system constructs the information field (see the parent patent for details). The ground state field is formed after the neuronal fragments are deployed to the information field in the initial equilibrium state. All neuronal fragments coexist within the information field and form implicit interactions with each other through field coupling. After the system naturally converges to reach an equilibrium state, a uniform, balanced, and silent inherent field space state is automatically formed. In the ground state field state, the information field has uniform field intensity and consistent phase across the entire domain. It does not carry any task structure or information features and only serves as the basic field environment for the deployment, perception, and subsequent work of neuronal fragments. It is important to clarify that this equilibrium field state is independent of the specific spatial distribution of the neuronal fragments—regardless of whether the neuronal fragments are diffusely distributed, regularly distributed, or clustered, as long as the system is free from external input signal interference, it will eventually converge to the ground state field with uniform field density and consistent phase, ensuring that each neuronal fragment is in an equivalent initial working environment, providing a stable foundation for subsequent activation judgment and collaborative computation.

[0008] When an external input signal enters the ground state field, the originally balanced ground state field is disturbed, breaking the original uniform equilibrium state and forming a field state with differences in field density and phase distribution, carrying external input information. This state is the initial distribution of the excitation field. The initial distribution of the excitation field is not fixed, but evolves continuously with the autonomous activation of neuronal fragments, the feedback of field signals from activated fragments, and the chain resonance process between fragments. As the core information medium of the entire information field, the excitation field undertakes the functions of information storage, transmission, and collaborative computation throughout the process. It is the core carrier for neuronal fragments to achieve dynamic networking and complete complex computational tasks, and its evolution process directly corresponds to the reasoning and computation process of the information field.

[0009] In this invention, field density is a core characteristic parameter of the information field points and the neuronal fragments deployed on these points. Specifically, it refers to a composite physical quantity existing at each point (i.e., the field point) in the field space, characterizing the strength and state of the field. Its essence is a concrete manifestation of information and the essential ontology of the information field in this invention. As the core carrier for information interaction between neuronal fragments and the information field, and between neuronal fragments themselves, field density allows neuronal fragments to perceive the field density within the information field in real time through their field-sensing interfaces, thereby acquiring global field information. Simultaneously, activated neuronal fragments project their own field density preferences outward through their field-excitation interfaces, feeding back into the field density distribution of the information field. This enables indirect information interaction between fragments without direct connections, aligning with the core mechanism of field coupling-driven mechanisms in this invention.

[0010] Specifically, field density is characterized by three core dimensions, each working together to represent the complete state of the field point and providing a foundation for information transmission and computation: 1. Field strength, indicating the magnitude of the field density and intuitively reflecting the strength of the field at that point; 2. Phase, representing the temporal characteristics and state shift of the field density, reflecting the dynamic changes in the field; 3. Mode, representing the spatial distribution shape of the field density, reflecting the spatial structural characteristics of the field. It should be noted that these three dimensions are only the core representation dimensions of field density, not the only representation method. Field density can also be further described using more vector parameters.

[0011] The neuron fragments in this invention are the smallest functional units with field coupling capabilities. Unlike the fixed nodes in traditional neural networks, these fragments do not rely on fixed topological connections and interact only through the information field. This is a core prerequisite for achieving connectionless, self-organizing networks. Each neuron fragment occupies a field point in the information field. Through its own field density preference, combined with the field density parameters of neighboring field points within its preset coupling radius, the neuron fragment completes field coupling calculation, activation judgment, and dynamic networking.

[0012] The neuron fragment has a field-sensing interface, a response kernel, and a field-activation interface. The field-sensing interface captures local field density information within its surrounding information field in real time, providing the basis for subsequent judgments and calculations. The response kernel, as the core functional module of the neuron fragment, has built-in inherent field density preference parameters and activation threshold parameters, and in preferred cases, also has built-in field coupling coefficient parameters. The response kernel is used to perform perception analysis, matching discrimination, activation judgment, field response calculation, and activation control on the field density information read by the field-sensing interface. It constitutes the minimum functional set of the neuron fragment and is also the core manifestation of the basic computational capabilities of the neuron fragment itself. Among them, the field response calculation is not a traditional software algorithm calculation, but a native responsive calculation based on field density matching and field resonance. The specific functions of the response kernel include: 1. Field density parsing function: acquiring local field density information in the information field through the field sensing interface, parsing parameters such as the amplitude, phase, and distribution pattern of the field density to form a feature representation that can be used for internal calculation; 2. Field density preference matching function: using its own inherent field density preference parameters as a benchmark, calculating the similarity between the parsed local field density features and its own field density preference to obtain the field density matching degree; 3. Activation intensity calculation function: calculating the activation intensity of the current neuron fragment according to preset rules based on the field density matching degree, field coupling coefficient, and local field density intensity. Without setting the field coupling coefficient, the activation intensity is directly calculated based on the field density matching degree and local field density intensity according to preset rules, which can also achieve effective calculation of activation intensity, but the calculation accuracy and field coupling interaction effect are slightly inferior. 4. Activation judgment function: compares the calculated activation intensity with its own inherent activation threshold to determine whether the neuron fragment has entered the activated state; 5. Field excitation control function: when the neuron fragment is activated, the response kernel drives the field excitation interface to project a field density signal consistent with its own field density preference, updating and strengthening the field density distribution of the information field; 6. State self-sustaining and adaptive function: during the network iteration process, the response kernel maintains the current active or silent state, and updates the activation judgment result in real time according to the continuous change of the global field density to achieve adaptive response.

[0013] The field excitation interface projects its own field density preference into the information field after the neuron fragment is activated, thereby updating the overall field density state of the information field and providing an interactive basis for the perception and activation of other neuron fragments. Neuron fragments have no fixed topological connections to each other, interacting only with the information field through field coupling. This breaks the limitations of fixed hierarchical structures and node dependencies in traditional neural networks, providing a structural foundation for subsequent dynamic networking and adaptive evolution.

[0014] As the smallest functional unit with field coupling capability, the inherent basic computing power of a neuron fragment is manifested in its minimal native functions such as micro-feature recognition, preference matching, threshold judgment, and field response reasoning. After multiple fragments are dynamically networked through field coupling, higher-order recognition, global reasoning, and complex solution capabilities emerge. This hierarchical design of computing power not only ensures the independence and basic functional integrity of individual neuron fragments but also realizes the higher-order computing capabilities after cluster networking, forming a complete system of "micro-basic computing power - macro-cooperative computing power". The micro-computing power and macro-computing power are automatically connected through field coupling without manual intervention.

[0015] Neuron fragments are the smallest functional units with field coupling capabilities. They cannot be further subdivided; further subdivision results in the loss of their independent field interaction functions, rendering them unable to respond, activate, or perform any effective computation independently. Maintaining this minimum functional granularity is crucial for the following beneficial technical effects: 1. Single and specialized function, resulting in higher decision efficiency: Each neuron fragment performs only the smallest-granularity field perception, activation decision, and field signal feedback functions. Its specialized function and simple logic enable rapid response and instantaneous activation, significantly improving the overall system response speed and decision efficiency. 2. No redundant structure, resulting in lower system energy consumption: The minimum functional granularity eliminates redundant computations and connections within the fragment, performing activation and field feedback operations only when necessary. This minimizes ineffective energy consumption, and combined with the information field energy circulation mechanism, achieves significant energy savings. 3. Highly flexible networking and stronger adaptability: The smaller the neuron fragment granularity, the richer the dynamic networking combinations. It can freely aggregate, diffuse, and recombine based on field density gradients and input information, adapting to different types and complexities of computational tasks, and exhibiting stronger generalization ability. 4. Enhanced anti-interference and anti-failure capabilities: Individual fragments are functionally independent and have a small granularity. Local fragment failures or disturbances only affect a very small range of field distributions, preventing overall network paralysis and significantly improving system stability and robustness. 5. Easy to generate, deploy, and expand: The smallest functional unit has a unified structure and standardized interfaces. It can be obtained by splitting from the parent neural network or directly generated through native construction, facilitating large-scale deployment, dynamic additions and removals, and system expansion. 6. Avoids complex fixed connections, simplifying the system architecture: The smallest functional granularity does not require complex internal connections and scheduling logic. Fragments interact indirectly only through field coupling, resulting in a simple and easy-to-implement system architecture that can achieve efficient computation without relying on high-density chip stacking.

[0016] Field coupling in this invention is the core mechanism for interaction between neuronal fragments and the information field, and also the foundation for indirect collaboration among neuronal fragments. Specifically, the core process of field coupling is as follows: neuronal fragments perceive field density information within the information field in real time through their field-sensing interfaces and can autonomously activate in response to changes in field density; once activated, the response kernel drives the field-excitation interface to project its own field density preference into the information field, thereby feeding back and updating the overall field density distribution of the information field. It is important to clarify that there are no direct connections or direct interactions between neuronal fragments in this invention. All collaborative linkages and information transmission among neuronal fragments are indirectly accomplished through the aforementioned field coupling process, which is one of the core features that distinguishes this invention from the fixed-node connections of traditional neural networks.

[0017] In this invention, field density preference refers to the preset target field density parameter in the neuronal fragment response kernel, and is also a set of field density features of the neuronal fragment's preferred response. Its core function is to characterize the neuronal fragment's tendency to respond to a specific field density. This field density preference is an inherent attribute of the neuronal fragment, determined by its preset function. Under the same field environment, neuronal fragments are more likely to be preferentially activated in field density regions close to their own field density preference. The degree of matching between the actual field density of the information field and the neuronal fragment's own field density preference directly determines the activation probability and activation intensity of the neuronal fragment. However, when the actual field density exceeds the upper limit of the neuronal fragment's field density preference, the activation intensity will not continue to rise, but will remain within the saturation threshold range. The core features of field density preference correspond to the three core dimensions of field density, specifically including the field strength (amplitude), phase (temporal offset), and local pattern (spatial distribution) of the neuronal fragment's preference, forming a precise correspondence with the characterization dimensions of field density. As the core criterion for determining whether a neuronal fragment is activated, the neuronal fragment only responds strongly to field density signals that match its own field density preference, while maintaining a low response or no response to mismatched field density signals, thus ensuring the specificity and accuracy of activation determination.

[0018] The activation threshold in this invention refers to the minimum field density threshold required for a neuronal fragment to transition from a quiescent state to an activated state. It is an inherent parameter built into the neuronal fragment's response kernel and, together with field density preference, constitutes the core basis for determining neuronal fragment activation. Its value is preset by the functional design of the neuronal fragment and does not dynamically change with the field evolution process. When a neuronal fragment senses the field density at a corresponding location in the information field through field coupling, and this field density is greater than or equal to the activation threshold, the activation condition is met. The neuronal fragment is then activated and projects its own field density preference signal outward through the field excitation interface, participating in dynamic networking and forming resonance clusters with other activated fragments to complete collaborative computation. When the local field density sensed by the neuronal fragment is lower than the activation threshold, the neuronal fragment remains quiescent, not participating in the activation, networking, or collaborative computation process, only maintaining basic field sensing functions, waiting for the field density to reach the activation threshold before initiating subsequent responses.

[0019] This invention also provides a method for dynamic networking calculation of neuronal fragments, comprising the following steps: A. Generating neuronal fragments; B. Deploy the neuronal fragments in an information field in an initial stable state, wherein the field point parameter structure of the information field is field density and mode gradient; construct a ground state field that is uniform across the entire domain and has stable field density; C. Encode the external input information and project it onto the ground state field to form an excitation field with a non-uniform field density distribution; D. Each neuron fragment senses the field density of its neighboring field points in parallel through field coupling, makes autonomous activation judgments, and achieves selective activation based on the judgment results; E. Activated neuronal fragments project their own field density preferences outward through field coupling to increase the field density of their neighborhood, thereby forming a field density gradient in the information field. The high field density region spreads, thereby chain-activating the previously inactive neuronal fragments in the vicinity. F. Field points in continuous regions between continuously activated neuronal fragments form high-density areas of field density. Neuronal fragments in these high-density areas are continuously activated and enhance the field density. After evolution, they naturally form stable resonance clusters, completing the first round of dynamic networking and collaborative computing. The calculation results are presented in the form of field density distribution. G. Decode the field density distribution of the information field and output the calculation results.

[0020] A resonance cluster refers to a group of field points with a stable field density distribution that forms when the field coupling strength of multiple neuronal fragments reaches a preset activation threshold during the evolution of the information field. The formation process of a resonance cluster includes the following stages: 1. When multiple neuronal fragments are activated due to the field density reaching their respective activation thresholds, each neuronal fragment projects its own field density preference signal into the information field through the field excitation interface. The projected signals superimpose and diffuse in the information field, forming local enhancement regions in areas where the field density matches.

[0021] 2. When the field density in the local enhancement region reaches the resonance condition (i.e., the field density deviation is lower than the preset threshold and the phase difference converges to the allowable range), multiple neuronal fragments in the region enter a steady-state coupling state and form a resonance cluster.

[0022] 3. The field points within the resonance cluster exhibit uniform field density distribution characteristics, including: information intensity tending to be consistent, phase difference converging to the allowable range, mode vectors converging to the same direction, and the cluster boundary being defined by the location of the maximum field density gradient.

[0023] 4. After the formation of a resonance cluster, the neuronal fragments within the cluster no longer rely on their individual preferred responses, but participate in subsequent calculations as a whole. The state (activation / silence) of the neuronal fragments within the cluster is determined by the field density state of the cluster and no longer responds independently to external signals. The stability and carrying capacity of the resonance cluster are determined by the number of neuronal fragments, the number of field points, the field density value, and the degree of convergence of the phase difference within it.

[0024] 5. Resonance clusters serve as intermediate carriers for intelligent computing, carrying a group of field points in the information field that have completed coupling and confirmation. The calculation results are obtained by decoding the field density distribution of the resonance clusters, i.e., extracting the field density parameters (such as mode vectors and information intensity) of the resonance cluster region and outputting the corresponding results. The lifespan of a resonance cluster is determined by the task computation time. After the task is completed, the resonance cluster automatically disbands, and the fragments within the cluster return to an independent activation waiting state, awaiting the next satisfaction of the field density conditions to reform a new resonance cluster.

[0025] Based on neuronal fragments, this method constructs a complete computational process of "generating fragments—deploying field space—input encoding—autonomous activation—dynamic networking—iterative convergence—decoding output," realizing collaborative computation driven by field coupling without fixed topology.

[0026] Furthermore, the method for generating neuronal fragments is as follows: 1. Select a pre-trained artificial neural network as the parent neural network; 2. The parent neural network is structurally decomposed according to the minimum functional unit standard to obtain basic neural units. Each basic neural unit inherits the field density preference parameter and activation threshold parameter corresponding to it in the parent neural network. 3. The basic neural units are de-fixed, removing their original hierarchical relationships, node dependencies, fixed transmission paths, and fixed weight connections; 4. Configure a unified field-sensing interface and field-activation interface for the basic neural units after defixed connections, and integrate their field density preference parameters and activation threshold parameters into the response kernel to form the neuronal fragments.

[0027] This step endows neurons with the ability to couple field fragments. Specifically, it enables them to perceive field density in the field space, make activation judgments based on their own field density preferences and activation parameters, and feed back field density signals to the field space and update the field density distribution after the neuronal fragments are activated.

[0028] The above method utilizes a pre-trained parent neural network to quickly generate neuron fragments. The generated neuron fragments directly inherit the corresponding parameters of their parent neural network, reducing the complexity and cost of redesigning neuron fragment parameters and improving generation efficiency.

[0029] Of course, neuronal fragments can also be generated natively without relying on the splitting of the parent neural network. The specific method is as follows: 1. Preset core intrinsic parameters for neuronal fragments, wherein the core intrinsic parameters include at least activation threshold and field density preference; 2. Configure the field sensing interface, field excitation interface, field coupling interaction logic, and minimum field response computation capability of the neuronal fragments; 3. Verify and calibrate the core inherent parameters and field coupling capability, and solidify them into the smallest functional unit with field coupling capability, so that it has the ability to sense and analyze, match and discriminate, activate and judge, calculate field response and excitation control.

[0030] Natively constructed neuron fragments can be flexibly designed with fragment parameters and functions according to the actual computing task requirements, realizing customized fragment generation and stronger adaptability.

[0031] Both of the aforementioned methods for generating neuronal fragments must meet the core characteristics of "minimum functional unit, no fixed topological connections between them, and the ability to perform field coupling and activation judgment." Both methods can participate in dynamic networking and collaborative computing through field coupling and field-space interaction, without affecting the implementation of the system's core functions. In practical applications, the appropriate generation method can be selected based on system deployment costs and task adaptation requirements.

[0032] Besides the two generation methods mentioned above, neuronal fragments can also emerge naturally through the long-term evolution of the information field. During the continuous evolution of the information field, the field density distribution adjusts over time, with local areas experiencing repeated increases and decreases in field density. Some field points, having repeatedly been in high-density areas, gradually form stable response patterns, eventually solidifying into the smallest functional units with field perception and response capabilities, becoming neuronal fragments capable of participating in subsequent network computations. This process requires no external parent network or manually preset parameters; it is an intelligent emergence driven by field evolution during system operation, characterized by strong adaptability and the absence of prior knowledge.

[0033] Furthermore, in step B, the ground state field is a globally uniform, steady-state zero-bias field with no field density gradient, no inherent bias, and no preset topology. Each neuronal fragment deployed in the ground state field is in the same initial environment and has no pre-polarization or pre-activation state. In step C, the external input information is mapped and encoded as continuously spatially distributed field density and mode gradient features, which are smoothly projected onto the ground state field in the form of a simulated field.

[0034] The steady-state zero-bias field provides a uniform and interference-free initial environment for neuronal fragments, ensuring the accuracy and consistency of neuronal fragment activation judgment and improving the reliability of calculation results. The encoding method of the external input signal realizes the adaptation of the input signal to the field space, ensuring that the signal can be effectively perceived and processed by the neuronal fragments. At the same time, the design of the continuous field density feature improves the smoothness and integrity of signal transmission. The explicit initial field and input signal constraints further refine the method flow, reduce interference factors in the calculation process, and make the whole method more reproducible. The zero-bias, topology-free ground state field further highlights the innovative point of this invention: no fixed connection and self-organizing network, which is significantly different from the initial settings of traditional neural networks.

[0035] Furthermore, in step D, the method for the neuronal fragment to perform autonomous activation determination and selective activation based on the determination result is as follows: each neuronal fragment determines the field density matching degree between its perceived real-time neighborhood field density and its own inherent field density preference, and then calculates the activation intensity based on the field density matching degree and the neighborhood field density intensity. When the activation intensity reaches or exceeds its activation threshold, the neuronal fragment autonomously triggers activation; when the activation intensity does not reach its activation threshold, the neuronal fragment remains inactive.

[0036] Field density matching degree refers to the similarity between the field density currently perceived by a neuronal fragment and its own field density preference. The value ranges from 0 to 1, with 1 for a perfect match and 0 for a complete mismatch. It is a core parameter for calculating fragment activation intensity. Activation intensity is a quantitative indicator characterizing the degree of activation of a neuronal fragment, and is calculated as: Activation Intensity = Field Density Matching Degree * Local Field Density Intensity.

[0037] The core of the autonomous activation judgment of neuronal fragments is the matching and discrimination between real-time field density and its own inherent field density preference, rather than traditional algorithm calculation. The matching and discrimination here is a responsive judgment made by the response kernel based on its own inherent parameters to the real-time field density information obtained by the field sensing interface, which is part of the field response calculation.

[0038] When the activation intensity reaches or exceeds the activation threshold, the neuron fragments autonomously trigger activation; when the activation intensity does not reach the threshold, they remain inactive. This judgment logic is simple, efficient, and enables parallel autonomous judgment by each fragment without centralized control, aligning with the core requirements of self-organizing networks. Furthermore, following the initial dynamic networking and collaborative computation in step F, there is step F1. Based on the current field density distribution, multiple rounds of field coupling evolution iterations are performed, with an upper limit set for the number of iterations. The field density distribution is continuously updated, and the active region and resonance cluster structure are reconstructed. When the global field density change rate is lower than a preset threshold for several consecutive cycles, it is determined that the information field has converged to a steady-state field density, and then step G is entered for decoding output. If the iteration upper limit is reached but the field density has not converged to a steady-state, step G is entered with the current field density state.

[0039] This optimization step, building upon the initial dynamic networking and collaborative computation in step F, adds a multi-round field coupling evolution iteration step (F1). This addresses the issues of insufficient computational accuracy and the information field not reaching a stable state in the initial networking round, further improving the accuracy and reliability of the computation results. The field coupling evolution iteration here is a process of continuous interaction and adaptive adjustment among neuron fragments through field coupling. It is not the gradient descent iteration of traditional neural networks and does not rely on error backpropagation. Instead, it achieves field density self-convergence through the activation-projection-reactivation cycle of fragments in the field. Its core is the dynamic evolution of field density, which aligns with the field coupling driving mechanism of this invention. During the iteration process, the resonance cluster structure is reconstructed, making the networking more adaptable and capable of handling complex computational tasks. Only when the information field converges to a steady state does the resonance cluster structure stabilize, and the field density distribution no longer changes significantly. At this point, the field density distribution accurately represents the computational results, and then the process proceeds to step G for decoding output, ensuring the accuracy of the computational results.

[0040] Furthermore, during the multi-round field coupling evolution iteration in step F1, and when decoding the field density distribution in step G, a snapshot of the field state of the current resonance cluster structure and field density distribution is captured and retained.

[0041] The captured data includes the current resonant cluster structure and field density distribution. The resonant cluster structure reflects the current dynamic network state, while the field density distribution reflects the current computational progress and result characteristics. The combined field state snapshot provides a complete record of key states during the computation process. Retaining these snapshots allows for subsequent tracking of the computation process, analysis of computational errors, optimization of fragment parameters, and iteration strategies. The capture and retention of field state snapshots enables traceability of the computation process, facilitating troubleshooting and computational error analysis, and improving system maintainability. Snapshot data can also serve as a reference for subsequent computational tasks, enabling the reuse and optimization of computational strategies and improving computational efficiency. Recording the dynamic changes in the resonant cluster structure and field density distribution provides data support for studying the evolution of field coupling and optimizing fragment design.

[0042] Furthermore, the response kernel of the neuronal fragment also incorporates an initial field coupling coefficient. In step D, the method for the neuronal fragment to autonomously activate and selectively activate based on the determination result is as follows: each neuronal fragment determines the field density matching degree between its perceived real-time neighborhood field density and its own inherent field density preference, and then calculates the activation intensity based on the field density matching degree, the initial field coupling coefficient, and the neighborhood field density intensity. When the activation intensity reaches or exceeds its activation threshold, the neuronal fragment autonomously triggers activation; when the activation intensity does not reach its activation threshold, the neuronal fragment remains inactive. In the multi-round field coupling evolution iteration process in step F1, the neuronal fragments that frequently participate in activation and are continuously included in the resonance cluster adaptively adjust their initial field coupling coefficient. After adjusting the initial field coupling coefficient, the neuronal fragments calculate their activation intensity according to the updated field coupling coefficient in the multi-round field coupling evolution iteration process, realizing the dynamic evolution of their field coupling intensity. The adjusted field coupling coefficient is retained as an inherent parameter of the neuronal fragment and can still be maintained after the information field recovers to the ground state.

[0043] The field coupling coefficient in this invention is a dimensionless parameter characterizing the strength of field coupling between neuronal fragments and the field space. Its core function is to quantify three key characteristics of neuronal fragments: first, their sensitivity to changes in field density; second, the strength of their feedback output of field density signals to the field space after activation; and third, the strength of indirect collaboration between neuronal fragments through the field space. Specifically, the field coupling coefficient determines the response amplitude of neuronal fragments to unit changes in field density: the larger the field coupling coefficient, the more sensitive the neuronal fragment is to changes in field density, the stronger the field feedback signal output after activation, and the easier it is to form stable resonance clusters with other fragments that match the field density preference, thereby improving the stability of the resonance clusters and the efficiency of collaborative computation. Its collaborative effect with the field space and other fragments is also more significant. Conversely, the smaller the field coupling coefficient, the less sensitive the neuronal fragment is to changes in field density, the weaker the field feedback signal output, and the lower its participation in dynamic networking and collaborative computation.

[0044] It should be clarified that the field coupling coefficient is a preferred parameter built into the neuron fragment response kernel. When this parameter is missing, the neuron fragments interact according to the default equal-strength field coupling mode, which can still fully realize the basic dynamic networking and collaborative computing functions, only with a reduction in networking accuracy and collaborative efficiency.

[0045] The initial field coupling coefficients are configured differently for neuronal fragments generated through different pathways, specifically in two ways: 1. The initial field coupling coefficient of the neuron fragments generated by splitting the parent neural network is obtained by normalizing the contribution of their corresponding neuron weights in the parent neural network. Specifically, the more important the neuron or the larger its weight in the parent neural network, the higher the initial field coupling coefficient of the resulting neuron fragments; conversely, the smaller the weight of a neuron at the edge of the parent neural network, the lower the initial field coupling coefficient of the resulting neuron fragments, ensuring that the initial field coupling coefficient matches the functional importance of the fragments.

[0046] 2. The initial field coupling coefficient of the neuron fragments generated by native construction is directly obtained through preset configuration. The initial value can be flexibly set according to the needs of specific computing tasks to adapt to the field coupling interaction requirements of different scenarios.

[0047] During the multi-round field coupling evolution iteration in step F1, the information field implicitly accumulates the core state information of each neuron fragment, specifically including activation frequency and residence time in the resonance cluster. Based on the above accumulated information, the information field differentially regulates neuron fragments in different states: for neuron fragments that are frequently activated and continuously included in the resonance cluster, the information field generates an adjustment potential based on their accumulated activity, adaptively increasing the initial field coupling coefficient of such neuron fragments; for neuron fragments with low activity and less participation in network coordination, their field coupling coefficient remains at its initial value and does not increase; for neuron fragments that have not participated in activation for a long time and have not entered the resonance cluster, their field coupling coefficient gradually decays with the field evolution process, thereby realizing the dynamic weakening of field coupling ability and the sparse allocation of field resources, effectively accelerating the convergence of the entire field space and reducing the redundant energy consumption of the system.

[0048] After adaptive enhancement, the field coupling coefficient strengthens the field coupling characteristics of the corresponding neuronal fragments, making them more likely to form field coupling associations with surrounding neuronal fragments and more easily excited by field density signals and quickly incorporated into resonance clusters. Based on this, when similar signals are input subsequently, the system can more quickly form stable field coordination patterns, achieving field-domain association learning. Furthermore, the field coupling coefficient, solidified through multiple rounds of field evolution iterations, is retained as an inherent parameter of the neuronal fragments for a long period. Even if the field space returns to the ground state, this solidified field coupling coefficient will not disappear. This characteristic allows the system to quickly reconstruct the historical resonance cluster structure when the same computational task recurs, thus possessing field-native long-term memory capabilities and improving the computational efficiency of repetitive tasks.

[0049] The present invention also discloses a dynamic networking computing system for neuronal fragments, the system being configured to execute and implement the above-described dynamic networking computing method for neuronal fragments.

[0050] This invention employs a field-coupled, neuronal fragment-driven architecture, overcoming the limitations of traditional neural networks that rely on fixed topologies, explicit weights, and hierarchical computation. Neuronal fragments have no direct connections but achieve dynamic self-organization through information fields, resulting in decentralized and naturally parallel computation, significantly reducing centralized scheduling overhead and achieving high efficiency and energy saving. Individual neuronal fragments possess basic computing power, and multi-fragment networks can generate higher-order inference capabilities. They also possess native field memory characteristics, allowing for rapid reconstruction of resonant structures for repetitive tasks. Overall, this architecture combines high efficiency, adaptability, long-term memory capabilities, and low energy consumption, truly realizing a life-like, growable, and self-healing intelligent network, while exhibiting strong resistance to attacks, crashes, and high stability. Furthermore, this architecture naturally supports cross-domain expansion, requiring no unified hardware environment or central coordinating node. It can be deployed on distributed computing nodes, edge devices, and heterogeneous cloud environments, achieving collaborative networking and unified computing across domain resources. When neuronal fragments move within the network or synchronize from fragment networks in different locations, they have the ability to carry field state information for migration. Without retraining, they can quickly reconstruct the resonance cluster structure in a new environment, achieving seamless cross-domain deployment and plug-and-play functionality, and possessing strong versatility and scalability. Attached Figure Description

[0051] Figure 1 This is a flowchart of Embodiment 1 of the present invention. Detailed Implementation

[0052] The following will further explain the concept, specific structure, and technical effects of the present invention in conjunction with the accompanying drawings, so as to fully understand the purpose, features, and effects of the present invention.

[0053] Example 1: Recognizing the digit content in a handwritten digit image Step A: Generating neuronal fragments This embodiment uses a pre-trained artificial neural network as the parent neural network. This parent neural network is a deep convolutional neural network, pre-trained on the ImageNet dataset, and contains 4 convolutional layers, 4 pooling layers, and 2 fully connected layers, totaling approximately 12 million parameters. The third convolutional layer in the parent neural network is selected as the splitting source layer, which contains 512 feature extraction nodes.

[0054] The parent neural network is structurally decomposed according to the minimum functional unit standard: the 512 feature extraction nodes of the third convolutional layer are decomposed one by one, and each node and its corresponding convolutional kernel weights are used as basic units to form 512 basic neural units. Each basic neural unit inherits its field density preference parameter (obtained by normalizing the distribution of the convolutional kernel weights corresponding to the node) and activation threshold parameter (determined by the average activation value of the node during the pre-training process, ranging from 0.45 to 0.75, with the specific distribution related to the node's functional type) from the parent neural network.

[0055] The basic neural units undergo de-fixed connectivity removal: This process removes the original hierarchical relationships, node dependencies, fixed transmission paths, and fixed-weight connections from each basic neural unit, making it no longer dependent on any fixed topology. After de-fixed connectivity removal, direct connections between basic neural units are no longer retained; only their inherited field density preference parameters and activation threshold parameters are preserved.

[0056] A unified interface is configured for the basic neural units after the fixed connections are removed, and the parameters are integrated into the response kernel: a unified field-sensing interface and field-emission interface are configured for each basic neural unit, and its field density preference parameters and activation threshold parameters are integrated into the response kernel to form neuronal fragments. In this embodiment, a total of 512 neuronal fragments are generated.

[0057] Step B: Deploy neuronal fragments in the information field of the initial steady state. A two-dimensional information field is constructed with a spatial size of 512×512 field points, a field point spacing of 1 unit, and a periodic boundary condition. The information field initialization parameters are: the field density at all field points is set to 0.1, and the mode gradient is set to zero vector. The 512 neuron fragments generated in step A are deployed in this ground-state field, with each fragment occupying a field point position. The initial positions of the fragments in the field space are uniformly distributed. After deployment, the information field is a globally uniform, steady-state, zero-biased field with no field density gradient, no inherent bias, and no preset topology. Each neuron fragment is in the same initial environment, without pre-polarization or pre-activation.

[0058] Step C: Encode the external input information and project it onto the ground state field The external input is a 28×28 pixel handwritten digit image (MNIST dataset sample), with pixel values ​​normalized to the 0-1 range. This image is encoded into continuous field density and mode gradient features adapted to the field space: each pixel value is mapped to the field density value of the corresponding region (pixel value 0.8 maps to field density 0.8), and the image edge gradient is mapped to the mode gradient direction. The encoded simulated field is smoothly projected onto the central region of the ground state field (784 field points, corresponding to a 28×28 pixel region). After projection, the field density in this region increases from 0.1 to the 0.2-0.9 range, and the mode gradient changes from a zero vector to a direction pointing towards the image edge, forming an excitation field with a non-uniform field density distribution.

[0059] Step D: Each neuronal fragment senses in parallel through field coupling and performs autonomous activation judgment. After the external signal is projected, each neuron fragment perceives the field density of its neighboring field points in parallel through the field-sensing interface, including two dimensions: information intensity and pattern gradient. The response kernel performs a field density matching degree determination (the matching degree ranges from 0 to 1, calculated using cosine similarity) on the perceived real-time neighboring field density and its own inherent field density preference. Then, it calculates the activation intensity (activation intensity = field density matching degree × neighboring field density intensity) based on the field density matching degree and the neighboring field density intensity. When the activation intensity reaches or exceeds its activation threshold, the neuron fragment autonomously triggers activation; otherwise, it remains inactive. In this embodiment, 47 out of 512 fragments are activated because they perceive a high matching degree (>0.6) between the field density and their own preference, while the remaining 465 fragments remain inactive.

[0060] Step E: Activated neuronal fragments project field density preferences outward through field coupling. The 47 activated neuronal fragments project their field density preferences (including information intensity, phase, and pattern vector) outward through the field excitation interface. The projection intensity is modulated by the initial field coupling coefficient. The projected signals superimpose and diffuse in the field space, forming a local field density enhancement region. The high field density region diffuses outward, causing neighboring previously inactive neuronal fragments to enter chain activation upon sensing the increase in field density. In this embodiment, the first round of chain activation adds 89 fragments to the activation, bringing the total number of activated fragments to 136.

[0061] Step F: Form a stable resonance cluster and complete the first round of dynamic networking and collaborative computing. The 136 continuously activated neuronal fragments form a high-density field region within their contiguous area, where the neuronal fragments continue to activate and amplify the field density. After field coupling evolution, fragment groups within the high-density region whose field coupling strength exceeds a preset threshold of 0.7 and whose phase difference is ≤0.05 enter a steady-state coupling state, naturally forming two stable resonance clusters. Resonance cluster 1 contains 94 fragments, corresponding to the main digital region in the image; resonance cluster 2 contains 42 fragments, corresponding to the background region in the image. The first round of dynamic networking and collaborative computation is completed, and the computation results are presented in the form of field density distribution.

[0062] Step G: Multi-round field coupling evolution iteration After the first round of networking is completed, the information field enters a multi-round field coupling evolution iteration. The maximum number of iterations is set to 50 rounds, and the convergence judgment threshold is that the global field density change rate is less than 0.001 for three consecutive rounds.

[0063] In each iteration, the current field density distribution is continuously updated, the original resonance cluster structure gradually disintegrates, and the fragments return to an independent activation waiting state. Based on the updated field density distribution, they are reactivated, recoupled, and form new resonance cluster structures. During the iteration process, fragments that frequently participate in activation and are continuously incorporated into the resonance cluster adaptively adjust their initial field coupling coefficients, gradually increasing them (from the initial value towards a preset upper limit, with the upper limit not exceeding 0.9); the field coupling coefficients of fragments that have not been activated for a long time gradually decrease (the lower limit is not lower than 0.2). The adjusted field coupling coefficients are retained as intrinsic parameters of the fragments and can still be maintained after the information field returns to its ground state.

[0064] In this embodiment, the convergence condition was met in the 9th round (the global field density change rate was below 0.001 for three consecutive rounds), and the iteration stopped. After convergence, the two resonance cluster structures stabilized, and the field density distribution no longer changed significantly. During the iteration process, the system captured and retained a snapshot of the field state of the current resonance cluster structure and field density distribution, providing a historical reference for subsequent repeated tasks.

[0065] Step G: Decode and output the calculation result After the information field converges to a steady state, the field density distribution is decoded and output. The average field density pattern vector of the two resonance cluster regions is extracted and compared with the cosine similarity of 10 preset digit category templates (0-9). The category with the highest matching degree is selected as the output result. In this embodiment, the output result is the digit "5" with a confidence level of 96.7%, which is determined to be a valid output. After the output is completed, the information field is reset to the ground state field, and the fragments return to the independent activation waiting state, waiting for the next round of external signal injection.

[0066] Example 2: Generating neuron fragments using native construction methods This embodiment uses a native construction method to generate neuron fragments, without relying on the pre-trained parent neural network splitting, and directly constructs the smallest functional unit with field coupling capability from preset parameters.

[0067] 1. Preset core intrinsic parameters of neuronal fragments This embodiment pre-generates 512 neuron fragments, and the core intrinsic parameters of each fragment include activation threshold and field density preference.

[0068] Activation Threshold: The activation threshold for each fragment is set independently within the range of 0.40-0.80, with the specific value determined based on the preset function type distribution. There are three function types: edge-sensitive (activation threshold 0.40-0.55, 128 in total), texture-sensitive (activation threshold 0.55-0.70, 256 in total), and global structure-sensitive (activation threshold 0.70-0.80, 128 in total). The ratio of fragments of the three types is 1:2:1.

[0069] Field density preference: The field density preference of each fragment includes three dimensions: information intensity preference, phase preference, and mode vector preference.

[0070] Information intensity preference: It is evenly distributed in the range of 0.2-0.9 and is negatively correlated with the activation threshold (the higher the activation threshold, the lower the information intensity preference, so as to avoid the activation threshold and preference values ​​being in the same direction, which would lead to an abnormally high activation probability).

[0071] Phase preference: randomly and uniformly distributed in the interval 0-2π.

[0072] Pattern vector preference: The pattern vector is a 10-dimensional vector, with each dimension taking values ​​in the range [-1, 1], normalized to a unit length. The pattern vector of edge-sensitive fragments is biased towards high-frequency components (the dimensions of the vector alternate between positive and negative values, with a high proportion of high-frequency modes), the pattern vector of texture-sensitive fragments is biased towards mid-frequency modes, and the pattern vector of global structure fragments is biased towards low-frequency modes (the values ​​of each dimension change smoothly).

[0073] Initial field coupling coefficient: The initial field coupling coefficient of each fragment is set in the range of 0.2-0.9, matching the function type. The coupling coefficient of edge-sensitive fragments is set to 0.2-0.4, the coupling coefficient of texture-sensitive fragments is set to 0.4-0.7, and the coupling coefficient of global structure fragments is set to 0.7-0.9.

[0074] Field sensing interface read frequency: The field sensing interface read frequency for all fragments is uniformly set to be synchronized with the information field evolution cycle, and read once per iteration. The read delay shall not exceed one field evolution cycle, and the read error shall not exceed ±0.01.

[0075] Field excitation interface projection intensity and range: The projection intensity is modulated by the field coupling coefficient, and the projection range is limited by the diffusion rules of the information field (diffusion coefficient 0.1, maximum diffusion radius 5 units). The projected signal contains three dimensions: fragment information intensity preference, phase preference, and mode vector preference.

[0076] 2. Interface and interaction logic for configuring neuronal fragments Configure a field-sensing interface, a field-excitation interface, field-coupling interaction logic, and minimum field response computation capability for each preset neuron fragment configuration.

[0077] Field sensing interface configuration: Configure the field sensing interface to read resolution of 10 bits, meaning the quantization precision of the field density value that can be read in each iteration is 0.001. The output of the field sensing interface is a three-channel signal, corresponding to information intensity, phase, and mode gradient vector, respectively.

[0078] Field excitation interface configuration: Configure the output intensity modulation coefficient of the field excitation interface to 0.8-1.2 times, and output a three-channel signal, corresponding to information intensity, phase, and mode vector, respectively. After fragment activation, the excitation interface continuously projects a field density preference signal outward, with the projection duration matching the activation duration, and projection terminates upon activation termination.

[0079] Field coupling interaction logic configuration: Configure the field coupling interaction logic between fragments as follows: fragments only interact indirectly through the information field, without retaining any direct connections. The coupling strength is jointly determined by the field coupling coefficient and the real-time field density matching degree, where coupling strength = field coupling coefficient × field density matching degree. The maximum value of the coupling strength does not exceed 1.0, and the minimum value is not lower than 0.

[0080] Minimum field response computation capability configuration: The minimum field response computation capability of the configured fragments includes: field density matching degree calculation (cosine similarity calculation), activation intensity calculation (activation intensity = field density matching degree × neighborhood field density intensity × field coupling coefficient), activation judgment (activation is triggered when activation intensity ≥ activation threshold), and field density preference projection (projecting its own field density preference outward after activation). All calculations are completed within the fragment and do not depend on external control.

[0081] 3. Verify and calibrate the core inherent parameters and field coupling capability. Functional verification and calibration were performed on 512 neuron fragments, which were then solidified into the smallest functional unit with field coupling capability.

[0082] Verification Test: Each fragment is activated sequentially in the simulated ground-state field, and its activation intensity, activation response time, field density preference projection intensity, and direction are recorded. Verification criteria: Activation intensity error not exceeding ±5%, activation response time not exceeding one field evolution cycle, field density preference projection direction error not exceeding ±3°, and projection intensity error not exceeding ±5%. In the test, the field density of the simulated ground-state field is set to 0.1, the mode gradient is set to zero vector, and all field points are in a uniform state. The test signal is a field density pulse projected at the fragment's location (information intensity 0.2-1.0, phase 0-2π, mode vector is a random vector), used to verify whether the activation response of the fragment under different field density conditions is consistent with the preset parameters.

[0083] Calibration and Adjustment: For fragments whose deviations exceed the allowable range during verification testing (activation intensity error > 5%, response time > 1 evolution cycle, projection direction error > 3°, projection intensity error > 5%), adjust their internal parameters (field density matching degree calculation coefficient, activation intensity calculation weight) to meet the standards, and recalibrate to meet the above verification standards. After calibration, solidify the core inherent parameters and interface configuration of each fragment.

[0084] Verification test results: Of the 512 fragments, 507 passed the verification test, and 5 fragments passed the verification test after calibration. All 512 fragments, after calibration, meet the verification criteria and possess independent field coupling capabilities, forming 512 minimum functional units with field coupling capabilities. Each unit has a field sensing interface, a response kernel, and a field excitation interface. The field density preferences, activation thresholds, initial field coupling coefficients, interface configurations, and computational logic for all fragments have been fixed. The fragments are now generated and can be deployed in the initially stable information field in subsequent steps.

[0085] The preferred embodiments of the present invention have been described in detail above. It should be understood that those skilled in the art can make numerous modifications and variations based on the concept of the present invention without creative effort. Therefore, all technical solutions that can be obtained by those skilled in the art based on the concept of the present invention through logical analysis, reasoning, or limited experimentation on the basis of existing technology should be within the scope of protection defined by the claims.

Claims

1. A neuronal fragment, characterized in that, The neuron fragments are the smallest functional units anchored in the field space and capable of field coupling, with each neuron fragment occupying a field point location. Each smallest functional unit has a field-sensing interface, a response kernel, and a field-excitation interface. The field-sensing interface reads the local field density information within the information field where the neuron fragment is located. The response kernel has built-in inherent field density preferences and activation thresholds, and is used to perform perceptual analysis, matching discrimination, activation judgment, field response calculation, and excitation control on the field density information read by the field-sensing interface. The field-excitation interface projects its own field density preferences outward after the neuron fragment is activated to update the field density state of the information field. The neuron fragments have no fixed topological connections to each other and interact with the information field only through field coupling.

2. A method for dynamic networking computation of neuronal fragments, characterized in that, Includes the following steps: A. Generating the neuronal fragments as described in claim 1; B. Deploy the neuronal fragments in an information field in an initial stable state, wherein the field point parameter structure of the information field is field density and mode gradient; construct a ground state field that is uniform across the entire domain and has stable field density; C. Encode the external input information and project it onto the ground state field to form an excitation field with a non-uniform field density distribution; D. Each neuron fragment senses the field density of its neighboring field points in parallel through field coupling, makes autonomous activation judgments, and achieves selective activation based on the judgment results; E. Activated neuronal fragments project their own field density preferences outward through field coupling to increase the field density of their neighborhood, thereby forming a field density gradient in the information field. The high field density region spreads, thereby chain-activating the previously inactive neuronal fragments in the vicinity. F. Field points in continuous regions between continuously activated neuronal fragments form high-density areas of field density. Neuronal fragments in these high-density areas are continuously activated and enhance the field density. After evolution, they naturally form stable resonance clusters, completing the first round of dynamic networking and collaborative computing. The calculation results are presented in the form of field density distribution. G. Decode the field density distribution of the information field and output the calculation results.

3. The method for dynamic networking of neuronal fragments according to claim 2, characterized in that, The generation of neuronal fragments in step A includes the following steps: A1. Select a pre-trained artificial neural network as the parent neural network; A2. The parent neural network is structurally decomposed according to the minimum functional unit standard to obtain basic neural units. Each basic neural unit inherits the field density preference parameter and activation threshold parameter corresponding to it in the parent neural network. A3. Perform de-fixed connection processing on the basic neural units to remove their original hierarchical relationships, node dependencies, fixed transmission paths and fixed weight connections; A4. Configure a unified field-sensing interface and field-activation interface for the basic neural units after defixed connections, and integrate their field density preference parameters and activation threshold parameters into the response kernel to form the neuronal fragments.

4. The method for dynamic networking of neuronal fragments according to claim 2, characterized in that, The method for generating fragmented neurons in step A is as follows: a1. Preset core intrinsic parameters of neuronal fragments, wherein the core intrinsic parameters include at least activation threshold and field density preference; a2. Configuration of the field sensing interface, field excitation interface, field coupling interaction logic, and minimum field response computation capability of the neuronal fragments; a3. Verify and calibrate the core inherent parameters and field coupling capabilities, and solidify them into the smallest functional unit with field coupling capabilities, enabling it to have the capabilities of perception analysis, matching discrimination, activation judgment, field response calculation and excitation control.

5. The method for dynamic networking of neuronal fragments according to claim 2, characterized in that, In step B, the ground state field is a steady-state zero-bias field that is uniform across the entire domain, has no field density gradient, no inherent bias, and no preset topology. Each neuronal fragment deployed in the ground state field is in the same initial environment and has no pre-polarization or pre-activation state. In step C, the external input information is mapped and encoded as field density and mode gradient features with continuous spatial distribution, and smoothly projected onto the ground state field in the form of a simulated field.

6. The method for dynamic networking of neuronal fragments according to claim 2, characterized in that, In step D, the method for the neuronal fragment to perform autonomous activation determination and selective activation based on the determination result is as follows: each neuronal fragment determines the field density matching degree between its perceived real-time neighborhood field density and its own inherent field density preference, and then calculates the activation intensity based on the field density matching degree and the neighborhood field density intensity. When the activation intensity reaches or exceeds its activation threshold, the neuronal fragment autonomously triggers activation; when the activation intensity does not reach its activation threshold, the neuronal fragment remains inactive.

7. The method for dynamic networking calculation of neuronal fragments according to any one of claims 2 to 6, characterized in that, After the first round of dynamic networking and collaborative computing in step F, there is step F1. Based on the current field density distribution, multiple rounds of field coupling evolution iteration are performed, while setting an upper limit for the number of iterations. The field density distribution is continuously updated, and the active region and resonance cluster structure are reconstructed. When the global field density change rate is lower than the preset threshold for several consecutive cycles, it is determined that the information field has converged to the field density steady state, and then step G is entered for decoding output. When the upper limit of iteration is reached but the field density has not converged to the steady state, step G is entered with the current field density state.

8. The method for dynamic networking of neuronal fragments according to claim 7, characterized in that, During the multi-round field coupling evolution iteration in step F1, and when decoding the field density distribution in step G, a snapshot of the field state of the current resonance cluster structure and field density distribution is captured and retained.

9. The method for dynamic networking of neuronal fragments according to claim 7, characterized in that, The response kernel of the neuron fragment also has an initial field coupling coefficient built in. In step D, the method for the neuron fragment to perform autonomous activation determination and selective activation based on the determination result is as follows: each neuron fragment determines the field density matching degree between its perceived real-time neighborhood field density and its own inherent field density preference, and then calculates the activation intensity based on the field density matching degree, the initial field coupling coefficient, and the neighborhood field density intensity. When the activation intensity reaches or exceeds its activation threshold, the neuron fragment autonomously triggers activation; when the activation intensity does not reach its activation threshold, the neuron fragment remains inactive. In the multi-round field coupling evolution iteration process in step F1, the neuron fragments that frequently participate in activation and are continuously included in the resonance cluster adaptively adjust their initial field coupling coefficient. After adjusting the initial field coupling coefficient, the neuron fragment calculates its activation intensity according to the updated field coupling coefficient in the multi-round field coupling evolution iteration process, realizing the dynamic evolution of its field coupling intensity. The adjusted field coupling coefficient is retained as an inherent parameter of the neuron fragment and can still be maintained after the information field recovers to the ground state.

10. A dynamic networking computing system for neuronal fragments, characterized in that, The system is configured to execute and implement the method of any one of claims 2 to 9.

Images