Distributed artificial intelligence model generation system and method based on intelligent wearable device

Abstract

The invention discloses a distributed artificial intelligence model generation system and method based on intelligent wearable equipment, relates to the technical field of block chains, and realizes distributed development of an artificial intelligence model related to the intelligent wearable equipment. The system comprises an intelligent wearable device which is used for obtaining a current training round, executing preset condition detection before training is started, and if the preset condition detection passes, sending application participation information; the aggregation node is used for obtaining participatable intelligent wearable equipment according to the current task type and the participation application information; the intelligent wearable device receiving the participation instruction is used for performing parameter updating on the global model by using the preprocessing features to obtain model parameters; the aggregation node is used for aggregating the model parameters to obtain global model parameters; the block chain node is used for receiving global model parameters of all training rounds, and evaluating to obtain to-be-aggregated global model parameters; and the aggregation node is used for aggregating the to-be-aggregated global model parameters to obtain target global model parameters and sending the target global model parameters to all the intelligent wearable devices.

Description

Technical Field

[0001] This invention relates to the field of blockchain technology, and in particular to a distributed artificial intelligence model generation system and method based on smart wearable devices. Background Technology

[0002] With advancements in hardware, smartwatches and other wearable devices have acquired basic AI inference capabilities. Since they are worn close to the body year-round, they collect sensitive data such as real-time health and activity levels. Compared to traditional cloud-based systems, local inference significantly reduces latency and bandwidth consumption, while protecting user privacy in scenarios like medical monitoring and activity feedback. This "on-site access + instant response" capability is crucial for user experience. Centralized development and construction of AI models related to wearable devices can no longer meet user needs.

[0003] Therefore, it is necessary to use blockchain technology to generate distributed artificial intelligence models based on smart wearable devices. Summary of the Invention

[0004] In view of this, the present invention provides a distributed artificial intelligence model generation system and method based on smart wearable devices, realizing the distributed development of artificial intelligence models related to smart wearable devices.

[0005] According to a first aspect of the present invention, a distributed artificial intelligence model generation system based on smart wearable devices is provided, the system comprising: a preset number of smart wearable devices, aggregation nodes, and blockchain nodes, wherein the preset number is greater than 1;

[0006] The smart wearable device is used to obtain the current training round. Before the current training round begins, it performs a preset condition detection. If the preset condition is met, it sends a participation application to the aggregation node.

[0007] The aggregation node is used to receive the participation application information sent by the smart wearable device, determine the smart wearable devices that can participate based on the current task type and the participation application information, and send a participation instruction to the smart wearable devices that can participate.

[0008] The smart wearable device is also used to receive the participation instruction, update the parameters of the global model using local preprocessing features, obtain model parameters, and send them to the aggregation node;

[0009] The aggregation node is also used to aggregate global model parameters based on the model parameters and send them to the blockchain node;

[0010] The blockchain node is used to receive the global model parameters corresponding to all training rounds, evaluate all the global model parameters, obtain the global model parameters to be aggregated that have passed the evaluation, and send the global model parameters to be aggregated to the aggregation node.

[0011] The aggregation node is also used to aggregate target global model parameters based on the global model parameters to be aggregated, and send the target global model parameters to the smart wearable device.

[0012] Preferably, the smart wearable device is also used to collect local data, perform noise filtering and cleaning, align timestamps and interpolate, extract features and reduce dimensionality to obtain local preprocessed features.

[0013] Preferably, the smart wearable device is specifically used to perform power detection, network quality detection, and computing power detection.

[0014] Preferably, the aggregation node is specifically used to score each item based on the participation application information, assign different preset weights to each item according to the current task type, calculate the total score, calculate the participation score based on the historical reliability score of the eligible smart wearable device and the total score, and obtain the eligible smart wearable device based on the participation score, wherein the participation application information includes the value corresponding to each preset condition.

[0015] Preferably, the aggregation node is specifically used to calculate the proportion of the amount of data used for training by each of the smart wearable devices to the total amount of data used for training by all the smart wearable devices, calculate the product of each proportion and the corresponding model parameter, and add all the products to obtain the global model parameter.

[0016] According to a second aspect of the present invention, a method for generating a distributed artificial intelligence model based on a smart wearable device is provided, the method comprising:

[0017] Obtain the current training round. Before the current training round begins, perform a preset condition check. If the preset condition is met, send a participation application to the aggregation node.

[0018] Receive the participation application information sent by all smart wearable devices, determine the smart wearable devices that can participate based on the current task type and the participation application information, and send a participation instruction to the smart wearable devices that can participate, wherein the number of smart wearable devices is a preset number;

[0019] Upon receiving the participation instruction, the global model parameters are updated using local preprocessed features to obtain model parameters, which are then sent to the aggregation node.

[0020] Global model parameters are obtained by aggregating the model parameters and then sent to the blockchain node.

[0021] Receive the global model parameters corresponding to all training rounds, evaluate all the global model parameters, obtain the global model parameters to be aggregated that have passed the evaluation, and send the global model parameters to be aggregated to the aggregation node;

[0022] Based on the global model parameters to be aggregated, the target global model parameters are aggregated and sent to the smart wearable device.

[0023] Preferably, before updating the parameters of the global model using local preprocessed features to obtain the model parameters, the method further includes:

[0024] Collect local data;

[0025] The local data is subjected to noise filtering and cleaning, timestamp alignment and interpolation, feature extraction and dimensionality reduction to obtain local preprocessed features.

[0026] Preferably, the preset condition detection includes:

[0027] Perform power detection, network quality detection, and computing power detection on the smart wearable device.

[0028] Preferably, the step of obtaining eligible smart wearable devices based on the current task type and the participation application information includes:

[0029] Individual scores are assigned based on the application information, wherein the application information includes the value corresponding to each preset condition;

[0030] Assign different preset weights to each item based on the current task type, and calculate the total score;

[0031] A participation score is calculated based on the historical reliability score of the eligible smart wearable device and the total score, and eligible smart wearable devices are obtained based on the participation score.

[0032] Preferably, the step of aggregating the model parameters to obtain the global model parameters includes:

[0033] Calculate the proportion of the amount of data used for training by each of the aforementioned smart wearable devices to the total amount of data used for training by all the aforementioned smart wearable devices.

[0034] Calculate the product of each of the stated proportions and the corresponding model parameters, and sum all the products to obtain the global model parameters.

[0035] By means of the above technical solution, the present invention provides a distributed artificial intelligence model generation system and method based on smart wearable devices. The system includes: a preset number of smart wearable devices, an aggregation node, and a blockchain node, wherein the preset number is greater than 1. The smart wearable devices are used to obtain the current training round, and before the start of the current training round, perform a preset condition check. If the preset condition is met, they send participation application information to the aggregation node. The aggregation node is used to receive the participation application information sent by the smart wearable devices, determine the eligible smart wearable devices based on the current task type and the participation application information, and send participation instructions to the eligible smart wearable devices. The wearable device is also used to receive the participation instruction, update the parameters of the global model using local preprocessing features, obtain model parameters, and send them to the aggregation node; the aggregation node is also used to aggregate global model parameters based on the model parameters and send them to the blockchain node; the blockchain node is used to receive the global model parameters corresponding to all training rounds, evaluate all the global model parameters, obtain global model parameters to be aggregated that have passed the evaluation, and send the global model parameters to be aggregated to the aggregation node; the aggregation node is also used to aggregate target global model parameters based on the global model parameters to be aggregated, and send the target global model parameters to the smart wearable device. Through the technical solution of this invention, on the one hand, multiple smart wearable devices use data in a distributed manner for local training, and then aggregate the model parameters obtained from their respective training. This allows the data to "remain on the device," with only the model parameters uploaded, thereby protecting data privacy. On the other hand, combined with the consensus mechanism of blockchain, it ensures that training contributions are verifiable and rewards are traceable, forming a decentralized, secure, transparent network that incentivizes continuous participation from edge nodes. Finally, multiple smart wearable devices undergo two screening processes to ensure the accuracy and efficiency of AI generation. The first screening is performed locally by the smart wearable devices; only those that pass the screening (i.e., those that pass the preset condition detection) can send participation application information. The second screening is performed by the aggregation node on all smart wearable devices that have sent participation application information, resulting in the smart wearable devices that can participate. This architecture reduces communication and energy consumption, ensures efficiency, meets real-time and privacy requirements, and promotes distributed intelligence to a wider device ecosystem.

[0036] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0037] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of this application. In the drawings:

[0038] Figure 1 This diagram illustrates the structure of a distributed artificial intelligence model generation system based on a smart wearable device, according to an embodiment of the present invention.

[0039] Figure 2 The illustration shows a flowchart of a distributed artificial intelligence model generation method based on a smart wearable device provided by an embodiment of the present invention;

[0040] Figure 3 This diagram illustrates a distributed artificial intelligence model generation device based on a smart wearable device, according to an embodiment of the present invention.

[0041] Figure 4 This diagram illustrates another distributed artificial intelligence model generation device based on a smart wearable device provided in an embodiment of the present invention. Detailed Implementation

[0042] The present invention will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0043] This embodiment provides a distributed artificial intelligence model generation system based on smart wearable devices, such as... Figure 1As shown, the system includes: a preset number of smart wearable devices, an aggregation node, and a blockchain node, wherein the preset number is greater than 1; the smart wearable devices are used to obtain the current training round, and before the start of the current training round, perform a preset condition check. If the preset condition is met, they send participation application information to the aggregation node; the aggregation node is used to receive the participation application information sent by the smart wearable devices, determine the eligible smart wearable devices based on the current task type and the participation application information, and send participation instructions to the eligible smart wearable devices; the smart wearable devices are also used to receive the participation instructions and utilize this... The preprocessing features of the local model are used to update the parameters of the global model, and the model parameters are obtained and sent to the aggregation node. The aggregation node is also used to aggregate the global model parameters according to the model parameters and send them to the blockchain node. The blockchain node is used to receive the global model parameters corresponding to all training rounds, evaluate all the global model parameters, obtain the global model parameters to be aggregated that have passed the evaluation, and send the global model parameters to be aggregated to the aggregation node. The aggregation node is also used to aggregate the target global model parameters according to the global model parameters to be aggregated and send the target global model parameters to the smart wearable device.

[0044] As an application scenario of distributed artificial intelligence models based on smart wearable devices, advanced sleep staging goes beyond basic sleep / wake detection, utilizing accelerometer, heart rate, heart rate variability (HRV), and respiratory rate variability (RRV) data to classify non-rapid eye movement (NREM) sleep and rapid eye movement (REM) sleep. Through federated learning models trained on numerous users, highly accurate personalized sleep models can be created.

[0045] As another application scenario for distributed artificial intelligence models based on smart wearable devices, aggregated (and anonymous) insights from health sub-networks (e.g., a widespread increase in fever-like body temperature readings, which can be detected by integrating audio processing, and abnormal fatigue patterns) can provide early signals for public health surveillance without revealing personal identities.

[0046] As another application scenario of distributed artificial intelligence models based on smart wearable devices, smartwatches can contribute to models that optimize smart home / office environments to improve comfort, productivity, or energy efficiency by sensing local environmental conditions (noise, light, temperature sensed by agents or future sensors) and user contexts (activity, stress).

[0047] Smart wearable devices can be smartwatches, smart glasses, smart bracelets, or smart rings. Distributed artificial intelligence models can be models for motion recognition or models for heart rate abnormality detection; there are no restrictions here.

[0048] In this embodiment, to reduce communication and energy consumption and improve efficiency, a screening process is performed on the local smart wearable device. Only smart wearable devices that pass the preset condition detection can apply to the aggregation node to participate in training by sending participation application information. After the aggregation node receives the participation application information sent by the smart wearable device, it performs another screening to obtain the smart wearable devices that can participate. Only the smart wearable devices that can participate can receive the participation instruction. Then, the smart wearable devices that can participate first obtain a global model snapshot from the aggregation node, that is, download the global model, and then use the local preprocessed features to perform local training, monitoring power consumption and calculation progress in real time. If the connection is lost or the power drops sharply, the training is interrupted and the status is reported.

[0049] When a blockchain node performs an evaluation, it can send global model parameters to the subnet. The subnet then evaluates the parameters according to the consensus mechanism, performs on-chain verification, and incorporates them into the global model. Specifically:

[0050] Submitting and Hiding Scores: Each participating blockchain node scores all submitted global model parameters, forming a set of "weight vectors." To prevent plagiarism of others' scores, validators first encrypt the score data before publishing it on the chain, but the scores themselves are temporarily invisible.

[0051] Score Disclosure: After a preset block period, validators publicly reveal the decrypted score and random salt value. This is then verified on-chain to match the previously encrypted content, thus validating the update's credibility. At this point, the valid score matrix is ​​complete and available for subsequent consensus. According to predetermined rules, participating smart wearable devices with scores exceeding a preset threshold (e.g., top 30%) will receive economic incentives (such as tokens), enhancing incentive drive. These incentives are recorded on the blockchain or awarded through tokens, with the consensus mechanism determining the reward distribution.

[0052] After consensus is reached, the evaluated global model parameters to be aggregated are broadcast through the chain, updated periodically, and sent to the aggregation node. The aggregation node then aggregates the parameters again and sends them to the smart wearable device, achieving a closed loop of synchronization between the global and local models.

[0053] Preferably, the smart wearable device is also used to collect local data, perform noise filtering and cleaning, align timestamps and interpolate, extract features and reduce dimensionality to obtain local preprocessed features.

[0054] In this embodiment, the smart wearable device, acting as an edge node in the distributed AI network, can collect local data. This local data includes sensor data such as heart rate, acceleration, and temperature. Preprocessing can also be performed locally. Preprocessing includes:

[0055] (1) Noise filtering and cleaning: Use algorithms such as low-pass filters or moving averages to remove outliers in local data and ensure data quality.

[0056] (2) Aligning timestamps and interpolation: Synchronize different local data streams and use interpolation methods to fill in short-term lost data to prevent misjudgment during subsequent model training.

[0057] (3) Feature extraction and dimensionality reduction: For example, the fast Fourier transform can be used to extract frequency domain features of acceleration, or the sliding window can be used to statistically analyze the rate of change of heart rate. Principal component analysis can be used to further reduce the input dimension and reduce the computational load of the back-end model.

[0058] On the local device side of the smart wearable device, the preprocessing process of cleaning, synchronization, and feature extraction is completed, generating structured lightweight input and providing a reliable and efficient data foundation.

[0059] Preferably, the smart wearable device is specifically used to perform power detection, network quality detection, and computing power detection.

[0060] In this embodiment, the preset condition detection includes power consumption detection, network quality detection, and computing power detection. Only when all three are passed can the preset condition detection be considered passed. The preset condition detection may also include other detections, which are not limited here.

[0061] Among them, the power detection function reads the remaining battery power of the smart wearable device. If it is lower than a preset threshold (such as 10%), the current training round is automatically terminated. In other words, power detection indicates that the battery is fully charged.

[0062] Network quality check: Evaluates the bandwidth and latency of the current connection (Wi-Fi / Bluetooth / Cellular). If data transmission cannot be completed or the communication time limit cannot be met, the current training round is automatically terminated. The size of the last uploaded file is divided by the current bandwidth. If the communication latency exceeds a first preset threshold (e.g., 1 second), the current training round is automatically terminated. In other words, passing the network quality check indicates a good connection.

[0063] Computational power test: Estimates the time required for local model training or inference, taking into account the available performance of the device chip and temperature status, to confirm that it can be completed within a preset time. (Calculated by dividing the total floating-point operations (FLOPS) of the previous model calculation by the available floating-point performance of the device (FLOPS / second)). If the time exceeds a second preset threshold (e.g., 100 milliseconds), the current training round is automatically terminated. In other words, the computational power test indicates that the computing power is sufficient.

[0064] Preferably, the aggregation node is specifically used to score each item based on the participation application information, assign different preset weights to each item according to the current task type, calculate the total score, calculate the participation score based on the historical reliability score of the eligible smart wearable device and the total score, and obtain the eligible smart wearable device based on the participation score, wherein the participation application information includes the value corresponding to each preset condition.

[0065] In this embodiment, the aggregation node can be a mobile phone or a home gateway. The application information includes the value corresponding to each preset condition. For example, the preset conditions are battery level, network quality (reflected in network bandwidth), and computing power (reflected in response latency). Accordingly, the application information includes battery level, network bandwidth, and response latency values. Taking three items as an example:

[0066] First, when scoring each item, convert all three items to a score from 0 to 100. The scoring rules for battery level are: 100% battery equals 100 points, and the lower the battery level, the lower the score; below 20% is 0 points. The scoring rules for network bandwidth are: the higher the bandwidth, the higher the score; reaching a cap (e.g., 100Mbps) equals 100 points. The scoring rules for response latency are: the lower the latency, the higher the score; below 50ms might be 100 points, and above 500ms is 0 points.

[0067] Then, based on the current task type, different preset weights are assigned. For example, for intensive tasks, "battery consumption" has the highest weight; for big data transmission tasks, "network bandwidth" has the highest weight; and for real-time interactive tasks, "response latency" has the highest weight. The total score is calculated using individual item scores and their corresponding weights.

[0068] Finally, the historical reliability score (a coefficient between 0 and 1) of the smart wearable device is multiplied by the weighted total score to obtain the final participation score. A smart wearable device with a stable historical performance that never drops the connection will have a historical reliability score close to 1, while a smart wearable device that frequently drops the connection will have a score close to 0.

[0069] Smart wearable devices with participation scores higher than the preset participation score will be selected as eligible smart wearable devices. Smart wearable devices that are not selected will be re-evaluated during off-peak hours in the region.

[0070] Preferably, the aggregation node is specifically used to calculate the proportion of the amount of data used for training by each of the smart wearable devices to the total amount of data used for training by all the smart wearable devices, calculate the product of each proportion and the corresponding model parameter, and add all the products to obtain the global model parameter.

[0071] In this embodiment, after the current training round ends, each participating smart wearable device uploads model parameters to the aggregation node. The model parameters include model gradients or model weights. Taking model weights as an example, the generation of model weight updates adopts sparse and structured updates to reduce communication overhead. For example, only important layers or threshold masks are uploaded. The aggregation node can also re-evaluate the communication success rate and discard invalid or incomplete model parameters, which is not limited here.

[0072] The aggregation node aggregates various model parameters. For example, if there are k smart wearable devices, and the global model parameters are used as global model weights, then the corresponding model parameters are used as model weights. When there are multiple global model parameters, each one is calculated using the following method:

[0073] Global model weights

[0074] Among them, w k It is the model weight of the kth eligible smart wearable device, n k N is the data volume of the k-th preprocessed feature that can participate in the preprocessing of smart wearable devices, and N is the total data volume of all preprocessed features that can participate in the preprocessing of smart wearable devices. It's a ratio.

[0075] Furthermore, a method for generating distributed artificial intelligence models based on smart wearable devices is provided, such as... Figure 2 As shown, the method includes:

[0076] 101. Obtain the current training round. Before the current training round begins, perform a preset condition check. If the preset condition is met, send a participation application to the aggregation node.

[0077] The preset condition detection includes: performing power detection, network quality detection, and computing power detection of the smart wearable device.

[0078] 102. Receive the participation application information sent by all smart wearable devices, determine the smart wearable devices that can participate based on the current task type and the participation application information, and send a participation instruction to the smart wearable devices that can participate, wherein the number of smart wearable devices is a preset number.

[0079] The step of obtaining eligible smart wearable devices based on the current task type and the participation application information includes: scoring each item based on the participation application information, wherein the participation application information includes the value corresponding to each preset condition; assigning different preset weights to each item based on the current task type and calculating the total score; calculating the participation score based on the historical reliability score of the eligible smart wearable devices and the total score; and obtaining the eligible smart wearable devices based on the participation score.

[0080] 103. Receive the participation instruction, update the parameters of the global model using the local preprocessing features, obtain the model parameters, and send them to the aggregation node.

[0081] Before updating the parameters of the global model using local preprocessed features to obtain the model parameters, the method further includes: collecting local data; performing noise filtering and cleaning, aligning timestamps and interpolating, extracting features and reducing dimensionality on the local data to obtain local preprocessed features.

[0082] 104. The global model parameters are obtained by aggregating the model parameters and sent to the blockchain node.

[0083] The step of aggregating the model parameters to obtain global model parameters includes: calculating the proportion of the amount of data used for training by each of the smart wearable devices to the total amount of data used for training by all the smart wearable devices; calculating the product of each proportion and the corresponding model parameter; and summing all the products to obtain global model parameters.

[0084] 105. Receive the global model parameters corresponding to all training rounds, evaluate all the global model parameters, obtain the global model parameters to be aggregated that have passed the evaluation, and send the global model parameters to be aggregated to the aggregation node.

[0085] 106. Based on the global model parameters to be aggregated, aggregate to obtain target global model parameters, and send the target global model parameters to the smart wearable device.

[0086] Through steps 101-106 of the implementation example, a decentralized AI network is built using blockchain technology, allowing anyone to participate in model training and verification and receive token rewards for their contributions. This model encourages open, transparent, and fair cooperation, and promotes the return of AI from the hands of a few giants to the community.

[0087] The validation process involves using locally preprocessed features for lightweight model inference. Specifically, preprocessed features are input into a lightweight model (e.g., a Tiny Machine Learning (TinyML) level CNN / RNN / Transformer, less than a few hundred KB, used for fall detection, abnormal heart rate classification, etc.). This eliminates the need to upload raw data to the cloud or blockchain, significantly reducing latency and bandwidth consumption while protecting privacy. Whether for training or validation, preprocessed features can be input into the lightweight model for updates, and the resulting model parameters can then be used as a basis for updating the global model's parameters. Preferably, a micro AI inference framework (such as TensorFlow Lite or ONNX Runtime) is used to generate immediate alerts or output anomaly messages upon detecting anomalies.

[0088] This invention provides a distributed artificial intelligence model generation system and method based on smart wearable devices. Through this technical solution, on the one hand, multiple smart wearable devices use data in a distributed manner for local training, and then aggregate the model parameters obtained from their respective training. This allows data to "remain on the device," with only model parameters uploaded, thus protecting data privacy. On the other hand, by combining the consensus mechanism of blockchain, it ensures that training contributions are verifiable and rewards are traceable, forming a decentralized, secure, transparent network that incentivizes continuous participation from edge nodes. Finally, multiple smart wearable devices undergo two screening processes to ensure the accuracy and efficiency of AI generation. The first screening is performed locally by the smart wearable devices; only those that pass the screening (i.e., those detected by preset conditions) can send participation application information. The second screening is performed by the aggregation node on all smart wearable devices that have sent participation application information, resulting in a list of eligible smart wearable devices. This architecture reduces communication and energy consumption, ensures efficiency, meets real-time and privacy requirements, and promotes distributed intelligence to a wider device ecosystem.

[0089] Furthermore, as Figure 2 The specific implementation of the method shown in this invention provides a distributed artificial intelligence model generation device based on a smart wearable device, such as... Figure 3 As shown, the device includes: an acquisition module 31, a receiving module 32, an update module 33, an aggregation module 34, an evaluation module 35, and a sending module 36;

[0090] The acquisition module 31 is used to acquire the current training round. Before the current training round begins, it performs a preset condition detection. If the preset condition is met, it sends a participation application information to the aggregation node.

[0091] The receiving module 32 is used to receive the participation application information sent by all smart wearable devices, determine the smart wearable devices that can participate based on the current task type and the participation application information, and send a participation instruction to the smart wearable devices that can participate, wherein the number of smart wearable devices is a preset number;

[0092] The update module 33 is used to receive the participation instruction, update the parameters of the global model using local preprocessed features, obtain model parameters, and send them to the aggregation node;

[0093] Aggregation module 34 is used to aggregate global model parameters based on the model parameters and send them to the blockchain node;

[0094] Evaluation module 35 is used to receive the global model parameters corresponding to all training rounds, evaluate all the global model parameters, obtain the global model parameters to be aggregated that have passed the evaluation, and send the global model parameters to be aggregated to the aggregation node.

[0095] The sending module 36 is used to aggregate the target global model parameters according to the global model parameters to be aggregated, and send the target global model parameters to the smart wearable device.

[0096] In specific application scenarios, a distributed artificial intelligence model generation module based on smart wearable devices, such as... Figure 4 As shown, the device also includes a preprocessing module 37, which is specifically used to collect local data; perform noise filtering and cleaning, align timestamps and interpolate, extract features and reduce dimensions on the local data to obtain local preprocessed features.

[0097] Accordingly, in order to perform preset condition detection, the acquisition module 31 is specifically used to perform power detection, network quality detection and computing power detection of the smart wearable device.

[0098] Accordingly, in order to obtain the eligible smart wearable devices based on the current task type and the participation application information, the receiving module 32 is specifically used to perform individual scoring based on the participation application information, wherein the participation application information includes the value corresponding to each of the preset conditions; assign different preset weights to each item according to the current task type, calculate the total score; calculate the participation score based on the historical reliability score of the eligible smart wearable devices and the total score, and obtain the eligible smart wearable devices based on the participation score.

[0099] Accordingly, in order to aggregate the model parameters to obtain global model parameters, the aggregation module 34 is specifically used to calculate the proportion of the amount of data used for training by each of the smart wearable devices to the total amount of data used for training by all the smart wearable devices; calculate the product of each proportion and the corresponding model parameter; and add all the products to obtain the global model parameters.

[0100] It should be noted that other corresponding descriptions of the functional units involved in the distributed artificial intelligence model generation device based on smart wearable devices provided in this embodiment can be found in [reference]. Figure 2 The corresponding description will not be repeated here.

[0101] Based on the above, Figure 2 Accordingly, this embodiment also provides a storage medium, which may be volatile or non-volatile, storing a computer program that, when executed by a processor, implements the above-described method. Figure 2 The method for generating distributed artificial intelligence models based on smart wearable devices is shown.

[0102] Based on this understanding, the technical solution of the present invention can be embodied in the form of a software product, which can be stored in a storage medium (such as a CD-ROM, USB flash drive, portable hard drive, etc.) and includes several instructions to cause a computer device (such as a personal computer, server, or network device, etc.) to execute the methods of various implementation scenarios of the present invention.

[0103] Based on the above, Figure 2 The method shown and Figure 3 , Figure 4 To achieve the above objectives, this embodiment also provides a computer device, which includes a storage medium and a processor; the storage medium is used to store a computer program; the processor is used to execute the computer program to implement the above-described virtual device embodiment. Figure 2 The method for generating distributed artificial intelligence models based on smart wearable devices is shown.

[0104] Optionally, the computer device may also include a user interface, a network interface, a camera, radio frequency (RF) circuitry, sensors, audio circuitry, a Wi-Fi module, etc. The user interface may include a display screen, input units such as a keyboard, etc., and optional user interfaces may also include USB interfaces, card reader interfaces, etc. The network interface may optionally include standard wired interfaces, wireless interfaces (such as Wi-Fi interfaces), etc.

[0105] Those skilled in the art will understand that the computer device structure provided in this embodiment does not constitute a limitation on the physical device, and may include more or fewer components, or combine certain components, or have different component arrangements.

[0106] The storage medium may also include an operating network communication module. An operating system is a program that manages the hardware and software resources of the aforementioned computer device, supporting the operation of information processing programs and other software and / or programs. The network communication module is used for communication between the various components within the storage medium, as well as for communication with other hardware and software within the information processing device.

[0107] Through the above description of the embodiments, those skilled in the art can clearly understand that the present invention can be implemented by means of software plus necessary general-purpose hardware platform, or it can be implemented by hardware.

[0108] This invention provides a distributed artificial intelligence model generation system and method based on smart wearable devices. Through this technical solution, on the one hand, multiple smart wearable devices use data in a distributed manner for local training, and then aggregate the model parameters obtained from their respective training. This allows data to "remain on the device," with only model parameters uploaded, thus protecting data privacy. On the other hand, by combining the consensus mechanism of blockchain, it ensures that training contributions are verifiable and rewards are traceable, forming a decentralized, secure, transparent network that incentivizes continuous participation from edge nodes. Finally, multiple smart wearable devices undergo two screening processes to ensure the accuracy and efficiency of AI generation. The first screening is performed locally by the smart wearable devices; only those that pass the screening (i.e., those detected by preset conditions) can send participation application information. The second screening is performed by the aggregation node on all smart wearable devices that have sent participation application information, resulting in a list of eligible smart wearable devices. This architecture reduces communication and energy consumption, ensures efficiency, meets real-time and privacy requirements, and promotes distributed intelligence to a wider device ecosystem.

[0109] Those skilled in the art will understand that the accompanying drawings are merely schematic diagrams of a preferred embodiment, and the modules or processes shown in the drawings are not necessarily essential for implementing the present invention. Those skilled in the art will understand that the modules in the apparatus of the embodiment can be distributed within the apparatus of the embodiment as described, or they can be located in one or more apparatuses different from this embodiment, with corresponding changes. The modules of the above-described embodiment can be combined into one module, or further divided into multiple sub-modules.

[0110] The serial numbers used above are for descriptive purposes only and do not represent the superiority or inferiority of the implementation scenarios. The above disclosures are merely a few specific implementation scenarios of the present invention; however, the present invention is not limited thereto, and any variations conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A distributed artificial intelligence model generation system based on smart wearable devices, characterized in that, The system includes: a preset number of smart wearable devices, aggregation nodes, and blockchain nodes, wherein the preset number is greater than 1; The smart wearable device is used to obtain the current training round. Before the current training round begins, it performs a preset condition detection. If the preset condition is met, it sends a participation application to the aggregation node. The aggregation node is used to receive the participation application information sent by the smart wearable device, determine the smart wearable devices that can participate based on the current task type and the participation application information, and send a participation instruction to the smart wearable devices that can participate. The smart wearable device is also used to receive the participation instruction, update the parameters of the global model using local preprocessing features, obtain model parameters, and send them to the aggregation node; The aggregation node is also used to aggregate global model parameters based on the model parameters and send them to the blockchain node; The blockchain node is used to receive the global model parameters corresponding to all training rounds, evaluate all the global model parameters, obtain the global model parameters to be aggregated that have passed the evaluation, and send the global model parameters to be aggregated to the aggregation node. The aggregation node is also used to aggregate target global model parameters based on the global model parameters to be aggregated, and send the target global model parameters to the smart wearable device.

2. The system according to claim 1, characterized in that, The smart wearable device is also used to collect local data, perform noise filtering and cleaning, align timestamps and interpolate, extract features and reduce dimensions to obtain local preprocessed features.

3. The system according to claim 1, characterized in that, The smart wearable device is specifically used to perform power detection, network quality detection, and computing power detection.

4. The system according to claim 1, characterized in that, The aggregation node is specifically used to score each item based on the participation application information, assign different preset weights to each item according to the current task type, calculate the total score, calculate the participation score based on the historical reliability score of the eligible smart wearable device and the total score, and obtain the eligible smart wearable device based on the participation score. The participation application information includes the value corresponding to each preset condition.

5. The system according to claim 1, characterized in that, The aggregation node is specifically used to calculate the proportion of the amount of data used for training by each of the smart wearable devices to the total amount of data used for training by all the smart wearable devices, calculate the product of each proportion and the corresponding model parameter, and add all the products to obtain the global model parameter.

6. A method for generating distributed artificial intelligence models based on smart wearable devices, characterized in that, The method includes: Obtain the current training round. Before the current training round begins, perform a preset condition check. If the preset condition is met, send a participation application to the aggregation node. Receive the participation application information sent by all smart wearable devices, determine the smart wearable devices that can participate based on the current task type and the participation application information, and send a participation instruction to the smart wearable devices that can participate, wherein the number of smart wearable devices is a preset number; Upon receiving the participation instruction, the global model parameters are updated using local preprocessed features to obtain model parameters, which are then sent to the aggregation node. Global model parameters are obtained by aggregating the model parameters and then sent to the blockchain node. Receive the global model parameters corresponding to all training rounds, evaluate all the global model parameters, obtain the global model parameters to be aggregated that have passed the evaluation, and send the global model parameters to be aggregated to the aggregation node; Based on the global model parameters to be aggregated, the target global model parameters are aggregated and sent to the smart wearable device.

7. The method according to claim 6, characterized in that, Before updating the global model parameters using local preprocessed features to obtain the model parameters, the method further includes: Collect local data; The local data is subjected to noise filtering and cleaning, timestamp alignment and interpolation, feature extraction and dimensionality reduction to obtain local preprocessed features.

8. The method according to claim 6, characterized in that, The execution of preset condition detection includes: Perform power detection, network quality detection, and computing power detection on the smart wearable device.

9. The method according to claim 6, characterized in that, The process of determining eligible smart wearable devices based on the current task type and the participation application information includes: Individual scores are assigned based on the application information, wherein the application information includes the value corresponding to each preset condition; Assign different preset weights to each item based on the current task type, and calculate the total score; A participation score is calculated based on the historical reliability score of the eligible smart wearable device and the total score, and eligible smart wearable devices are obtained based on the participation score.

10. The method according to claim 6, characterized in that, The process of aggregating global model parameters based on the model parameters includes: Calculate the proportion of the amount of data used for training by each of the aforementioned smart wearable devices to the total amount of data used for training by all the aforementioned smart wearable devices. Calculate the product of each of the stated proportions and the corresponding model parameters, and sum all the products to obtain the global model parameters.

Images