191 results about "Model aggregation" patented technology

The invention relates to the technical field of agricultural information management, and particularly discloses an agricultural information management system and method based on a big data platform, and the method comprises the steps: firstly deploying a multi-source data collection module at an edge calculation node, and obtaining and standardizing the soil moisture content, meteorological environment and equipment operation data in real time; secondly, constructing a local dynamic irrigation strategy model, and realizing multi-objective optimization through a reinforcement learning algorithm; establishing a federated learning framework at the cloud, dynamically distributing node weights by adopting an attention mechanism, and realizing model aggregation of privacy protection in combination with secure multi-party computing; an optimal irrigation instruction is generated through a multi-source data fusion engine, and a three-level response exception handling mechanism is established; and finally, a closed-loop feedback system containing short-term incremental learning and long-term architecture optimization is formed. The corresponding management system comprises six functional modules, namely a data acquisition module, a local modeling module, a federated learning module, a real-time decision-making module, an abnormal monitoring module and a closed-loop optimization module.

The invention discloses a multi-source threat intelligence privacy fusion processing method and system, and relates to the technical field of network security threat intelligence analysis and processing, and the method comprises the steps: collecting threat intelligence, and carrying out the cleaning, standardization and desensitization; establishing a threat index semantic model and aligning cross-modal features; the central coordination module completes model aggregation after organization node local training; and based on the aggregation model fusion intelligence, semantic association and attack link reconstruction are established, and a result is generated and fed back for optimization. The technical problems that in cross-organization fusion processing of multi-source heterogeneous threat intelligence, data formats are not uniform, and privacy protection and intelligence collaborative analysis contradictions exist, so that threat detection is not comprehensive and inaccurate, and accurate evaluation and effective management and control requirements are difficult to meet are solved. The technical effects that the multi-source heterogeneous threat intelligence is effectively fused on the premise of privacy protection, the comprehensiveness and accuracy of threat detection are improved, and the requirements for accurate assessment and effective management and control of threats in a cross-organization scene are met are achieved.

The invention belongs to the technical field of block chains and federated learning, and particularly discloses a dynamic incentive federated learning method based on a trusted execution environment and a block chain. The method comprises the following steps: firstly, constructing a core computing security area by utilizing TEE, and executing key links such as model aggregation, client screening and contribution measurement in a hardware isolated trusted environment; and then, a block chain and an intelligent contract technology are adopted as a decentralized trust root to realize effective management of identities of participants, tamper-proof records of key certificates and automatic distribution of economic incentives. Secondly, establishing a set of dynamic excitation and reputation mechanism executed in the TEE, and performing credible quantification and automatic reward on the contribution of the client based on multi-dimensional indexes such as model quality, historical reputation, asset pledge, participation stability and the like; and finally, end-to-end data encryption is realized through a session key mechanism in the TEE, so that the data privacy of the client is effectively protected.

The invention discloses a non-independent identically distributed data asynchronous federal learning method based on an improved aggregation algorithm. The method comprises the steps that a server initializes a global model and issues the global model to all clients; and the client performs local training on the received global model by using local data, and uploads the model and model parameters to the server after training is completed. Then, the server adjusts a model lag degree based on a client data volume proportion, calculates model difference consistency, client historical contribution stability, old degree penalty of the client model and cosine similarity of the client model and the global model based on parameters of the client model and the current global model, and generates an asynchronous federal aggregation factor accordingly; and updating the global model parameters to generate a new global model. And finally, testing the global model by the server, and judging whether the learning process is stopped or not. According to the method, fair and effective model aggregation can be realized, and the model convergence stability and the final model detection precision are improved.

The invention discloses a heterogeneous data financial intelligent model collaborative construction device and method and related equipment, and relates to the technical field of model processing, and an institution grouping device of a public control unit clusters and groups financial institutions participating in model construction by using clustering, data distribution speculation and other related technologies. The mechanisms of the same type in the grouping result have similar data distribution characteristics, and the mechanisms in the same group have data with complementary distribution characteristics, that is, missing data knowledge is mutually compensated, and finally good training data with the same distribution are formed on the whole. The grouping result is input into a first model scheduler to schedule mechanism selection of local training of the federal model, the model is guided to complete training and learn complete data knowledge in different grouped mechanisms, and a financial intelligent model meeting result expectation is output through multiple rounds of local model training, correct model scheduling and model aggregation. The method solves the influence of heterogeneous data, and constructs a financial intelligent model with high accuracy.

The invention discloses an air-space asynchronous federated edge learning-based general computing resource management method, which comprises the following steps of: 1, establishing an air-space asynchronous federated edge learning system model assisted by a low-orbit satellite, and designing a layered federated model aggregation mechanism; 2, describing a communication-sensing-computing resource management problem with optimal global model convergence, and considering terminal equipment energy consumption and training time delay constraints; step 3, decoupling the problem into three sub-problems by adopting alternative optimization, namely satellite receiving denoising factor optimization, terminal transmitting power control and sensing data batch size design; and step 4, solving the three sub-problems based on mathematical optimization methods such as quadratic constraint quadratic programming and fractional programming, and obtaining an optimization result of general inductance computing resource management through iteration. According to the method, an aggregation mechanism of asynchronous federated learning is reasonably designed to balance contributions of different terminals to model convergence, and meanwhile, an efficient problem decoupling and optimization method is provided to realize joint allocation of multi-domain resources of the general inductance calculation, so that the federated learning performance is improved.

The invention relates to a multi-party trusted cooperation method and system based on a DAG block chain and federal learning. According to the method, a task publisher publishes a global initial model and task metadata in a DAG block chain, and starts a collaborative training task; a task training party dynamically selects Tip nodes from a DAG block chain cooperation module based on freshness, accessibility and data distribution similarity, aggregates models according to a set Tip selection rule-based model aggregation strategy to obtain a local training model, and trains the local training model; after the model is aggregated, a model aggregation path is uploaded to a DAG block chain cooperation module to generate a corresponding Hash commitment; after model training is completed, model parameters and model updating metadata are encrypted and then uploaded to a DAG block chain cooperation module; after the DAG block chain cooperation module confirms that no tampering exists, Tip node updating is executed by utilizing the updating data; and the task issuer issues a training termination command to terminate training according to a preset global training termination condition.

The invention provides a dynamic threshold adjustment method based on a self-adaptive alarm rule engine, and belongs to the field of intelligent power grids, and the method comprises the steps: collecting equipment operation data in real time by an edge node, and building a personalized threshold model; the central server initiates a model aggregation request according to a preset period, each edge node uploads a personalized threshold model parameter increment, calculates an aggregated global model parameter, adds noise to the aggregated global model parameter to obtain the global model parameter, and establishes a global model; inputting the standardized feature vector into a global model to obtain a prediction probability, and calculating a basic threshold according to the prediction probability; if the prediction probability exceeds a basic threshold value, generating a weight vector for the feature vector after standardization processing to obtain a weighted feature; calculating an anomaly degree, calculating a smoothness anomaly degree, calculating an adjustment factor, and adjusting the basic threshold value to obtain a dynamic threshold value; the invention further provides a dynamic threshold adjusting system. The problem that a traditional fixed threshold value cannot cope with equipment parameter volatility enhancement is solved.

The invention relates to the technical field of project management informatization, and discloses a project supervision cost consultation integrated management system, which comprises a dynamic risk modeling module, a risk probability distribution and gradient matrix generated by fusing multi-source data through a dynamic Bayesian network, and a federal collaborative learning module, the system comprises a model aggregation module, a multi-target optimization module, a block chain auditing module, a closed-loop feedback module, a block chain auditing module, a block chain auditing module, a block chain auditing module and a block chain auditing module, wherein the model aggregation under multi-party privacy protection is realized based on a timestamp-driven algorithm and gradient confusion encryption; the multi-target optimization module is used for performing cost, construction period, quality and risk tolerance four-dimensional optimization by using an improved NSGA-III algorithm; and dynamically correcting system parameters according to an audit result to form a closed loop. The reliability and the intelligent level of the consultation management system are improved by fusing the multi-source data quantification risk through the dynamic Bayesian network, combining federal collaborative learning with gradient encryption, risk gradient optimization decision and closed-loop mechanism dynamic correction.

The invention discloses a federated learning dynamic aggregation method, system and device, a medium and a product. According to the method, a twinborn layer is constructed on a central server, model aggregation deduction is performed in advance in a virtual environment by establishing a digital twinborn model of a client, and then an optimal aggregation strategy is screened out; in each round of iteration, the client maps self state information and local model parameters to a twinborn layer, the twinborn layer clusters the client by using a K-means dynamic clustering algorithm, the aggregation effect of different cluster combinations is evaluated, and an optimal aggregation strategy is selected for pre-aggregation; therefore, the server globally aggregates the client model according to the strategy provided by the twinborn layer. According to the embodiment of the invention, the model convergence speed and precision of federal learning can be remarkably improved, the communication delay and energy consumption are reduced, the method effectively adapts to the isomerism of clients, and an efficient and flexible solution is provided for distributed machine learning.

The application provides an asynchronous federated learning system and a neural network framework, and relates to the technical field of picking robots. The asynchronous federated learning system solves the ''short board effect'' of synchronous federated learning, and improves the performance of the global model on non-independent and identically distributed data through dynamic clustering and adaptive model aggregation. At the same time, the neural network framework enhances the perception ability and decision accuracy of the picking robot in a complex agricultural scene, thereby improving the collaborative training efficiency of the asynchronous federated learning system in the picking robot in the smart farm.

The invention relates to the technical field of artificial intelligence, and discloses a model training method, and clients generate different aggregation strategies for each client on the premise of not sharing privacy information by uploading key model parameters and graph-level samples. A representative sample with privacy protection property is constructed at a client side through an adversarial learning mode, and the representative sample is uploaded to a server for subsequent personalized model aggregation. According to the personalized global aggregation strategy, graph-level representation is extracted from uploaded information on a server side, and graph similarity between clients is estimated, so that a personalized aggregation strategy is formulated through collaborative integration of multiple clients, the clustering performance of a local model is enhanced, the relationship between the clients can be accurately captured, and the clustering effect is improved. When personalized recommendation is carried out on a trained clustering model, the accuracy of personalized recommendation is improved. The invention further discloses a model training device, a storage medium and a program product.

The invention relates to a full-life-cycle intelligent monitoring and health assessment method and system for substation equipment, and belongs to the technical field of substation equipment monitoring. Therefore, the health degree information of each piece of substation equipment is obtained through the life cycle monitoring model of the substation equipment, early warning is carried out according to the health degree information of each piece of substation equipment, early warning information is generated, and finally a related maintenance scheme is generated based on the early warning information in combination with the aging evolution factor data in the target area. According to the method, the data of the collection equipment is processed through federated learning, the size of the collected data is dynamically adjusted by combining the state and operational capability of the actual equipment, and the potential contribution of local data distribution of the equipment to the improvement of the current global model is comprehensively evaluated, so that the deviation of model aggregation of the federated learning is reduced; the prediction accuracy of the life cycle monitoring model is improved, and the reliability of early warning is improved.

The invention provides a personalized federated learning method based on dynamic sparse structure adjustment, which is oriented to a non-independent identically distributed data scene and solves the problems of high communication and calculation overhead and low convergence speed in federated learning. According to the method, model rarefaction and personalized modeling are combined, and an initial sparse structure is constructed according to the parameter scale of each network layer under the constraint of a fixed overall sparse rate in an initialization stage; in the training process, a dynamic sparse updating mechanism based on gradient information is introduced, and adaptive pruning and regeneration are carried out on model connection so as to continuously optimize the model structure. In order to avoid falling into a suboptimal sparse mode, the pruning and connection regeneration proportion is dynamically adjusted, and the structure searching capability is enhanced. In the model aggregation stage, a mask perception parameter aggregation strategy is provided, only activation parameters are aggregated, and aggregation deviation is reduced. Experimental results show that the method significantly reduces the communication and calculation overhead while ensuring the model precision, accelerates the model convergence speed, and is suitable for a federal learning scene with limited resources.

The application relates to the technical field of estimation of maximum frequency deviation of power grid nodes and is used for solving the problem that the existing estimation method of maximum frequency deviation of power grid nodes cannot reflect the time-space difference characteristics of frequency response of an interconnected power system. First, a two-region frequency response model TAFR considering the time-space difference characteristics of frequency of power grid nodes is established, a time-domain analytical expression describing the node frequency difference mechanism is obtained by adopting the mode analysis and average system frequency model aggregation idea, and a maximum frequency deviation estimation method reflecting the time-space difference is proposed on the basis of the analytical expression. The established TAFR model considers the distribution of tie-line parameters, system inertia and damping, and equivalently processes the new energy frequency modulation parameters, compared with the existing frequency dynamic response model, the model has improved accuracy and applicability while ensuring the time-space difference.

The application discloses a federal learning system based on a FISCO group mechanism, which comprises a supervision node group and multiple levels of client groups, for any client, obtains and updates a local model according to a global model obtained through initialization or a last learning iteration; for any client group, based on a practical Byzantine fault tolerance consensus mechanism and a malicious node screening mechanism based on cosine similarity, an updated local model is obtained; based on a blockchain or a supervision node, an updated global model is aggregated; until a preset federal learning convergence condition is met, otherwise, the next round of federal learning is entered. The application divides different levels of client groups, guarantees the privacy of the blockchain system while retaining security and decentralization, and the introduction of the supervision node can not only play a penetrating supervision across agencies, but also greatly improve the efficiency of model aggregation.

This specification discloses a method and apparatus for defending against poisoning attacks based on federated learning, comprising: a central server randomly selecting clients to participate in training from all clients participating in federated learning, and synchronizing a global model with the clients participating in training as an initial global model; the clients participating in training performing local training on the initial global model to obtain the upload gradients of normal clients; the central server extracting features from long-term historical gradients to obtain score index sorting to remove malicious clients from the clients participating in training, obtaining candidate clients; the central server performing model aggregation based on the candidate clients, the clients participating in training, and the upload gradients, and updating the model parameters of the global model, the updated global model serving as a new round of initial global model, iterating until a preset number of iterations is reached to obtain a trained global model.

The invention discloses a doubly-fed wind turbine generator-oriented wind power plant model aggregation method and device and a medium. The method relates to the field of wind power generation, and solves the problem that wind power plant model aggregation cannot simultaneously guarantee analysis precision and reduce calculation burden requirements. Based on a hybrid wind power plant structure, a dynamic mathematical model of a single machine and a whole wind power plant is obtained, contribution of each WTG to power grid injection current is quantified into a dynamic weight coefficient, the coefficient is updated in real time along with operation conditions, and the limitation that only a single machine type is concerned or dynamic characteristics are simplified in a traditional method is broken through. The method comprises the following steps: constructing an equivalent model for electrical and control mechanical parameters of a wind driven generator, a wind energy capture characteristic of a turbine and an impedance distribution characteristic of a line, ensuring that an aggregation model retains a core physical mechanism of an original wind power plant, and replacing detailed modeling of dozens of to hundreds of WTGs with a single-unit equivalent model. And the problem of calculation burden of large-scale wind power plant modeling is solved on the premise of ensuring the precision.

The application relates to a task self-adaption and federal learning system for edge side heterogeneous task sequences, comprising the following steps: 1, a knowledge extraction module extracts each client local model into a compact knowledge, forms a knowledge distillation model, and sends the knowledge distillation model to a server; 2, the server uses a heterogeneous model selection module to search a task memory palace, the task memory palace comprises task knowledge storage, searches similar task knowledge storage, adds new task knowledge, and finds a part of the most similar task knowledge for each client, and the like. The learning system has the advantages that the learning system is used for the problem that task differences of various edge devices in a real edge environment are large, in the background, the application is started from the perspective of task similarity, key data structures and four modules are used to ensure that similar tasks are aggregated, and the precision of local model training is improved under the condition that the calculation cost and the communication cost are low.

The invention discloses an equipment maintenance robot group for mutual learning of maintenance experience and a working method thereof. The system comprises a cloud server, at least three maintenance robots, an edge computing node and a distributed sensor network. The cloud server stores a global maintenance experience library, and a block chain-database hybrid architecture is adopted to ensure that data cannot be tampered and is efficiently stored. The maintenance robot integrates a multi-modal data acquisition module, a federated learning module, a heterogeneous communication module, a maintenance execution module and an energy management module. The edge computing node is responsible for local experience verification, conflict resolution and model aggregation, and cloud load is reduced. And the distributed sensor network feeds back the equipment state in real time to assist in fault pre-judgment. Through the steps of multi-modal data acquisition and preprocessing, local experience generation, encryption and uploading, edge verification and aggregation, global collaborative optimization, dynamic task allocation and execution and the like, robot group cross-regional maintenance experience mutual learning is realized, and the intelligent level and overall efficiency of equipment maintenance are improved.

The invention relates to the technical field of computer vision, in particular to a distributed collaborative safety helmet detection model training system, which comprises a data acquisition and preprocessing unit, a model training unit, a model aggregation unit, a model deployment and reasoning unit and a safety and communication coordination unit. According to the method, node weights are dynamically calculated through contribution-credibility dual evaluation, and local parameters are fused in combination with a federated average algorithm, so that negative effects of low-contribution and low-credibility nodes on a global model are effectively avoided, and a global optimization model which is high in robustness, high in detection precision and resistant to hostile attacks is generated; and meanwhile, the aggregation universality is improved by adapting to a non-independent identically distributed data scene.

The present application relates to a kind of safe robust feature joint method, belong to artificial intelligence field.The method includes: edge area division;System initialization: initialize global model, key generation and the key parameter of executing protocol;Device data equalization: the data distribution of equipment is collected, and the sampling mode of each class is generated for enhanced sampling or down sampling with reference to global data distribution, and the sample proportion of enhanced sampling and down sampling is set;Local model training: according to each class belongs to minority class set or majority class set, corresponding sampling strategy is adopted to execute random gradient descent algorithm, and gradient information is secret sharing;Edge security robust aggregation: design lightweight security aggregation protocol to support the implementation of robust aggregation algorithm based on gradient secret value;Cloud global model aggregation: receiving the local model aggregation result returned by each edge server, and global model aggregation is carried out using federated average algorithm.The present application improves the robustness and security of federated learning global model.

The application discloses a method for defending against poisoning attacks in blockchain federated learning supporting privacy protection, and relates to five links of initialization, model training and blinding, malicious detection, model aggregation and model updating. In the process of malicious detection, the method accumulates abnormal fluctuations in the form of historical model updates to highlight the characteristics of malicious behavior and realize accurate identification of malicious behavior, thereby effectively defending against poisoning attacks and improving the robustness of the global model. In addition, the poisoning attack defense effect of the method will not be affected by the protection of model updates, which can realize malicious detection using original features and avoid direct exposure of participant data privacy, to a certain extent, alleviate the contradiction that the original features of model updates are hidden after the implementation of privacy protection, which is not conducive to the accurate identification of malicious detection, thereby further enhancing the overall security of blockchain federated learning.

The invention relates to an intelligent charging pile demand prediction method based on graph convolution under a federated learning architecture, and belongs to the technical field of electric vehicle charging. The method aims at solving the problems of model misinterpretation risk, insufficient time-space dependence complexity modeling and data privacy protection in a traditional prediction scheme. According to the technical scheme, the method comprises the steps that a central server executes multi-modal meta-learning pre-training of economic law constraints to generate a global basic model; the intelligent charging pile locally processes the charging occupancy rate and price data through a time value fusion module, generates encryption features and uploads the encryption features to the edge server; an edge server constructs a multi-scale adjacent graph to perform graph convolution space feature extraction, and long and short time sequence dependence is modeled in combination with a time mode attention mechanism; and cloud edge-end collaborative model aggregation is realized through a federal learning strategy of adaptive weight. The method has the technical effects that price demand misjudgment is fundamentally eliminated, multi-dimensional spatio-temporal feature collaborative modeling is realized, and data privacy security and system robustness are guaranteed.

The invention discloses a double-classification head voiceprint recognition training method based on federal learning. The process of the training method comprises the steps that a client provides local speaker category information, a server integrates global speaker categories, the server initializes a global classifier and a global encoder, the server distributes the global classifier and the global encoder, and the client receives the global classifier and the global encoder and then carries out training together with a local classifier. In the local training process, global classifier parameters are updated through a moving average algorithm, cooperative training of the global classifier and a client local classifier is provided, and the model performance is remarkably improved. In the model fusion process, in order to solve the influence caused by data distribution deviation, a global classifier carries out weighted averaging according to categories contained in each client. According to the method, the local model with better performance can be obtained, so that the global model performance loss during model aggregation is reduced, and a forward feedback mechanism is formed for subsequent model training.

Embodiments of the present application disclose a data clustering and summarizing method and device, equipment and a storage medium. The method comprises: obtaining a target suggestion dataset corresponding to a suggestion type problem insight task; performing vectorization clustering on the target suggestion dataset to obtain a target suggestion clustering result corresponding to the target suggestion dataset; using a target large model to summarize the same type of target suggestion data in the target suggestion dataset based on the target suggestion clustering result, to obtain target summary description data corresponding to each suggestion category in the target suggestion dataset. In the clustering insight scene of the suggestion type problem, through the vectorization clustering and large model summarizing manner, not only the artificial cost required for the suggestion data clustering and summarizing can be saved, and the efficiency of the suggestion type problem sorting can be improved, but also the same type of suggestion data can be expressed and summarized according to the suggestion clustering result, so that the user can efficiently and intuitively understand the key points of each suggestion category in the suggestion clustering result.

A communication method and apparatus. The method comprises: a first communication apparatus sending a first model feature to a second communication apparatus, wherein the first model feature is used for representing model features of a first model, such that the second communication apparatus can execute a corresponding operation on the basis of the first model feature. For example, the second communication apparatus may aggregate, on the basis of the first model feature, a plurality of models which comprise the first model. There is no restriction on whether the plurality of aggregated models have identical model structures, thereby improving the flexibility of model aggregation.

This invention relates to a self-learning method and system for home environment regulation based on multimodal feedback and scene awareness. The method collects the user's physiological characteristics and body movement signals through non-contact sensing devices to identify the user's current activity scene. The system integrates the user's explicit intervention operations with implicit comfort feedback based on physiological stability, dynamically assigning fusion weights for explicit and implicit feedback according to the activity scene to generate target feedback labels. An online incremental learning algorithm is used to update the local environmental control model, and a privacy-preserving model aggregation is performed in the cloud through a federated learning mechanism. This invention solves the problems of traditional home control relying on single commands and lacking physiological feedback mechanisms, achieving intelligent closed-loop control that balances user privacy and personalized comfort.

The invention discloses a multi-mode aggregation optimization method and equipment for a multi-mode database and a medium, belongs to the technical field of multi-mode databases, and aims to solve the technical problem of how to reduce coding loss caused by unified calculation of multi-mode data, improve the overall query efficiency of the database and improve the overall query efficiency of the database. According to the technical scheme, a push-down agg operator is added on a time sequence calculation engine, plan conversion is carried out through the push-down agg operator, data connection of different modes of a multi-mode database is optimized, and the query speed during aggregation calculation exists; the method specifically comprises the following steps: performing prepolymerization calculation through a push-down agg operator to obtain a prepolymerization calculation result, and reducing the data volume output by a scan operator; uniformly coding the result of the pre-polymerization calculation through a join operator to obtain uniformly coded data; performing connection calculation on the uniformly coded data through a relation calculation engine to obtain a connection calculation result; and performing second aggregation calculation on the connection calculation result through an agg operator to obtain a final aggregation calculation result.

Provided are a communication method, a terminal device, and a network device. The communication method is applied to a machine learning model and comprises: a terminal device receiving a global model parameter related to a first model sent by a network device; the terminal device sending a first parameter to the network device, wherein the first parameter is determined on the basis of the global model parameter; the terminal device receiving a plurality of candidate parameters sent by the network device; and on the basis of the plurality of candidate parameters, the terminal device determining a local model aggregation parameter that is to be sent to the network device, wherein the terminal device is one of a plurality of terminal devices, the plurality of candidate parameters are determined on the basis of a plurality of first parameters sent by the plurality of terminal devices, and a plurality of local model aggregation parameters of the plurality of terminal devices are used for training the first model.