44 results about "Model sharing" patented technology

Methods of enabling Artificial Intelligence (Al) models to be trained in a privacy preserving manner are disclosed. A method comprises the formation of virtual cluster per privacy level and selection of a cluster head that performs the training and coordination per privacy level. A method comprises training an Al primary model on a private user data set. The Al primary model may be used to label a public data set. A secondary model may be trained using the labeled public data set. The secondary model may be shared with a network or other UEs.

In one implementation, a method of decoding point cloud data for a point cloud is presented. A signal indicative of a number of sub-blocks, N1 is decoded, and a neural network is configured to have N1 sub-blocks. In particular, each sub-block of the N1 sub-blocks includes an upsampling function and at least a neural network layer, and each of the N1 sub-blocks is configured with the same neural network parameters. The point cloud data is decoded based on the neural network. At the encoder side, the signal indicative of N1 is encoded, and the neural network is configured to have N1 sub-blocks. In particular, each sub-block of the N1 sub-blocks includes a downsampling function and at least a neural network layer, and each of the N1 sub-blocks is configured with the same neural network parameters. The point cloud data is encoded based on the neural network.

The invention provides a federal learning model copyright protection method and system based on feature fusion, and the method comprises the steps: 1, receiving a watermark triggering sample, and inputting the watermark triggering sample into a pre-established client model for training; 2, updating and aggregating the local shared model parameters and the local classification layer parameters to obtain updated global model shared parameters and personalized classification layer parameters; performing watermark feature extraction on the updated global model sharing parameters based on a feature layer watermark embedding mechanism to obtain a global watermark; 3, performing multi-bit quantization on the personalized classification layer parameters to obtain a set of binary signatures; and performing copyright verification based on the black box watermark and the global watermark, and determining a malicious client based on a set of binary signatures. According to the method, illegal copying and distribution of the model can be effectively restrained, and a key technical support is provided for promoting healthy and sustainable development of a data and model cooperation normal form.

The present invention belongs to the field of federated learning technology and discloses a method, apparatus, device, and medium for constructing a random model sharing pool for federated learning. The method comprises: randomly selecting at least one base model from the model sharing pool; randomly assigning the at least one base model to multiple participants; obtaining the model parameters of the local model recorded by each participant after training, aggregating the model parameters of all local models corresponding to the same base model after training, and obtaining a global model after the aggregation of each base model; then randomly selecting at least one global model and re-randomly assigning it to multiple participants to update the local model of each participant, and each participant then performs the next iterative training with the updated local model until the global models corresponding to each base model converge. The present invention can provide better guarantees for privacy protection in federated learning and provide a better privacy protection solution for digital governance.

The invention relates to a federated learning method and system based on dynamic clustering and double-ring cooperative training, and relates to the field of federated learning. The problems that a clustering mechanism is insufficient in adaptability and low in knowledge sharing efficiency are solved. The method comprises the following steps: step 1, receiving non-sensitive features from each industrial control device participating in federal learning; 2, clustering the industrial control equipment according to the similarity of each non-sensitive feature to form a plurality of federal sub-clusters; 3, independently executing inner ring knowledge distillation training by each federal sub-cluster, executing outer ring model sharing training among the federal sub-clusters, and judging whether a training ending condition is met or not; and step 4, updating the non-sensitive features, and repeating the step 2, the step 3 and the step 4 until a training ending condition is met.

The invention relates to the technical field of data sharing, in particular to a GIM model sharing method based on a digital intelligent design management platform. The method comprises the following steps: acquiring a GIM model data set, analyzing the GIM model data set, and determining a core construction data set and a three-dimensional geometric data set; obtaining a project participant responsibility division information set, analyzing the core construction data set based on the project participant responsibility division information set, and determining a core data encryption strategy set; based on the project participant responsibility division information set, analyzing the three-dimensional geometric data set, and determining a three-dimensional model identification information set; and according to the core data encryption strategy set and the three-dimensional model identification information set, distributing a corresponding local online fusion type security data sharing strategy to each project participant, and outputting a data sharing report. According to the method, good balance between data security and collaborative high efficiency in a GIM model sharing process is realized through a local online fusion type security data sharing strategy.

The invention discloses a device for controlling a model question and answer result through role definition, and the device sets different roles through a role model setting module, and carries out the output limitation of the roles. The role automatic verification module verifies the content output by the role model by using a verification rule; a model result output comparison module performs similarity verification on files according to output results of different roles, compares similar text contents in different roles one by one, and confirms role model versions; performing question-answering by using a role model question-answering module; meanwhile, the model optimization module can configure output elements and form a specific role model according to the current role of the user, and defines a content generation method taking figures, things, events, targets and scenery as the center by performing condition combination setting on the model, so that the accuracy is greatly improved; meanwhile, sharing among different role versions is achieved through the model sharing module, and continuous optimization learning is conducted on the role template through the model continuous learning module.

The application discloses a kind of model sharing method, device, equipment, medium and product based on multi-chain structure, which comprises the following steps: model training is carried out on the federal learning network of the multi-chain structure created to the node participating in model sharing, to generate federal learning model;Wherein, the federal learning network includes a plurality of data blockchains supporting cross-chain communication, and the data blockchains correspond to the nodes one by one;According to the data distribution characteristics of each node in the federal learning network, the federal learning model is segmented to obtain a plurality of model blocks, and the plurality of model blocks are encrypted to obtain a plurality of encrypted model blocks;Wherein, each data distribution characteristic corresponds to at least one model block;Each encrypted model block is uploaded to the data blockchain of the corresponding node for storage and sharing;The application combines the advantages of blockchain technology and federal learning, realizes the federal learning model sharing of multi-chain structure, can effectively improve the efficiency of model sharing and reduce communication overhead.

The present invention relates to a method for intelligent and secure sharing of vehicle-road collaborative data, and belongs to the field of mobile communication technology. First, local redundant data and noise data are screened to avoid overfitting. Secondly, a malicious model detection scheme is designed to provide a defense method against label flipping attacks; then, a pre-learning scheme based on dynamic mutual distillation technology is established to improve the generalization ability of the model and reduce the communication load; finally, a personalized aggregation strategy based on CAV is proposed. After the last CAV mutual distillation is completed, the final mutual distillation model is broadcast to all remaining CAVs, and the CAVs evaluate the accuracy of the mutual distillation model and the local training model based on their respective local data sets. It is used to solve the problems of low model accuracy caused by heterogeneous data in the Internet of Vehicles scenario, high model sharing communication overhead, and label flipping attacks on federated learning. This solution improves the generalization ability of the model and maximizes sharing efficiency under the premise of ensuring safe and low-load CAV collaborative training.

The invention provides a system and a method for deploying a model based on a cluster. The resource utilization rate of the model after being deployed in the cluster can be improved. The system based on the cluster deployment model comprises a scheduling controller which is used for receiving a reasoning request, adjusting resource distribution of a multi-model shared container group and distributing the reasoning request to the multi-model shared container group to which a corresponding model belongs; the multi-model sharing container group is in communication connection with the dispatching controller, and the multi-model sharing container group is used for simultaneously operating a plurality of models of the same type; and the distributed memory is in communication connection with the multi-model sharing container group, and the distributed memory is used for storing model file data and model metadata of the multi-model sharing container group.

The invention relates to a personalized federal learning scheme based on model similarity and historical performance weighting. The personalized federal learning scheme specifically comprises the following steps: step 1, initializing model parameters by each client; 2, the clients independently train models on a local data set, and model parameters are sent to other clients through a partial model sharing mechanism; step 3, after the client receives the model parameters of the other clients, calculating the cosine similarity among the model parameters, and dynamically adjusting the aggregation weight in combination with the historical performance improvement rate of the client; 4, the client selects the most similar K model parameters for aggregation according to the adjusted aggregation weight, and local model parameters are updated; and step 5, the client checks the personalized model, if the personalized model converges or reaches a preset training round, training is stopped, and otherwise, the step 2 is returned.

The invention discloses a GPU (Graphics Processing Unit) video memory and computing unit multi-model sharing method, which comprises the following steps of: S1, splitting each network layer for carrying out data processing on different models into a universal layer and a distinguishing layer, and respectively deploying each universal layer and / or each distinguishing layer of each model on a corresponding GPU computing unit; and S2, identifying a network layer splitting structure of the model to be operated, calling the first GPU computing unit to execute data processing on the identified universal layer through a load balancing strategy, and calling the second GPU computing unit to execute data processing on the identified distinguishing layer. By splitting and deploying the model, the model adopts the same universal layer, different distinguishing layers and model feature structures to realize multiplexing, video memory exclusive occupation is avoided, video memory multiplexing is realized, more models can be deployed, and fragmented video memory space is fully utilized.

The invention provides a WebGL and URDF-based mechanism animation model self-building sharing method, system and device, relates to the technical field of data sharing platforms, and comprises a preprocessing stage, a loading stage, a running stage and an interaction stage. According to the technical scheme, the dependence of traditional mechanism simulation software on specific equipment and complex environments is eliminated, and smooth and high-precision mechanism motion simulation can be completed in a universal web browser by calling the GPU computing power of a local computer of a user. Through a Web technical architecture, a user can access the system through a browser anytime and anywhere and conveniently carry out real-time parameter adjustment and multi-view dynamic observation, so that the movement mechanism and the working principle of a mechanism are intuitively and deeply understood. Besides, an open model sharing ecology is also constructed, different users are supported to upload and share an autonomously constructed mechanism model, and the movement process of a complex mechanism is vividly demonstrated through custom animations, so that the communication and propagation of technical knowledge are facilitated.

The invention relates to an enterprise intelligent modeling platform based on a big language model, and the platform comprises a model capability platform which is used for achieving the integration with a machine learning platform and a big data platform, achieving the configuration arrangement function of a big language model / big language model agent, and providing the support of data, basic algorithm capability and big language model capability for conversational modeling; and the application capability platform is used for realizing related functions of manual modeling, dialogue-type inherent model calling, dialogue-type intelligent modeling, dialogue retrieval of existing knowledge, model sharing and agent arrangement on the basis of the model capability platform. The method can be widely applied to the technical field of enterprise-level artificial intelligence application.

A secure clean room environment for sharing artificial intelligence (Al) models between multiple parties uses access control mechanisms and role-based access control for governing access to the Al models within the clean room environment. Isolated sandbox environments within the clean room environment allow authorized parties to test and use the Al models, monitor and log all activities performed within the clean room environment, and enforce usage controls on the Al models. The clean room leverages a retrieval-augmented generation (RAG) framework, multi-agent architecture, and confidential computing to enable safe, purpose-specific machine learning (ML) model sharing. A centralized knowledge graph and confidential virtual machines (VMs) create isolated environments where partners can securely access and query ML models.

The present application discloses a model sharing method, device, and storage medium for multiple edge servers, relating to the field of artificial intelligence technology. The method includes: issuing a baseline model to multiple edge servers; in response to receiving model training information reported by a first edge server for training the baseline model, obtaining reward values ​​corresponding to each of multiple second edge servers other than the first edge server; determining at least one shared edge server corresponding to the first edge server from the multiple second edge servers based on the reward values ​​corresponding to each of the multiple second edge servers and a preset reward threshold; determining address information corresponding to each of the at least one shared edge server, and issuing the address information corresponding to each of the at least one shared edge server to the first edge server, wherein the address information is used by the first edge server to send model sharing information to the shared edge servers. This method can improve the model accuracy of each edge server.

This invention provides a copyright protection method and system for federated learning models based on feature fusion. The method includes: Step 1, receiving watermark trigger samples and inputting them into a pre-established client model for training; Step 2, updating and aggregating the local shared model parameters and local classification layer parameters to obtain updated global model shared parameters and personalized classification layer parameters; extracting watermark features from the updated global model shared parameters based on a feature layer watermark embedding mechanism to obtain a global watermark; Step 3, performing multi-bit quantization on the personalized classification layer parameters to obtain a set of binary signatures; verifying copyright based on the black-box watermark and the global watermark, and identifying malicious clients based on the set of binary signatures. This invention can effectively curb the illegal copying and distribution of models, providing key technical support for promoting the healthy and sustainable development of data and model collaboration paradigms.

In one implementation, a method of decoding point cloud data for a point cloud is presented. A signal indicative of a number of sub-blocks, N1 is decoded, and a neural network is configured to have N1 sub-blocks. In particular, each sub-block of the N1 sub-blocks includes an upsampling function and at least a neural network layer, and each of the N1 sub-blocks is configured with the same neural network parameters. The point cloud data is decoded based on the neural network. At the encoder side, the signal indicative of N1 is encoded, and the neural network is configured to have N1 sub-blocks. In particular, each sub-block of the N1 sub-blocks includes a downsampling function and at least a neural network layer, and each of the N1 sub-blocks is configured with the same neural network parameters. The point cloud data is encoded based on the neural network.

The invention discloses a full-cycle asset security guarantee method and device for a nuclear power industrial model, and the method comprises the following steps: carrying out the preprocessing of a data set, optimization of model parameters, and setting of a corresponding defense strategy for data poisoning in a model compiling stage, i.e., a model generation stage; in the model transmission stage, a secure transmission protocol is set, a virtual private network is adopted, encryption transmission and uploading are carried out on the model, a secret key is managed and stored, and integrity verification is carried out on transmitted model data; in the model storage stage, encryption and access authority control are performed on the stored model data, and a data backup and recovery strategy is set for the stored model data; performing access authority control in a model operation calling stage, and performing operation calling on the model by adopting isolation and sandbox technologies; access authority control is carried out in the model sharing stage, and a corresponding application program interface is set for model sharing. According to the method, the asset safety of the nuclear power industrial model in the whole life cycle can be guaranteed.

The present disclosure relates to a model sharing method, an apparatus, a device, and a readable storage medium. Provided in the present disclosure is the model sharing method, used in a group management device of a first user equipment (UE) group, the first UE group comprising the group management device and a plurality of member UEs. The method comprises: receiving a model acquisition request sent by a first member UE, the first member UE being any one of the plurality of member UEs; and sending a model acquisition response to the first member UE, the model acquisition response being used for the first member UE to acquire a required target model. Provided is the model sharing method based on a UE group.

The invention discloses a model sharing method and device for multiple edge servers and a storage medium, and relates to the technical field of artificial intelligence. The method comprises the following steps: issuing a reference model to a plurality of edge servers; in response to the received model training information reported by the first edge server and used for training the reference model, obtaining reward values corresponding to a plurality of second edge servers except the first edge server in the plurality of edge servers; determining at least one shared edge server corresponding to the first edge server from the plurality of second edge servers according to the reward values corresponding to the plurality of second edge servers and a preset reward threshold value; address information corresponding to the at least one shared edge server is determined, the address information corresponding to the at least one shared edge server is issued to the first edge server, and the address information is used for the first edge server to send model sharing information to the shared edge server. According to the method, the model accuracy of each edge server can be improved.

The application particularly relates to a machine translation model crowdsourcing incremental learning method, which comprises the following steps: providing a translation model array composed of multiple machine translation models, introduction and details of each translation model; model owners can select whether to open crowdsourcing training, if the crowdsourcing training is opened, the system automatically preprocesses and cleans data uploaded by participants for participating in joint training; data quality scoring is performed, and a contribution value of the batch of data to model training is calculated according to the effective data amount and the quality score; after the participants confirm the contribution value, the model incremental training process is entered through a front-end interactive module; after the incremental training is completed, the model is updated to the machine translation array, the model related information is updated, and the opening is continued, and other participants can participate in the training. According to the method, data resources and the model after incremental training can be greatly shared through joint training and model sharing of multiple persons.

The invention relates to a federal prototype learning method fusing personalization and generalization in the Internet of Vehicles, and belongs to the technical field of mobile communication. For the problems of low model precision, poor fairness and fuzzy class boundary caused by data heterogeneity of the Internet of Vehicles, a double-layer architecture based on a DAG block chain is proposed to realize model security sharing, and the contribution degree distribution of a model is optimized by measuring the influence between vehicles through a dynamic hierarchical aggregation algorithm. And a federal prototype learning algorithm is designed to fuse generalization prototype and local prototype features. And a feature enhancement layer dynamic contrast learning mechanism is adopted, so that the model synchronously captures public knowledge, collaborative knowledge and personalized features, and the problem of unbalance between personalization and generalization is effectively solved. According to the method, the training efficiency, the classification precision and the fairness of the intelligent driving decision model are remarkably improved, and the method has outstanding advantages in the aspects of data safety, class boundary definition and asynchronous cooperative training.

The present invention relates to the field of artificial intelligence technology, and particularly to a management method, device, computer device and storage medium for machine learning models. The method includes determining a data set and a model corresponding to the configuration information of a process node; the process node is any process node in a process, and the configuration information includes execution conditions pointing to the data set and the model; establishing an association relationship between the process node, the data set and the model; wherein the association relationship can be used to indicate the data set and the model called when the process node is executed. By matching the data sets and models required when each process node is executed throughout the process and establishing an association relationship between each process node, the data set and the model to form a closed loop, the present invention can closely associate data production and data use in an environment where multiple models share data sets, improve the efficiency and effect of data use, and further improve the management efficiency of long-cycle processes.

The present disclosure describes methods for using information sharing based on a preconfigured AI / ML (artificial intelligence / machine learning)-based model in a wireless mobile communication system including a base station (e.g., gNB) and a mobile station (e.g., a UE). When an AI / ML model is applied to a radio access network, signaling model information supporting multiple UEs and / or applications can be heavily burdened when conditions applicable to ML vary across different UEs / applications. Therefore, model sharing can be performed adaptively between the network and the UE, reducing signaling overhead while improving model performance.

The application discloses a training method and device of a longitudinal federal neural network based on MPC. In order to solve the problem that the longitudinal federal split learning is vulnerable to model completion attack, the first n layers of the neural network adopt model fragmentation technology, and there is no complete bottom model, so the model completion attack cannot be initiated. In order to solve the efficiency problem caused by model fragmentation, after the n layers, the model fragmentation and the fragmentation of the calculation result are added by using the existing model sharing patent idea, that is, the training and reasoning mode of a single server is used to speed up the running speed. Since n is usually 1 or 2, the training and reasoning efficiency of the whole system is close to the effect of plaintext, and the security and privacy characteristics are maintained. For the scene of multiple label providers, the model and the model result can be split and combined, and the model and the result are split again to protect the privacy of the label. And the overall structure is still applicable to the application scene of split learning.

The embodiment of the invention provides a model sharing authorization method, a model reasoning method, electronic equipment and a storage medium, and the model sharing authorization method comprises the steps: obtaining a model sharing obtaining request sent by a network function consumer, and the model sharing obtaining request at least carries the consumer information of the network function consumer; sending the consumer information and a consumer authentication request to a security subject; and in response to authentication success information fed back by the security subject according to the consumer authentication request and the consumer information, sharing a model to the network function consumer. According to the embodiment of the invention, the security of model sharing in a communication network is realized, and the privacy leakage risk caused by model sharing is reduced.

The application provides a model sharing management method and device, equipment and a storage medium. The method comprises the following steps: obtaining a model demand of a to-be-subscribed model sent by a model user, wherein the model demand of the to-be-subscribed model comprises a type demand of the to-be-subscribed model and a deployment environment type demand; determining a target model and use data of the target model according to the model demand of the to-be-subscribed model and a published model, wherein the published model is a to-be-published model stored in a model sharing platform and matching audit results meet publishing requirements, and the matching audit results represent the results of matching and auditing of the model data of the to-be-published model and a publishing request sent by a model provider by the model sharing platform; and sending the use data of the target model to the model user. The method improves the efficiency of the user using the model and increases the universal applicability of the model sharing platform.

The invention discloses a model sharing service method, device, system and equipment, a medium and a program product, and relates to the technical field of digital networking, and the method applied to a model sharing platform comprises the steps: carrying out the preprocessing of task request information sent by a model demand node, and obtaining at least one piece of target task information; each piece of target task information in the at least one piece of target task information is matched with multiple private domain models in a pre-registered model library, a first private domain model corresponding to each piece of target task information is determined, and the at least one piece of target task information is sent to a model service node where the corresponding first private domain model is located; receiving a task processing result corresponding to the target task information, wherein the task processing result is obtained based on the first private domain model; and fusing task processing results, and sending response information to the model demand node. According to the embodiment of the invention, the problem of insufficient adaptive capability of model security of a large model in the related art can be solved.

Provided in the present application are a model sharing authorization method, a model inference method, an electronic device, and a storage medium. The model sharing authorization method comprises: acquiring a model sharing acquisition request sent by a network function consumer, wherein the model sharing acquisition request at least carries consumer information of the network function consumer; sending the consumer information and a consumer authentication request to a security main body; and in response to authentication success information, which is fed back by the security main body on the basis of the consumer authentication request and the consumer information, sharing a model with the network function consumer.