751 results about "Federated learning" patented technology

Aspects of the disclosure provide a method performed by a central Network Data Analytics Function (NWDAF) in a communications network. The communications network comprises one or more local NWDAFs configured to develop a model using federated learning, in which each local NWDAF stores a copy of the model and trains the copy of the model by inputting training data into a machine-learning process. The method comprises receiving, from the one or more local NWDAFs, a respective local model update comprising an update to values of one or more parameters of the model generated by training a respective copy of the model using machine-learning. The method further comprises combining the local model updates received from the one or more local NWDAFs to obtain a combined model update.

Aspects described herein relate to signals transmitted in federated learning (FL) of a neural network (NN) using wireless communications. A transmitting device can generate an analog modulation signal to indicate an updated value for one or more coefficients for a federated learning (FL) procedure for a neural network, and the transmitting device can transmit the analog modulation signal over time and frequency resources configured for a UE group of multiple UE groups, where each UE group of the multiple UE groups is associated with a range of values of a measured signal metric. The receiving device can receive a combined signal from one or more UE groups and can normalize the received signal by power to determine the combined coefficient update for the NN.

This application discloses a method, device, and medium for collaborative training of intelligent agents based on a trusted data space. The method includes: classifying and grading heterogeneous financial data from participating parties to generate standardized data; assigning unique identifiers to generate a standardized data catalog and identifying heterogeneous data formats to convert them into a unified format dataset; storing the dataset in a trusted data space to record and document the full lifecycle operations of the unified format dataset, generating operation logs; and encrypting the dataset based on its sensitivity level and computational complexity to generate encrypted data; loading the encrypted data and the agent training code into a trusted execution environment and monitoring operational status indicators to generate an environment health report; coordinating multiple participating parties to train models using local encrypted data based on a federated learning framework and uploading model parameters to the trusted data space; and aggregating and processing model parameters to generate global model parameters and distributing them to multiple participating parties to update their local models.

This invention relates to the field of federated learning privacy and security technology, and proposes an enhanced federated learning gradient leakage attack method based on singular value decomposition, comprising three steps: gradient denoising projection, noise variable optimization, and structure regularization. Gradient denoising projection performs singular value decomposition on the intercepted perturbed gradient, extracts the principal direction to construct a projection matrix, and restricts the gradient matching process to the signal-dominated subspace. Noise variable optimization introduces auxiliary noise variables, jointly calculates the virtual gradient with virtual data, projects it onto the principal subspace, calculates the gradient projection loss, and sets independent learning rates for joint optimization of the virtual data and noise variables. Structure regularization calculates the total variational regularization loss of the reconstructed image, weights it with the gradient projection loss to form a total loss function, iteratively updates the virtual data, and then reconstructs and recovers the original training data. Even if the client gradient is perturbed, this invention can still reconstruct and recover the client's original training data with high quality from the perturbed gradient.

This invention provides a method and system for modeling defect data in a dual-cylinder rotary engine manufacturing line based on federated learning, relating to the fields of intelligent manufacturing and federated learning. The method includes: collecting raw defect data from the production line machine end and standardizing it to obtain standardized defect feature data; extracting the kinematic posture sequence of cylinder block cutting and the dynamic load spectrum of the assembly contact surface from the standardized defect feature data to construct a mechanical motion constraint basis; projecting the standardized defect feature data onto the basis to form mechanical mapping features; and sequentially performing operations including calculating the implicit semantic affinity between the vectors of each dimension of the mechanical mapping features using an interactive attention coupling mechanism, quantifying feature redundancy using an adaptive distance-weighted attenuation model, and constraining the feature distribution offset using an isoangular regularization reference feature space. This invention can achieve secure data sharing and feature alignment across production lines, improving the adaptability and detection accuracy of the defect identification model.

This invention proposes a multi-label federated forgetting method based on feature separation. This method trains an initial global model through federated learning and distributes it to clients. On each client side, based on label attention weights, the activation dimension of all data in the feature space is calculated. By constructing a mechanism to separate private and public features of the target class, features of the target class in the feature space are extracted, and label features highly relevant to the target class and not belonging to the common background are selected. On the server side, the most influential K-dimensional features among all clients are aggregated and distributed to the clients. During the forgetting phase, the client freezes the classification heads and some shared backbone network parameters corresponding to non-target classes. Based on the features distributed by the server, constraints are optimized only for features related to the target class, weakening the model's ability to discriminate the target class while keeping the non-target class feature space as stable as possible. Finally, the server side performs weighted aggregation of the model parameters after forgetting from each client to obtain a global model that completes the target class forgetting.

This invention relates to a method and apparatus for detecting and correcting federated learning data. The method includes: acquiring intermediate data uploaded by each participating party, the intermediate data including location information of data to be supplemented, suspicious data, and indicators related to the suspicious data; performing a global analysis based on the indicators related to the suspicious data to filter the suspicious data and determine the data to be corrected; determining a problem data matrix based on the location information of the data to be supplemented and the data to be corrected; acquiring the associated data corresponding to each problem data in the problem data matrix from each participating party, analyzing the associated data, and determining a correction matrix for each participating party, so as to provide data processing services based on the correction matrix of the participating parties and the local data of the participating parties. The technical solution provided by this invention can effectively solve the technical problems of data accuracy, consistency, and security by combining data quality, data validity detection, and data protection mechanisms.

This invention discloses a privacy-preserving distributed collaborative auditing system and method based on blockchain and federated graph neural networks. The invention proposes a decentralized collaborative auditing architecture. First, local graph neural networks deployed on each audit node extract transaction subgraph features. Without sharing original data, a global audit anomaly detection model is collaboratively trained using a reputation-weighted federated learning algorithm. Second, a hybrid consensus mechanism combining dynamic proof-of-stake and reputation values ​​is employed to complete the on-chain recording of audit logs and the verification of model parameters. Finally, zero-knowledge proof technology is used to verify the authenticity of cross-institutional transactions while protecting trade secrets. This invention achieves the usability without visibility of audit data, significantly improving the ability to identify complex risks such as fraud and money laundering in cross-institutional related transactions and enhancing the credibility of audit results while ensuring data privacy and security.

The present application relates to the technical field of federated learning, and particularly to a blood drug concentration anomaly analysis method based on federated learning, comprising the following steps: a center server generates and issues a proxy data set without privacy risk to a plurality of medical institution clients; and receiving local prediction logits generated and uploaded by the plurality of medical institution clients after predicting the proxy data set by using local heterogeneous models. In the present application, a dynamically evolving global prototype memory bank is combined for anomaly judgment, a normal prototype benchmark reflecting the latest global state is constructed by periodically fusing feature distributions extracted by each node at different time stages, so that anomaly detection no longer depends on static or isolated rules, but is based on a dynamically updated and collectively agreed dynamic standard, thereby improving the robustness and accuracy of anomaly judgment.

A method (200) for performing Federated Learning, FL, in an FL system (100) comprising an FL collecting server (110) and a plurality of contributing servers (120a- d), wherein the method is performed by the FL collecting server (110) The method comprises receiving (210), from at least one contributing server (120a-d) of the plurality of contributing servers (120a-d), at least one first model update contribution, and at least one first assessment weight, obtaining (220) a first updated model of the global model, obtaining (230) a first performance metric of the first updated model; updating (240) the at least one reputation weight to at least one first updated reputation weight, receiving (250), from the at least one contributing server (120a-d), at least one second model update contribution, and at least one second assessment weight, and obtaining (260) a second updated model of the global model.

The application provides a longitudinal federated learning training method and system. The method comprises: a plurality of clients respectively obtain local feature embeddings through respective local models; a server side trains a global fusion network based on the local feature embeddings and pre-set label data, obtains a back propagation gradient, and generates quantization configuration parameters through an intelligent agent decision; the server side distributes the back propagation gradient and the quantization configuration parameters to the corresponding clients, drives the clients to update the local models according to the back propagation gradient and the quantization configuration parameters, and completes the longitudinal federated learning training. Through the end-to-end framework of "heterogeneous local model + longitudinal feature interaction + DQfD intelligent agent bit width scheduling", the application realizes the comprehensive optimization of communication and calculation resources while ensuring the global accuracy, and improves the efficiency and practicality of the longitudinal federated learning in a complex environment.

A federated learning framework including household agents configured to continuously learn model parameters for managing charging periods and discharging periods of household batteries, and microgrid agents to maximize use of local energy based on a pricing policy, including accessing power from other microgrids when there is insufficient local energy to cover local demand. and selling surplus energy to the other microgrids when power generation by the microgrid surpasses the local demand. Each household machine learning agent is configured to control household energy demand from and supply to a microgrid which they are connected in order to minimize household energy cost while adapting to changes in the energy price that is determined based on the pricing policy of the microgrid agent that encourages reduction of carbon emission. A federated learning engine combines the model parameters from the household machine learning agents to update a global household machine learning agent.

The application belongs to the field of artificial intelligence security, and specifically discloses a defense method for specific target forgetting attack on vertical federated learning, which comprises the following steps: each client trains a student model by using a locally stored local data set to generate embedding features, and uploads the embedding features to a server; the server receives the embedding features from each client, calculates the gradient of the loss on the embedding features based on a preset global task loss, and generates a targeted feature according to the gradient, and distributes the targeted feature to each client; each client receives the targeted feature, and updates the student model by minimizing the feature alignment loss between the embedding features generated by the student model and the received targeted feature; when the client obtains a new data set, the client performs incremental learning by using the new data set to update the student model. The application can dynamically adapt to environmental changes and effectively resist the defense mechanism of the specific target forgetting attack with strong concealment.

This invention relates to a decentralized federated learning optimization method and system based on a consensus Kalman filter, belonging to the fields of privacy computing and data security. The invention aims to optimize and improve the effectiveness of decentralized federated learning methods based on differential privacy. It comprises two core components: a gradient filtering algorithm based on a consensus Kalman filter and an overshoot filtering algorithm based on a consensus Kalman filter. The former uses a Kalman filter to filter gradient information during model training, reducing noise bias in the gradient, increasing gradient accuracy and stability, and thus accelerating the model's convergence speed and accuracy. The latter addresses the overshoot problem caused by momentum optimization methods by using a Kalman filter with exponentially decaying system variance to filter overshoot noise in the model parameters, mitigating the impact of overshoot on model convergence speed and accuracy, ensuring accelerated model convergence in the early stages and stable model convergence in the later stages, thereby achieving higher accuracy.

Techniques for training a global model for use at a host of oilfield application sites based on raw data obtained from the application sites without direct exposure of the data to the global model. The techniques include developing and distributing a global model with a predetermined set of parameter weights. The model is then locally employed at each application site by a local computer which maintains the integrity of the acquired data during performance of the oilfield application. The data is used to update the parameter weights based on real-time circumstances. Thus, the parameter weights may be transmitted to the centralized computer for updating of the global model. Further, the updated global model may continue to direct other applications and the process continued in a beneficial feedback loop manner.

The application discloses a heterogeneous federated learning method based on space-time data distillation, which is used for solving the problem that related technologies cannot fully solve the time problem of knowledge reservation, resulting in knowledge forgetting when training a new global model, and the unique space-time characteristics of each client data are not fully considered, so that the efficiency and convergence of federated learning are poor. The server receives the local encoding data obtained by encoding based on the local data sent by each client for data-free global distillation, obtains a first global model, and distributes the first global model to each client; the client performs partial parameter local distillation and non-real-time distillation according to the local data and the first global model combined with dynamic temperature adjustment optimization, generates a local model, and sends the local model back to the server; the server updates the first global model according to each local model through a data-free global distillation strategy, obtains a second global model, and distributes the second global model to each client for next round training.

The application provides a remote intelligent operation monitoring method and system of a smart substation, relates to the technical field of intelligent operation and maintenance of a substation, and relates to multi-source heterogeneous sensing, deep feature modeling, fault prediction evaluation and model self-optimization.The application collects electrical, environmental and meteorological data through heterogeneous sensors, constructs a structured original data set, carries out time series prediction in combination with a convolution-LSTM model, and outputs a fault probability and a confidence interval by using a Transformer fusion network, so that online early warning and response control are realized, and the application has the self-adaptive updating capacity driven by federated learning.

The application discloses an adaptive clustering federated learning modeling method for precision medicine, and relates to the technical field of precision medicine, and comprises the following steps: data collection and preprocessing and model construction and training; the application accurately determines the optimal number of the global model through adaptive clustering, and divides clustering groups according to the similarity of the client model parameters, realizes independent training of the groups, effectively improves the adaptability of the model to different client data characteristics, samples and label distribution differences, avoids the problem that a single model has poor performance in some clients, and significantly enhances the ability of the model to capture complex medical patterns; the model generalization is optimized through grouped training, so that the model can better cope with new data distribution, under the premise of ensuring the privacy and security of medical data, the accuracy and reliability of the model in precision medical scenes such as disease diagnosis and prognosis prediction are greatly improved.

The application relates to the technical field of security protection, can be applied to a business system platform such as financial technology and medical health, and discloses a Byzantine attack defense method, device and equipment based on federated learning and a medium. The method comprises the following steps: acquiring local models and sample data sets of a plurality of target clients, performing forward propagation on the sample data sets by using the local models, constructing a plurality of probability matrices, determining the Euclidean distance between any two probability matrices, constructing a distance matrix, performing hierarchical clustering on the target clients, obtaining a plurality of client clusters, acquiring a probability standard matrix of a standard model, determining the similarity between the probability matrices and the probability standard matrix one by one, marking the types of the to-be-marked matrices according to the similarity, updating the information of the client clusters, obtaining updated clusters, aggregating all the updated clusters, and obtaining a final defense model. The application can effectively improve the defense efficiency and defense scene coverage rate for the Byzantine attack.

The application provides a communication method, device and equipment and a storage medium. The method comprises the following steps: a network device sends a training time delay constraint instruction to a terminal device participating in federated learning; the terminal device receives the training time delay constraint instruction from the network device; and the training time delay constraint instruction is used to indicate the reporting deadline of a model training result. Since the network device can send the training time delay constraint instruction to the terminal device participating in the federated learning, the training time delay constraint instruction is used to indicate the reporting deadline of the model training result, so that in the case that there are two or more terminal devices participating in the federated learning, different terminal devices can upload the model training result according to the training time delay constraint instruction, so as to reduce the time difference of the model training result uploaded by different terminal devices participating in the federated learning, thereby ensuring the overall training time delay of the federated learning, reducing the possibility of late reporting of the terminal device, and helping to reduce the waste of network resources.

This invention belongs to the field of electronic digital data processing technology, specifically relating to a low-altitude communication data anomaly detection system and method based on edge intelligence. The system parses protocol fields from the raw bitstream of the low-altitude communication link layer, extracts type identifiers and payload lengths to construct a protocol state machine transition matrix, and utilizes a Long Short-Term Memory (LSTM) network to extract state transition probability sequence features. Geographically adjacent edge nodes are set as micro-federated learning groups. Each node, after training using local features, only uploads network gating weight parameters to its neighboring nodes for weighted aggregation to update its local model. An alarm is triggered when the real-time state transition probability sequence deviates from the normal transition matrix and exceeds a preset topology threshold. This invention can identify protocol state machine logic errors and transition anomalies, reducing the amount of communication data required for collaborative model updates between edge nodes.

This invention discloses a federated learning method and system for edge-aware networks aimed at refined urban and rural governance, belonging to the field of federated learning technology. The method includes: receiving local model update parameters, physical layer state metadata, and spatiotemporal context identifiers from multiple flexible electronic signs; calculating the effective data collection rate based on the physical layer state metadata, calculating the spatiotemporal consistency metric based on the spatiotemporal context identifier, and fusing them to generate spatiotemporal quality factors for each node; comparing the spatiotemporal quality factors with a preset quality threshold, and if the value is less than the threshold, using a historical reliable parameter library to perform weighted hybrid correction on the current update; calculating a comprehensive aggregation weight based on the local data sample size and spatiotemporal quality factors of each node, performing a weighted average to generate a new round of global model, and distributing it. This invention solves the performance degradation problem of federated learning caused by data intermittency in edge-aware networks by introducing spatiotemporal quality factors, combining historical knowledge completion, and dynamic weighted aggregation.

The application discloses a method for optimizing the flight path and resources of a UAV based on an ISAC system and federated learning, and belongs to the technical field of UAV air-ground communication, joint calculation and perception. The application constructs a multi-UAV wireless communication model with the minimum energy consumption of the UAV as the target, and the communication energy consumption of the UAV, the position and speed of the UAV, the communication demand of the UAV, and the bandwidth and computing resources that can be allocated by the UAV as constraints. The application can more accurately estimate the position of the ground terminal, more quickly solve the communication demand of the dense ground terminal, and allocate deep learning tasks to multiple UAVs for calculation, fully utilize the computing resources of the UAV, and reduce the computing pressure of a single UAV. The application also jointly optimizes the bandwidth of the multiple UAVs and the ground terminal, the trajectory of the UAV, and the deep learning task calculation, so that the wireless communication rate is maximized while the energy consumption of the UAV is minimized, and the energy efficiency of the communication between the UAV and the ground terminal is improved.

A new and improved technology capable of further improving inference accuracy while securing security of a machine learning model by federated learning is proposed. Provided is an information processing apparatus including: a learning unit that learns an inference model by federated learning; and an acquisition unit that acquires, from a plurality of terminals, privacy-protected data obtained by executing privacy protection processing on local data obtained by each of the plurality of terminals, in which the learning unit is configured to: perform learning of the inference model on the basis of the privacy-protected data; and distribute information regarding the inference model including a hyperparameter of the inference model on the basis of a result of the learning to the plurality of terminals, the acquisition unit is configured to acquire, from the plurality of terminals, update information of the inference model obtained by performing the learning of the inference model using the local data as learning data, and the learning unit is configured to update the inference model using the update information.

The application relates to a federal learning-based information transmission system, comprising a center server and a plurality of participants for executing a training task, task information of the training task comprising first type information stored in the center server and second type information stored only in each participant; when the center server receives a request of a user for viewing the first type information, corresponding first type information is provided according to the authority of the user; when the center server receives a request of the user for viewing the second type information, a link channel connected to a corresponding participant for accessing the second type information is provided according to the authority of the user. When the user accesses the second type information, the user is provided with a link channel for accessing the second type information, so that the second type information is prevented from being uploaded and downloaded on the center server data, and the viewing task log is prevented from affecting the data throughput and the speed of federal learning training.

This invention discloses a method for architecture generation and hierarchical alignment-based precise aggregation for heterogeneous federated learning, comprising the following steps: 1) initializing a global supernet; 2) constructing an architecture exploration environment based on reinforcement learning; 3) performing offline generative exploration in the supernet search space; 4) constructing a Pareto optimal candidate architecture pool; 5) determining the set of clients participating in training and dividing resource levels according to resource budget; 6) selecting the optimal nested subnet for each resource level; 7) distributing the selected subnet architecture and parameters to clients for local training; 8) generating parameter masks for each client; 9) performing precise aggregation of gradients uploaded by clients based on the masks and updating the global supernet weights; 10) iterating through communication rounds until the model converges or reaches a preset number of rounds. This invention solves the suboptimal performance problem caused by traditional heuristic architecture generation methods, as well as the parameter space mismatch and aggregation interference problems caused by incompatibility of heterogeneous subnet structures.

This application relates to a method, system, and medium for diagnosing high-loss anomalies in distribution networks based on federated transfer learning. The method includes the following steps: dividing the distribution network into N regions, with each region acting as a federated learning client to collect distribution network data from each region; establishing a correlation analysis model between electrical characteristics and high-loss anomalies through data preprocessing and feature engineering to provide high-quality input data for federated learning; each client using a 1D-CNN+LSTM hybrid model for local training after data processing; a Bayesian hierarchical aggregation center receiving model parameters or update information from clients in each region, performing aggregation calculations to generate or update the global model; and rapidly deploying the global model to the target region clients for diagnosing the causes of high-loss anomalies in the target region. This application achieves efficient collaboration of model performance and lightweight, rapid deployment, significantly improving the automation level and operational efficiency of high-loss anomaly diagnosis.

The application provides an image generation method based on federated learning and an optimal time allocation strategy, comprising the following steps: determining the sparsity according to the client communication budget and the data volume, and randomly initializing the diffusion model parameters and the personalized sparse mask; determining the neighbor clients of each client according to different optimal time allocation strategies, and determining the mask intersection parameter aggregation mode of the current client and the neighbor clients according to the current round and the aggregation probability, to obtain the aggregated initialization model; performing gradient update on the initialized model parameters after aggregation according to the data batch and the parameter mask state, to obtain the updated model parameters; and calling a gradient information dynamic search algorithm to update the sparse mask in each client based on the updated model parameters. The application can protect the training data privacy of users of each party, can jointly use the data of each party to improve the generalization ability of the model, and can efficiently allocate limited communication resources under the condition of bandwidth limitation.