A training method, device and electronic equipment based on a blockchain-based training architecture

Abstract

The application belongs to the field of artificial intelligence, and provides a training method and device of a training architecture based on a blockchain and electronic equipment, the method comprising: generating a training task through a task management contract of a blockchain in the training architecture, the training task comprising training task parameters, the training task parameters comprising: a stimulus category and an input sequence; executing neural training based on the received training task through a plurality of biological neuron network nodes in the training architecture to generate node response results of the plurality of biological neuron network nodes; and dynamically adjusting the training task parameters and a stimulation scheme based on an evaluation result and an aggregation result through a training coordinator in the training architecture. The training method provided in the application embodiment can not only execute neural training based on the received training task to generate node response results of the plurality of biological neuron network nodes, but also dynamically adjust the training task parameters and the stimulation scheme based on the evaluation result and the aggregation result.

Description

Technical Field

[0001] This invention belongs to the field of artificial intelligence, and in particular relates to a training method, apparatus and electronic device based on a blockchain-based training architecture. Background Technology

[0002] With the development of neuromorphic computing, biological neural networks, as computing carriers with advantages of synaptic plasticity and low power consumption, have gradually become an important direction in artificial intelligence research. However, due to their physical properties being highly dependent on the culture environment and exhibiting heterogeneity, it is difficult for multiple biological neural networks to be trained collaboratively in a unified model.

[0003] Furthermore, the lack of transparency and unreliability of traditional centralized management methods in collaborative training of biological neural networks has become a significant bottleneck restricting its development.

[0004] Therefore, how to provide a training method that can collaboratively train multi-neuron networks is a technical problem to be solved. Summary of the Invention

[0005] Therefore, it is necessary to address the shortcomings of existing training methods in coordinating the training of multi-neuron networks by providing a training method, device, storage medium, and electronic device based on a blockchain-based training architecture.

[0006] In a first aspect, embodiments of the present invention provide a training method based on a blockchain-based training architecture, the method comprising: Training tasks are generated through a task management contract on the blockchain within the training architecture. These training tasks include training task parameters, which include: stimulus types and input sequences. Based on the received training task, neural training is performed using multiple biological neuron network nodes in the training architecture to generate node response results for multiple biological neuron network nodes. The training coordinator in the training architecture dynamically adjusts the training task parameters and stimulation scheme based on the evaluation results and aggregation results. The evaluation results are obtained by evaluating the training effect of the node response results of multiple biological neuron network nodes, and the aggregation results are obtained by aggregating the node response results of multiple biological neuron network nodes according to the set logic.

[0007] Optionally, before generating the training task through the task management contract of the blockchain, the method further includes: The task management contract and the data storage contract are deployed on the blockchain, wherein the task management contract is used to publish and manage the corresponding training tasks, and the data storage contract is used to store the training results of the multi-neuron network nodes.

[0008] Optionally, any one of the plurality of biological neuron network nodes includes: a local controller, a stimulation controller, a microelectrode array, and a neuron culture unit; the method further includes: The following operations are performed sequentially via the local controller, the stimulation controller, the microelectrode array, and the neuron culture unit: Perform the following operations through the local controller: In response to receiving the training task, the electrical stimulation command control logic is triggered. Receive the training data returned by the microelectrode array; The training data, signal features, and training summary are integrated to generate response data; The response data is uploaded to the blockchain; The following operations are performed via the stimulation controller: Based on the received training task, a preset current stimulation signal and a preset voltage stimulation signal are generated; and voltage pulses and current pulses are output to the microelectrode array. The following operations are performed using the microelectrode array: Receive stimulation signals from the stimulation controller and apply electrical stimulation to nerve tissue; Real-time acquisition of training data for the neural network, the training data including: discharge mode and reaction timing; The following operations are performed using the neuron culture unit: Receive electrical stimulation from a microelectrode array and perform training based on the electrical stimulation.

[0009] Optionally, the training coordinator includes: a data collection module, a response aggregation module, a training evaluation module, a policy generation module, a task publishing module, and a feedback incentive module; the method further includes: The following operations are performed sequentially through the data collection module, the response aggregation module, the training evaluation module, the policy generation module, the task publishing module, and the feedback incentive module: The following operations are performed through the data collection module: Listen to blockchain data storage contract events and read node response data uploaded from multiple biological neural network nodes in the blockchain; The following operations are performed through the response aggregation module: The node response data is aggregated and analyzed according to a preset algorithm to obtain the aggregation result; and global training features are extracted based on the aggregation result. The training evaluation module performs the following operations: The current training state is evaluated based on the aggregation results to obtain the evaluation results, which include: convergence and performance metrics. The strategy generation module performs the following operations: Based on the evaluation results, multiple training parameters are dynamically adjusted, including at least: stimulation frequency, intensity, and duration; Perform the following operations through the task publishing module: By calling the blockchain task management contract, new training tasks and new training parameters are published; The feedback incentive module performs the following operations: The node response quality is evaluated based on the node response data to obtain an evaluation result; and corresponding on-chain incentive tokens are allocated based on the evaluation result.

[0010] Optionally, before generating the training task through the task management contract of the blockchain, the method further includes: A training architecture for collaborative training of multi-neuronal networks based on blockchain is constructed. The training architecture includes: multiple biological neuron network nodes, a blockchain capable of generating tokens, and a training coordinator. Each of the multiple biological neuron network nodes is configured with a corresponding stimulus generator, an EEG monitoring interface, and a local controller.

[0011] Optionally, after dynamically adjusting the training task parameters and stimulus scheme based on the evaluation results and aggregation results, the method further includes: The response results of the multiple biological neuron network nodes are evaluated to obtain target response nodes that meet preset quality requirements; and corresponding on-chain incentive tokens are allocated to the target response nodes.

[0012] Optionally, the method further includes: Initiate training in response to the training strategy; After each round of training, each biological neuron network node submits a corresponding training summary; the training summary includes at least: the unique identifier of the current node, the unique identifier of the current task, the neural stimulation pattern determined by the type of stimulation sequence in this round, the delay from stimulation to the first neural response, the average firing frequency per unit time, the entropy value of the change in neuronal synaptic weight distribution, the timestamp, the hash calculated from the potential waveform or encoded sequence, and the signature generated in the trusted execution environment. Based on the training summary, the training coordinator sequentially performs the following steps: filtering valid data; filtering each biological neuron network node to obtain activated nodes; and generating the training strategy for the next round. Filter valid data by performing the following steps: Signature verification is performed by verifying the match between the signature and the node's identity. Check whether the current task is within its validity period based on the time window; Verify that the task matches the task identifier; The next training strategy is generated by performing the following operations in sequence: The stimulation parameters are optimized, the incentive nodes are selected, and the training strategy for the next round is encoded and published by calling the task management contract; Specifically, the stimulation parameters are optimized by performing the following operations: based on the high-scoring nodes of the previous round, effective stimulation types and stimulation parameters are extracted to obtain effective stimulation types and stimulation parameters, so as to generate the stimulation sequence for the next round based on the stimulation parameters.

[0013] Secondly, embodiments of the present invention provide a training device based on a blockchain-based training architecture, the device comprising: The blockchain task management contract in the training architecture is used to generate training tasks, which include training task parameters, including stimulus types and input sequences. The training architecture includes multiple biological neuron network nodes, which are used to perform neural training based on the received training task to generate node response results for multiple biological neuron network nodes. The training coordinator in the training architecture is used to dynamically adjust the training task parameters and stimulation scheme based on the evaluation results and the aggregation results. The evaluation results are obtained by evaluating the training effect of the node response results of multiple biological neuron network nodes, and the aggregation results are obtained by aggregating the node response results of multiple biological neuron network nodes according to the set logic.

[0014] Thirdly, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed in a computer, causes the computer to perform the method of the first aspect.

[0015] Fourthly, an electronic device is provided, including a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to implement the method of the first aspect.

[0016] In this embodiment of the invention, a training task is generated through a blockchain task management contract in the training architecture. The training task includes training task parameters, such as stimulus types and input sequences. Based on the received training task, neural training is performed on multiple biological neuron network nodes in the training architecture to generate node response results for the multiple biological neuron network nodes. Furthermore, a training coordinator in the training architecture dynamically adjusts the training task parameters and stimulus scheme based on evaluation results and aggregation results. The evaluation results are obtained by assessing the training effect of the node response results of the multiple biological neuron network nodes, and the aggregation results are obtained by aggregating the node response results of the multiple biological neuron network nodes according to a set logic. The training method provided by this embodiment of the invention can not only perform neural training based on the received training task to generate node response results for multiple biological neuron network nodes, but also dynamically adjust the training task parameters and stimulus scheme based on evaluation results and aggregation results. Attached Figure Description

[0017] Exemplary embodiments of the present invention can be more fully understood by referring to the accompanying drawings. The drawings are provided to further illustrate the embodiments of the present invention and form part of the specification. They are used together with the embodiments of the present invention to explain the invention and do not constitute a limitation thereof. In the drawings, the same reference numerals generally represent the same components or steps.

[0018] Figure 1 A flowchart illustrating a training method based on a blockchain-based training architecture according to an exemplary embodiment of the present invention; Figure 2 This is a schematic diagram of the structure of a training system based on a blockchain training architecture. Figure 3 This is a schematic diagram of the structure of a BNN node; Figure 4 This is a schematic diagram of the structure of a training device 400 based on a blockchain-based training architecture according to an exemplary embodiment of the present invention. Detailed Implementation

[0019] Exemplary embodiments of the invention will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the invention are shown in the drawings, it should be understood that the invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided to enable a more thorough understanding of the invention and to fully convey the scope of the invention to those skilled in the art.

[0020] It should be noted that, unless otherwise stated, the technical or scientific terms used in this invention should have the ordinary meaning as understood by one of ordinary skill in the art.

[0021] Furthermore, the terms "first" and "second," etc., are used to distinguish different objects, not to describe a specific order. Additionally, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to those processes, methods, products, or devices.

[0022] This invention provides a training method and apparatus, a computer-readable medium, and an electronic device based on a blockchain-based training architecture, which will be described below with reference to the accompanying drawings.

[0023] Please refer to Figure 1 It illustrates flowcharts of training methods based on blockchain training architectures provided by some embodiments of the present invention, such as... Figure 1 As shown, the training method based on the blockchain training architecture provided in this embodiment of the invention may include the following steps: Step S101: Generate a training task through the task management contract of the blockchain in the training architecture. The training task includes training task parameters, which include: stimulus type and input sequence.

[0024] In one example, before generating the training task through the blockchain task management contract, the training method based on the blockchain-based training architecture provided in this embodiment of the invention may further include the following steps: Deploy task management contracts and data storage contracts on the blockchain. The task management contract is used to publish and manage the corresponding training tasks, and the data storage contract is used to store the training results of the multi-neuron network nodes.

[0025] In one example, before generating the training task through the blockchain task management contract, the training method based on the blockchain-based training architecture provided in this embodiment of the invention may further include the following steps: A training architecture for collaborative training of multi-neuronal networks based on blockchain is constructed. The training architecture includes: multiple biological neuron network nodes, a blockchain capable of generating tokens, and a training coordinator. Each of the multiple biological neuron network nodes is configured with a corresponding stimulus generator, EEG monitoring interface, and local controller.

[0026] Step S102: By training multiple biological neuron network nodes in the training architecture, neural training is performed based on the received training task to generate node response results of multiple biological neuron network nodes.

[0027] In one example, any one of the multiple biological neuron network nodes includes: a local controller, a stimulation controller, a microelectrode array, and a neuron culture unit.

[0028] The training method based on the blockchain training architecture provided in this embodiment of the invention may further include the following steps: The following operations are performed sequentially through the local controller, stimulation controller, microelectrode array, and neuron culture unit: Perform the following operations through the local controller: In response to receiving a training task, the electrical stimulation command control logic is triggered. Receive training data returned by the microelectrode array; The training data, signal features, and training summary are integrated to generate response data; Upload the response data to the blockchain; Perform the following actions via the stimulus controller: Based on the received training task, a preset current stimulation signal and a preset voltage stimulation signal are generated; and voltage pulses and current pulses are output to the microelectrode array. Perform the following operations using a microelectrode array: It receives stimulation signals from the stimulation controller and provides electrical stimulation to nerve tissue; Real-time acquisition of neural network training data, including discharge patterns and reaction timing; Perform the following operations using the neuron culture unit: It receives electrical stimulation from a microelectrode array and performs training based on the electrical stimulation.

[0029] Step S103: The training task parameters and stimulation scheme are dynamically adjusted based on the evaluation results and aggregation results by the training coordinator in the training architecture. The evaluation results are obtained by evaluating the training effect of the node response results of multiple biological neuron network nodes. The aggregation results are obtained by aggregating the node response results of multiple biological neuron network nodes according to the set logic.

[0030] In one example, the training coordinator includes: a data collection module, a response aggregation module, a training evaluation module, a policy generation module, a task publishing module, and a feedback incentive module.

[0031] The training method based on the blockchain training architecture provided in this embodiment of the invention may further include the following steps: The following operations are performed sequentially through the data collection module, response aggregation module, training and evaluation module, policy generation module, task deployment module, and feedback and incentive module: Perform the following operations through the data collection module: Listen to blockchain data storage contract events and read node response data uploaded from multiple biological neural network nodes in the blockchain; Perform the following actions by responding to the aggregation module: The node response data is aggregated and analyzed according to a preset algorithm to obtain the aggregation result; and global training features are extracted based on the aggregation result. Perform the following operations using the training evaluation module: The current training state is evaluated based on the aggregation results, and the evaluation results are obtained. The current training state includes: convergence and performance indicators. Perform the following operations through the policy generation module: Based on the evaluation results, multiple training parameters are dynamically adjusted. These multiple training parameters include at least: stimulation frequency, intensity, and duration. Perform the following operations through the task publishing module: By calling the blockchain task management contract, new training tasks and new training parameters are published; Perform the following operations through the feedback incentive module: The quality of node responses is evaluated based on node response data to obtain evaluation results; and corresponding on-chain incentive tokens are allocated based on the evaluation results.

[0032] In one example, after dynamically adjusting the training task parameters and stimulus scheme based on the evaluation results and aggregation results, the training method of the blockchain-based training architecture provided in this embodiment of the invention may further include the following steps: The response results of multiple biological neuron network nodes are evaluated to obtain target response nodes that meet preset quality requirements; and corresponding on-chain incentive tokens are allocated to the target response nodes.

[0033] In one example, the training method based on the blockchain training architecture provided in this embodiment of the invention may further include the following steps: Initiate training in response to the training strategy; After each round of training, each biological neuron network node submits a corresponding training summary; the training summary includes at least: the current node's unique identifier, the current task's unique identifier, the neural stimulation pattern determined by the type of stimulus sequence in this round, the delay from stimulation to the first neural response, the average firing frequency per unit time, the entropy value of the change in neuronal synaptic weight distribution, the timestamp, the hash calculated from the potential waveform or encoded sequence, and the signature generated in the trusted execution environment. The training coordinator, based on the training summary, sequentially performs the following steps: filtering valid data; filtering each biological neuron network node to obtain activated nodes; and generating the training strategy for the next round. Filter valid data by performing the following steps: Signature verification is performed by verifying the match between the signature and the node's identity. Check whether the current task is within its validity period based on the time window; Verify that the task matches the task identifier; The next training strategy is generated by performing the following operations in sequence: The stimulation parameters are optimized, the incentive nodes are selected, and the training strategy for the next round is encoded and published by calling the task management contract; Specifically, the stimulation parameters are optimized by performing the following operations: based on the high-scoring nodes of the previous round, effective stimulation types and stimulation parameters are extracted to obtain effective stimulation types and stimulation parameters, so as to generate the stimulation sequence for the next round based on the stimulation parameters.

[0034] In a specific application scenario, the submitted training summary is as follows:

[0035] In practical applications, the evaluation scoring model can use the following formula: Formula (1); In the above formula (1), , indicating whether learning has occurred; response_latency_ms is the delay from stimulus to the first neural response, used to judge reaction speed and stimulus effectiveness; mean_spike_rate is the average firing rate per unit time, used to measure the activity level of neurons; coefficients α, β, and γ can be set according to task objectives (such as enhancing reaction speed, synaptic adjustment, etc.).

[0036] In a specific application scenario, after evaluating all nodes, the coordinator generates the next round of training strategy, including: Step a1: Optimize the stimulation parameters.

[0037] Based on the high-scoring nodes from the previous round, extract their most effective stimulus types and parameters, and use them to generate the stimulus sequence for the next round. If a stimulus leads to generally low potential and high surge, increase its probability of occurrence; if a stimulus causes a significant increase in synaptic entropy, it indicates that learning has occurred and is preferentially retained.

[0038] Step a2: Select the excitation node.

[0039] Select only a subset of nodes to participate in the incentive: Select the Top K nodes based on their scores. K can be set according to the needs of different application scenarios, and no specific restrictions are imposed here.

[0040] Step a3: The new strategy is coded and released.

[0041] The new strategy is coded and published by calling the task management contract.

[0042] like Figure 2 As shown, the training system based on the blockchain training architecture includes the following modules: Multiple biological neuron network nodes (BNN modules), each equipped with a stimulus generator, an EEG monitoring interface, and a local controller.

[0043] like Figure 3 The diagram shown is a schematic of the structure of a BNN node.

[0044] Each BNN node is essentially a composite system of experimentation and computation, and the node composition is as follows: Figure 3 As shown, it can be divided into four main modules: Local Controller: The local controller is typically an embedded device or a lightweight server used for blockchain interaction and control of experimental processes. It receives tasks by periodically polling or subscribing to blockchain events, triggering the electrical stimulation command control logic upon receiving a task. Simultaneously, it receives training data returned by the microelectrode array, analyzes signal characteristics, generates summaries, integrates this data into response data, and uploads all response data to the blockchain.

[0045] Stimulation controller: Essentially a circuit control module used to receive task instructions, generate preset current and voltage stimulation signals, and output voltage and current pulses to the microelectrode array.

[0046] Microelectrode arrays: Essentially, they are electrode arrays (miniature sensors) that serve as the physical interface for contacting nerve cells, closely attached to nerve tissue. They have two main functions: first, to receive stimulation signals from the stimulation controller and electrically stimulate the nerve tissue; and second, to collect training data such as the firing patterns and reaction timing of the neural network in real time.

[0047] Neuron culture unit: contains living neurons (such as rat neural stem cells) that receive electrical stimulation from a microelectrode array for training.

[0048] The training method based on the blockchain training architecture provided in this embodiment of the invention can effectively guarantee the data credibility of BNN nodes.

[0049] In practical applications, each BNN node autonomously generates, collects, and calculates training results before uploading them to the blockchain, which carries the risk of data falsification. The blockchain-based training architecture training method provided in this invention employs a trusted execution environment. The data transmission and computation of both the local controller and the stimulus controller are executed within this environment, ensuring the authenticity of stimuli and responses, thereby significantly improving training accuracy.

[0050] A trusted execution environment was chosen because the electrophysiological data output by biological neural networks (such as synaptic entropy, sequential firing, and noise distribution) is high-dimensional, floating-point, and unsuitable for direct conversion into ZKP (Zero Knowledge Proof) circuits. The ZKP generation process places excessive demands on the local controller computing power of BNN nodes, which is not conducive to deployment in biological experimental equipment.

[0051] The task management contract and data storage contract are deployed on the blockchain. The task management contract is used to publish and manage the corresponding training tasks, and the data storage contract is used to store the training results of the multi-neuron network nodes.

[0052] In practical applications, task management contracts are run for task scheduling, response summary recording, node performance evaluation, and reward distribution.

[0053] Train the coordinator to analyze task execution results and generate aggregation strategies.

[0054] The training coordinator is responsible for receiving neural response return data uploaded from multiple BNN nodes, performing training effect evaluation and response aggregation, and then dynamically adjusting training task parameters and stimulus schemes. All training transmission data is recorded within this module. The training coordinator includes the following modules: Data collection module: Monitors blockchain data storage contract events and reads response data uploaded by multiple BNN nodes from the blockchain. Response aggregation module: Performs aggregation analysis on node responses according to a preset algorithm to extract global training features.

[0055] Training evaluation module: Evaluates the current training status based on the aggregated results, such as convergence and performance metrics.

[0056] Strategy generation module: Dynamically adjusts training parameters such as stimulus frequency, intensity, and duration based on evaluation results.

[0057] Task publishing module: Calls the blockchain task management smart contract to publish new training tasks and parameters.

[0058] Feedback Incentive Module: This module allocates on-chain incentive tokens based on the quality of node responses, encouraging active participation. In practical application scenarios, the following methods are adopted: Figure 2 The training process of the training system shown is as follows: Step b1: The task management contract on the blockchain generates training task parameters (including stimulus types, input sequences, etc.). Step b2: After receiving the task, each biological neuron network node performs neural training and collects response data. Then, it extracts electrical activity indicators (such as response delay, average firing rate, synaptic change entropy, etc.), generates a training summary, and uploads it to the blockchain. Step b3: The training coordinator aggregates the response results of multiple BNN nodes according to the set logic and generates the guidance strategy for the next round of training; Step b4: The blockchain allocates on-chain token incentives to nodes with high-quality responses based on response quality assessment.

[0059] In practical applications, the incentive mechanism involves issuing tokens to nodes based on participation, response speed, and learning improvement.

[0060] It should be noted that blockchain, as a distributed ledger technology, has the characteristics of decentralization, immutability, and verifiability, making it naturally suitable as a coordination and recording platform among multiple heterogeneous nodes.

[0061] The training method provided in this invention not only performs neural training based on received training tasks to generate node response results for multiple biological neuron network nodes, but also dynamically adjusts training task parameters and stimulation schemes based on evaluation and aggregation results. Furthermore, the training method utilizes blockchain to achieve trusted task scheduling and response recording in multi-BNN collaborative training. Additionally, the training method effectively avoids privacy leaks and performance bias issues caused by centralized scheduling centers. Moreover, the training method supports neural networks trained under different physical conditions to participate in training together, enhancing the robustness of the training architecture. Furthermore, the training method does not require trust in the node operators; even if BNN nodes are run by different laboratories, as long as their trusted execution environment is trustworthy, their participation results can be accepted by the training architecture. Furthermore, the training method constructs on-chain trajectories for neural response data, which can be used for monitoring, interpretability research, and retraining traceability. Finally, the training method has good scalability and can be extended to biological distributed inference networks in the future.

[0062] In the above embodiments, a training method based on a blockchain-based training architecture is provided. Correspondingly, the present invention also provides a training apparatus based on a blockchain-based training architecture. The training apparatus based on a blockchain-based training architecture provided in the embodiments of the present invention can implement the above-described training method based on a blockchain-based training architecture. This training apparatus based on a blockchain-based training architecture can be implemented through software, hardware, or a combination of both. For example, the multi-version code processing apparatus may include integrated or separate functional modules or units to execute the corresponding steps in the above methods.

[0063] Please refer to Figure 4 This illustration shows a schematic diagram of a training apparatus based on a blockchain-based training architecture provided by some embodiments of the present invention. Since the apparatus embodiments are substantially similar to the method embodiments, the description is relatively simple; relevant details can be found in the description of the method embodiments. The apparatus embodiments described below are merely illustrative.

[0064] like Figure 4 As shown, the training device 400 based on the blockchain training architecture may include: The blockchain task management contract 401 in the training architecture is used to generate training tasks. The training tasks include training task parameters, which include: stimulus types and input sequences. Multiple biological neuron network nodes 402 in the training architecture are used to perform neural training based on the received training task to generate node response results of multiple biological neuron network nodes; The training coordinator 403 in the training architecture is used to dynamically adjust the training task parameters and stimulation scheme based on the evaluation results and the aggregation results. The evaluation results are obtained by evaluating the training effect of the node response results of multiple biological neuron network nodes, and the aggregation results are obtained by aggregating the node response results of multiple biological neuron network nodes according to the set logic.

[0065] In some embodiments of the present invention, the training device 400 based on the blockchain training architecture may further include: Deployment module (in) Figure 4 (Not shown in the image) is used to deploy a task management contract and a data storage contract on the blockchain before generating training tasks through the blockchain's task management contract. The task management contract is used to publish and manage the corresponding training tasks, and the data storage contract is used to store the training results of the multi-neuron network nodes.

[0066] In some embodiments of the present invention, the training device 400 based on the blockchain training architecture may further include: Any node in a multi-branch neuronal network includes: a local controller, a stimulation controller, a microelectrode array, and a neuron culture unit; First execution module (in) Figure 4 (Not shown in the image), used to perform the following operations sequentially through the local controller, stimulation controller, microelectrode array, and neuron culture unit: Perform the following operations through the local controller: In response to receiving a training task, the electrical stimulation command control logic is triggered. Receive training data returned by the microelectrode array; The training data, signal features, and training summary are integrated to generate response data; Upload the response data to the blockchain; Perform the following actions via the stimulus controller: Based on the received training task, a preset current stimulation signal and a preset voltage stimulation signal are generated; and voltage pulses and current pulses are output to the microelectrode array. Perform the following operations using a microelectrode array: It receives stimulation signals from the stimulation controller and provides electrical stimulation to nerve tissue; Real-time acquisition of neural network training data, including discharge patterns and reaction timing; Perform the following operations using the neuron culture unit: It receives electrical stimulation from a microelectrode array and performs training based on the electrical stimulation.

[0067] In some embodiments of the present invention, the training device 400 based on the blockchain training architecture may further include: The training coordinator includes: a data collection module, a response aggregation module, a training evaluation module, a policy generation module, a task publishing module, and a feedback and incentive module; The second execution module (in) Figure 4 (not shown in the image), used to perform the following operations sequentially through the data collection module, response aggregation module, training and evaluation module, policy generation module, task publishing module, and feedback and incentive module: Perform the following operations through the data collection module: Listen to blockchain data storage contract events and read node response data uploaded from multiple biological neural network nodes in the blockchain; Perform the following actions by responding to the aggregation module: The node response data is aggregated and analyzed according to a preset algorithm to obtain the aggregation result; and global training features are extracted based on the aggregation result. Perform the following operations using the training evaluation module: The current training state is evaluated based on the aggregation results, and the evaluation results are obtained. The current training state includes: convergence and performance indicators. Perform the following operations through the policy generation module: Based on the evaluation results, multiple training parameters are dynamically adjusted. These multiple training parameters include at least: stimulation frequency, intensity, and duration. Perform the following operations through the task publishing module: By calling the blockchain task management contract, new training tasks and new training parameters are published; Perform the following operations through the feedback incentive module: The quality of node responses is evaluated based on node response data to obtain evaluation results; and corresponding on-chain incentive tokens are allocated based on the evaluation results.

[0068] In some embodiments of the present invention, the training device 400 based on the blockchain training architecture may further include: Build modules (in) Figure 4 (Not shown in the image) is used to build a training framework for collaborative training of multi-neuronal networks based on blockchain before generating training tasks through blockchain task management contracts. The training framework includes: multiple biological neuron network nodes, a blockchain capable of generating tokens, and a training coordinator. Each of the multiple biological neuron network nodes is configured with a corresponding stimulus generator, EEG monitoring interface, and local controller.

[0069] In some embodiments of the present invention, the training device 400 based on the blockchain training architecture may further include: Evaluation module (in) Figure 4 (not shown in the text) is used to: evaluate the response results of multiple biological neuron network nodes after dynamically adjusting the training task parameters and stimulation scheme based on the evaluation results and aggregation results, and obtain the target response nodes that meet the preset quality requirements; Allocation module (in) Figure 4 (Not shown in the image), used for: allocating corresponding on-chain incentive tokens to the target response node.

[0070] In some embodiments of the present invention, the training device 400 based on the blockchain training architecture may further include: Startup module (in) Figure 4 (Not shown in the image) In response to the training strategy, training is initiated; Submit module (in) Figure 4 (not shown in the image) is used to allow each biological neuron network node to submit a corresponding training summary after each round of training. The training summary includes at least: the unique identifier of the current node, the unique identifier of the current task, the neural stimulation pattern determined by the type of stimulation sequence in this round, the delay from stimulation to the first neural response, the average firing frequency per unit time, the entropy value of the change in neuronal synaptic weight distribution, the timestamp, the hash calculated from the potential waveform or encoded sequence, and the signature generated in the trusted execution environment. The third execution module (in) Figure 4 (Not shown in the image) is used by the training coordinator to sequentially perform the following based on the training summary: filtering valid data; filtering each biological neuron network node to obtain activated nodes; and generating the training strategy for the next round: Filter valid data by performing the following steps: Signature verification is performed by verifying the match between the signature and the node's identity. Check whether the current task is within its validity period based on the time window; Verify that the task matches the task identifier; The next training strategy is generated by performing the following operations in sequence: The stimulation parameters are optimized, the incentive nodes are selected, and the training strategy for the next round is encoded and published by calling the task management contract; Specifically, the stimulation parameters are optimized by performing the following operations: based on the high-scoring nodes of the previous round, effective stimulation types and stimulation parameters are extracted to obtain effective stimulation types and stimulation parameters, so as to generate the stimulation sequence for the next round based on the stimulation parameters.

[0071] In some embodiments of the present invention, the training device 400 based on the blockchain training architecture provided in the present invention is based on the same inventive concept and has the same beneficial effects as the training method based on the blockchain training architecture provided in the foregoing embodiments of the present invention.

[0072] According to another embodiment, a computer-readable storage medium is also provided, on which a computer program is stored, which, when executed in a computer, causes the computer to perform a combination Figure 1 The method described.

[0073] According to another embodiment, an electronic device is also provided, including a memory and a processor, wherein the memory stores executable code, and when the processor executes the executable code, it implements a combination... Figure 1 The method described.

[0074] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using hardware, software, firmware, or any combination thereof. When implemented in software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A training method based on a blockchain-based training architecture, characterized in that, The method includes: Training tasks are generated through a task management contract on the blockchain within the training architecture. These training tasks include training task parameters, which include: stimulus types and input sequences. Based on the received training task, neural training is performed using multiple biological neuron network nodes in the training architecture to generate node response results for multiple biological neuron network nodes. The training coordinator in the training architecture dynamically adjusts the training task parameters and stimulus schemes based on the evaluation and aggregation results. The training coordinator includes: a data collection module, a response aggregation module, a training evaluation module, a policy generation module, a task publishing module, and a feedback incentive module. The method further includes: The following operations are performed sequentially through the data collection module, the response aggregation module, the training evaluation module, the policy generation module, the task publishing module, and the feedback incentive module: The following operations are performed through the data collection module: Listen to blockchain data storage contract events and read node response data uploaded from multiple biological neural network nodes in the blockchain; The following operations are performed through the response aggregation module: The node response data is aggregated and analyzed according to a preset algorithm to obtain the aggregation result; and global training features are extracted based on the aggregation result. The training evaluation module performs the following operations: The current training state is evaluated based on the aggregation results to obtain the evaluation results, which include: convergence and performance metrics. The strategy generation module performs the following operations: Based on the evaluation results, multiple training parameters are dynamically adjusted, including at least: stimulation frequency, intensity, and duration; Perform the following operations through the task publishing module: By calling the blockchain task management contract, new training tasks and new training parameters are published; The feedback incentive module performs the following operations: The node response quality is evaluated based on the node response data to obtain an evaluation result; and corresponding on-chain incentive tokens are allocated based on the evaluation result.

2. The training method according to claim 1, characterized in that, Before generating the training task through the task management contract of the blockchain, the method further includes: The task management contract and the data storage contract are deployed on the blockchain, wherein the task management contract is used to publish and manage the corresponding training tasks, and the data storage contract is used to store the training results of the multi-neuron network nodes.

3. The training method according to claim 1, characterized in that, Any one of the plurality of biological neuron network nodes includes: a local controller, a stimulation controller, a microelectrode array, and a neuron culture unit; the method further includes: The following operations are performed sequentially via the local controller, the stimulation controller, the microelectrode array, and the neuron culture unit: Perform the following operations through the local controller: In response to receiving the training task, the electrical stimulation command control logic is triggered. Receive the training data returned by the microelectrode array; The training data, signal features, and training summary are integrated to generate response data; The response data is uploaded to the blockchain; The following operations are performed via the stimulation controller: Based on the received training task, a preset current stimulation signal and a preset voltage stimulation signal are generated; and voltage pulses and current pulses are output to the microelectrode array. The following operations are performed using the microelectrode array: Receive stimulation signals from the stimulation controller and apply electrical stimulation to nerve tissue; The training data of the neural network is acquired in real time, and the training data includes: discharge mode and reaction timing; The following operations are performed using the neuron culture unit: Receive electrical stimulation from a microelectrode array and perform training based on the electrical stimulation.

4. The training method according to claim 1, characterized in that, Before generating the training task through the task management contract of the blockchain, the method further includes: A training architecture for collaborative training of multi-neuronal networks based on blockchain is constructed. The training architecture includes: multiple biological neuron network nodes, a blockchain capable of generating tokens, and a training coordinator. Each of the multiple biological neuron network nodes is configured with a corresponding stimulus generator, an EEG monitoring interface, and a local controller.

5. The training method according to claim 1, characterized in that, After dynamically adjusting the training task parameters and stimulus scheme based on the evaluation results and aggregation results, the method further includes: The response results of the multiple biological neuron network nodes are evaluated to obtain target response nodes that meet preset quality requirements; and corresponding on-chain incentive tokens are allocated to the target response nodes.

6. The training method according to claim 1, characterized in that, The method further includes: Initiate training in response to the training strategy; After each round of training, each biological neuron network node submits a corresponding training summary; the training summary includes at least: the unique identifier of the current node, the unique identifier of the current task, the neural stimulation pattern determined by the type of stimulation sequence in this round, the delay from stimulation to the first neural response, the average firing frequency per unit time, the entropy value of the change in neuronal synaptic weight distribution, the timestamp, the hash calculated from the potential waveform or encoded sequence, and the signature generated in the trusted execution environment. Based on the training summary, the training coordinator sequentially performs the following steps: filtering valid data; filtering each biological neuron network node to obtain activated nodes; and generating the training strategy for the next round. Filter valid data by performing the following steps: Signature verification is performed by verifying the match between the signature and the node's identity. Check whether the current task is within its validity period based on the time window; Verify that the task matches the task identifier; The next training strategy is generated by performing the following operations in sequence: The stimulation parameters are optimized, the incentive nodes are selected, and the training strategy for the next round is encoded and published by calling the task management contract; Specifically, the stimulation parameters are optimized by performing the following operations: based on the high-scoring nodes of the previous round, effective stimulation types and stimulation parameters are extracted to obtain effective stimulation types and stimulation parameters, so as to generate the stimulation sequence for the next round based on the stimulation parameters.

7. A training device based on a blockchain training architecture, characterized in that, The device includes: The blockchain task management contract in the training architecture is used to generate training tasks, which include training task parameters, including stimulus types and input sequences. The training architecture includes multiple biological neuron network nodes, which are used to perform neural training based on the received training task to generate node response results for multiple biological neuron network nodes. The training coordinator in the training architecture is used to dynamically adjust the training task parameters and stimulus schemes based on the evaluation results and aggregation results. The training device based on the blockchain training architecture may further include: The training coordinator includes: a data collection module, a response aggregation module, a training evaluation module, a policy generation module, a task publishing module, and a feedback and incentive module; The second execution module is used to perform the following operations sequentially through the data collection module, response aggregation module, training evaluation module, policy generation module, task deployment module, and feedback incentive module: Perform the following operations through the data collection module: Listen to blockchain data storage contract events and read node response data uploaded from multiple biological neural network nodes in the blockchain; Perform the following actions by responding to the aggregation module: The node response data is aggregated and analyzed according to a preset algorithm to obtain the aggregation result; and global training features are extracted based on the aggregation result. Perform the following operations using the training evaluation module: The current training state is evaluated based on the aggregation results, and the evaluation results are obtained. The current training state includes: convergence and performance indicators. Perform the following operations through the policy generation module: Based on the evaluation results, multiple training parameters are dynamically adjusted. These multiple training parameters include at least: stimulation frequency, intensity, and duration. Perform the following operations through the task publishing module: By calling the blockchain task management contract, new training tasks and new training parameters are published; Perform the following operations through the feedback incentive module: The quality of node responses is evaluated based on node response data to obtain evaluation results; and corresponding on-chain incentive tokens are allocated based on the evaluation results.

8. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed in a computer, causes the computer to perform the method described in any one of claims 1 to 6.

9. An electronic device, characterized in that, The method includes a memory and a processor, wherein the memory stores executable code, and the processor executes the executable code to implement the method according to any one of claims 1 to 6.

Images