CN114491615A - Blockchain-based Asynchronous Vertical Federated Learning Fair Incentive Mechanism Method - Google Patents
Blockchain-based Asynchronous Vertical Federated Learning Fair Incentive Mechanism MethodDownload PDFInfo
- Publication number
- CN114491615A CN114491615ACN202111488605.6ACN202111488605ACN114491615ACN 114491615 ACN114491615 ACN 114491615ACN 202111488605 ACN202111488605 ACN 202111488605ACN 114491615 ACN114491615 ACN 114491615A
- Authority
- CN
- China
- Prior art keywords
- participants
- model
- information
- local
- participant
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/62—Protecting access to data via a platform, e.g. using keys or access control rules
- G06F21/6218—Protecting access to data via a platform, e.g. using keys or access control rules to a system of files or objects, e.g. local or distributed file system or database
- G06F21/6245—Protecting personal data, e.g. for financial or medical purposes
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/27—Replication, distribution or synchronisation of data between databases or within a distributed database system; Distributed database system architectures therefor
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F21/00—Security arrangements for protecting computers, components thereof, programs or data against unauthorised activity
- G06F21/60—Protecting data
- G06F21/64—Protecting data integrity, e.g. using checksums, certificates or signatures
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0241—Advertisements
- G06Q30/0251—Targeted advertisements
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Computing Systems (AREA)
- Databases & Information Systems (AREA)
- Business, Economics & Management (AREA)
- Evolutionary Computation (AREA)
- Bioethics (AREA)
- Computer Security & Cryptography (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- Computer Hardware Design (AREA)
- Strategic Management (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Molecular Biology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Computational Linguistics (AREA)
- Accounting & Taxation (AREA)
- Development Economics (AREA)
- Finance (AREA)
- Entrepreneurship & Innovation (AREA)
- Game Theory and Decision Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Economics (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
Description
Translated fromChinese技术领域technical field
本发明属于深度学习和隐私安全技术领域，具体涉及一种基于区块链的异步纵向联邦学习公平激励机制方法。The invention belongs to the technical field of deep learning and privacy security, and in particular relates to a blockchain-based asynchronous vertical federated learning fair incentive mechanism method.
背景技术Background technique
近年来，得益于大量数据资源和丰富的计算资源，深度学习凭借优异的性能在众多领域得到广泛的应用，例如人脸识别、自动驾驶和机器翻译等领域。数据隐私安全逐渐成为国家和地区关注的热点话题，并且制定了众多数据安全保护法规。数据隐私保护法规的限制使得企业之间的数据孤岛现象日益加剧，因此，探索一种保护隐私的机器学习技术正成为学术界和工业界关注的焦点话题。为解决上述数据孤岛问题，提出了一种保护数据隐私的分布式机器学习，其通过数据不出本地、模型出本地的方式实现分布式机器学习。In recent years, thanks to a large number of data resources and abundant computing resources, deep learning has been widely used in many fields with its excellent performance, such as face recognition, automatic driving, and machine translation. Data privacy security has gradually become a hot topic of national and regional attention, and many data security protection regulations have been formulated. The limitation of data privacy protection regulations has made the phenomenon of data silos among enterprises increasingly intensified. Therefore, exploring a privacy-preserving machine learning technology is becoming a focus of attention in academia and industry. In order to solve the above-mentioned data island problem, a distributed machine learning method to protect data privacy is proposed.
这种保护隐私的联邦学习技术被广泛应用于现实的广告精准投放业务中。在广告投放业务中根据参与方的数据分布的差异，联邦学习分为横向联邦学习、纵向联邦学习和联邦迁移学习。横向联邦学习适用于参与广告投放的企业之间样本id空间不同、特征空间相同的场景，例如来自不同地区的具有相同业务的电商之间合作训练机器学习模型。纵向联邦学习适用于参与广告投放的企业之间样本id空间相同、特征空间不同的场景，例如来自相同地区的两家不同功能的电商平台之间合作训练机器学习模型进行广告精准投放。联邦迁移学习适用于企业之间样本空间和特征空间重叠都较少的场景。由于来自不同领域的企业(具有相同特征空间、样本空间不同)之间合作日益紧密，目前纵向联邦学习受到学术界和工业界的广泛关注。This privacy-preserving federated learning technology is widely used in real-world advertising precision delivery business. In the advertising business, federated learning is divided into horizontal federated learning, vertical federated learning and federated transfer learning according to the differences in the data distribution of the participants. Horizontal federated learning is suitable for scenarios where the sample id space is different and the feature space is the same between enterprises participating in advertisement placement, for example, e-commerce companies from different regions with the same business cooperate to train machine learning models. Vertical federated learning is suitable for scenarios where the sample id space is the same and the feature space is different between enterprises participating in advertisement placement, such as the cooperation between two e-commerce platforms with different functions from the same region to train machine learning models for accurate advertisement placement. Federated transfer learning is suitable for scenarios where the sample space and feature space overlap between enterprises are small. Due to the increasingly close cooperation between enterprises from different fields (with the same feature space but different sample spaces), vertical federated learning has received extensive attention from academia and industry.
多家企业联合进行纵向联邦学习实现广告精准投放目的，该过程主要分为三步：首先在初始化过程中将参与方(包括参与方和主参与方)之间进行成员对齐，参与方利用本地模型在本地数据集上提取嵌入特征；然后发送给协作方进行聚合并利用顶部模型完成剩下的前向传播；最后根据求解概率向量和主任务真实的标签交叉损失函数，反向传播并更新顶部模型参数和参与方的本地模型参数。参与联合训练的企业中具有标签的参与方被定义为主参与方，只具有特征信息的参与方被定义为参与方。A number of enterprises jointly conduct vertical federated learning to achieve the purpose of accurate advertising. This process is mainly divided into three steps: first, in the initialization process, members are aligned between the participants (including the participants and the main participant), and the participants use the local model. Embedding features are extracted on the local dataset; then sent to the collaborators for aggregation and the top model is used to complete the remaining forward propagation; finally, the top model is back-propagated and updated according to the solution probability vector and the true label cross loss function of the main task parameters and local model parameters of the parties. In the enterprises participating in the joint training, the participants with tags are defined as the main participants, and the participants with only characteristic information are defined as participants.
然而，现有适用于广告精准投放业务的纵向联邦学习方法存在3个问题：1.纵向联邦学习中的算法是同步设计的，当部分设备宕机时难以维持广告投放业务的正常进行；2.纵向联邦学习中参与方的本地模型都由参与者本地进行维护，测试阶段难以由单一参与方发起测试任务，投放广告的业务唤醒受限；3.纵向联邦学习方法中由于缺少对广告投放精准度的激励机制，参与方的积极性会影响广告投放业务的整体性能。However, there are three problems with the existing vertical federated learning methods suitable for the accurate advertising business: 1. The algorithms in the vertical federated learning are designed synchronously, and it is difficult to maintain the normal operation of the advertising delivery business when some devices are down; 2. The local models of the participants in the vertical federated learning are all maintained locally by the participants, and it is difficult for a single participant to initiate a test task during the test phase, and the business wake-up of advertisement placement is limited; 3. In the vertical federated learning method, due to the lack of accuracy of advertisement placement The incentive mechanism of the participants will affect the overall performance of the advertising business.
去中心化的区块链技术为解决上述基于纵向联邦学习的广告精准投放的问题提供了一种解决思路。区块链技术是一种分布式账本，以透明的方式记录数据处理和查询过程，具有去中心化、可溯源以及不可篡改的特性。区块链可以分为公有链、联盟链、私有链三类。其中联盟链采用了一种混合的组网机制，对于网络内的节点具有部分的控制权。联盟链保留了公有链的部分透明、公开、防篡改等特征，并且具备权限管理、身份认证等特点，受到广泛关注，主要侧重于区块链在数据安全、可信认证等方面的应用。Decentralized blockchain technology provides a solution to the above-mentioned problem of accurate advertising based on vertical federated learning. Blockchain technology is a distributed ledger that records data processing and query processes in a transparent manner, with the characteristics of decentralization, traceability and immutability. Blockchain can be divided into three categories: public chain, consortium chain, and private chain. Among them, the alliance chain adopts a hybrid networking mechanism, which has partial control over the nodes in the network. The alliance chain retains some of the features of the public chain, such as transparency, openness, and tamper resistance, and has the characteristics of authority management and identity authentication.
鉴于纵向联邦学习在广告精准投放领域存在广泛的应用，这其中不仅仅包含企业之间的商业合作还涉及到物联网中的IOT设备的边缘计算，与此同时，其算法自身的隐私安全需要得到加固和强化，并且需要设计一套奖罚分明的激励机制促进纵向联邦学习中的参与方为联合训练提供更高质量的数据。因此结合区块链技术提出一种具有公平激励机制的异步纵向联邦学习方法具有重要价值。Given that vertical federated learning has a wide range of applications in the field of accurate advertising, it not only includes business cooperation between enterprises, but also involves edge computing of IOT devices in the Internet of Things. At the same time, the privacy and security of its algorithm itself needs to be obtained. Reinforcement and reinforcement, and need to design a set of incentives with clear rewards and punishments to promote participants in vertical federated learning to provide higher-quality data for joint training. Therefore, it is of great value to propose an asynchronous vertical federated learning method with a fair incentive mechanism combined with blockchain technology.
发明内容SUMMARY OF THE INVENTION
鉴于上述，本发明的目的在于提供一种基于区块链的异步纵向联邦学习公平激励机制方法，以提高广告投放的精准度。In view of the above, the purpose of the present invention is to provide a blockchain-based asynchronous vertical federated learning fair incentive mechanism method to improve the accuracy of advertisement placement.
为实现上述发明目的，本发明提供的技术方案为：In order to realize the above-mentioned purpose of the invention, the technical scheme provided by the present invention is:
一种基于区块链的异步纵向联邦学习公平激励机制方法，包括以下步骤：A blockchain-based asynchronous vertical federated learning fair incentive mechanism method, including the following steps:
步骤1，具有广告推荐需求的参与方注册申请，任务协调者通过区块链部署纵向联邦学习任务，参与方之间数据对齐和依据纵向联邦学习任务部署用于提取样本数据的嵌入表示信息的本地模型；Step 1: Participants with advertising recommendation requirements apply for registration, the task coordinator deploys the vertical federated learning task through the blockchain, the data alignment between the participants and the localization of the embedded representation information for extracting sample data according to the vertical federated learning task are deployed. Model;
步骤2，参与方利用本地数据训练本地模型，并上传样本数据的嵌入表示信息至区块链；
步骤3，选择所有参与方中的一者为主参与方，主参与方依据时间戳从区块链收集满足预设数量的完整嵌入表示信息后进行嵌入表示信息聚合，并利用聚合的嵌入表示更新用于广告推荐的顶部模型和本地模型的梯度信息，为本地模型的梯度信息生成新区块保存；Step 3: Select one of all participants as the main participant. The main participant collects the complete embedded representation information from the blockchain according to the timestamp and aggregates the embedded representation information, and uses the aggregated embedded representation to update. The gradient information of the top model and the local model used for advertisement recommendation, and generate a new block for the gradient information of the local model to save;
步骤4，验证委员会对新区块的合法性进行验证，并广播通过验证的新区块，同步更新区块链中的账本信息，参与方从账本信息中下载本地模型的梯度信息进行下一轮本地训练；Step 4: The verification committee verifies the legitimacy of the new block, broadcasts the verified new block, updates the ledger information in the blockchain synchronously, and the participants download the gradient information of the local model from the ledger information for the next round of local training ;
步骤5，纵向联邦学习任务结束时，参与方和主参与方上传本地模型和顶部模型至新区块保存；Step 5: When the vertical federated learning task ends, the participants and the main participant upload the local model and the top model to a new block for saving;
步骤6，验证委员会对参与方的数据质量贡献度进行评分，并根据贡献度得分为参与方分配激励值。Step 6: The verification committee scores the data quality contribution of the participants, and allocates incentive values to the participants according to the contribution scores.
在一个实施例中，参与方注册申请时，将自身数据集大小和算力情况及网络通信速率作为注册信息上传；In one embodiment, when a participant applies for registration, upload its own data set size, computing power and network communication rate as registration information;
任务协调者通过区块链部署纵向联邦学习任务时，任务协调者分配签名的公钥和私钥，此时参与方具有支出激励值的数字签名，任务协调者在区块链中创建创世块，创世块包含纵向联邦学习任务信息；When the task coordinator deploys the vertical federated learning task through the blockchain, the task coordinator assigns the public key and private key of the signature. At this time, the participants have the digital signature of the spending incentive value, and the task coordinator creates the genesis block in the blockchain. , the genesis block contains vertical federated learning task information;
参与方与附近的计算节点进行匹配通信，1个计算节点同时匹配多个参与方参与纵向联邦学习任务，参与方通过计算节点从创世块下载纵向联邦学习任务信息进行本地部署。Participants communicate with nearby computing nodes. One computing node matches multiple participants to participate in the vertical federated learning task at the same time. The participants download the vertical federated learning task information from the genesis block through the computing node for local deployment.
在一个实施例中，参与方之间通过基于RSA加密技术和哈希算法完成数据秘密对齐匹配。In one embodiment, the secret alignment and matching of data is completed between the participants based on RSA encryption technology and hash algorithm.
在一个实施例中，参与方利用本地数据训练部署的本地模型，完成样本数据的1个轮次的前向传播以获得样本数据的嵌入表示信息，参与方将带有签名的嵌入表示信息、本地资源运行消耗、以区块链事务的形式上传给关联的计算节点。In one embodiment, the participants use the local data to train the deployed local model, and complete one round of forward propagation of the sample data to obtain the embedded representation information of the sample data. Resource running consumption is uploaded to associated computing nodes in the form of blockchain transactions.
在一个实施例中，计算节点收到带有签名的嵌入表示信息，验证签名的合法性并发布事务，主参与方监听通过计算节点收集参与方发布的事务，获取参与方发布的嵌入表示信息，主参与方监听机制包括监听参与方上传事务的时间戳以及嵌入表示信息的完整性，当主参与方监听到前n个参与方完成嵌入表示信息的上链，聚合拼接n个参与方的嵌入表示信息，形成聚合的嵌入表示信息；与此同时，监听的计数器重置为0，一旦监听到新上链的嵌入表示信息计数器加1。In one embodiment, the computing node receives the embedded representation information with the signature, verifies the validity of the signature and publishes the transaction, the main participant listens to the transaction published by the participating parties collected through the computing node, and obtains the embedded representation information published by the participants, The monitoring mechanism of the main participant includes monitoring the timestamp of the transaction uploaded by the participant and the integrity of the embedded representation information. When the main participant monitors that the first n participants have completed the uploading of the embedded representation information, the embedded representation information of the n participants is aggregated and spliced. , forming the aggregated embedded representation information; at the same time, the monitoring counter is reset to 0, and the counter is incremented by 1 once the newly uploaded embedded representation information is monitored.
在一个实施例中，主参与方利用聚合的嵌入表示更新用于广告推荐的顶部模型和本地模型的梯度信息时，将聚合的嵌入表示更新输入顶部模型，经过计算得到输出向量，将该输出向量经过分类器映射输出预测概率向量，通过该预测概率向量与真实标签的交叉熵作为损失函数，来更新顶部模型，同时生成各本地模型的梯度信息。In one embodiment, when the main participant uses the aggregated embedding representation to update the gradient information of the top model and the local model for advertisement recommendation, the aggregated embedding representation is updated into the top model, and an output vector is obtained through calculation, and the output vector The predicted probability vector is output through the classifier mapping, and the cross-entropy between the predicted probability vector and the real label is used as the loss function to update the top model, and the gradient information of each local model is generated at the same time.
在一个实施例中，参与方从账本信息中下载新顶部模型的更新梯度信息，在本地模型实现反向传播算法，并利用更新梯度信息更新本地模型参数，完成本地训练。In one embodiment, the participant downloads the updated gradient information of the new top model from the ledger information, implements the back-propagation algorithm in the local model, and uses the updated gradient information to update the local model parameters to complete local training.
在一个实施例中，验证委员会对参与方的数据质量贡献度采用以下两种打分机制：In one embodiment, the verification committee adopts the following two scoring mechanisms for the data quality contribution of the participants:
互信息评测打分机制，计算真实标签与参与方上传的嵌入表示信息之间的互信息数值，该互信息数值作为数据质量贡献度评分的一部分；The mutual information evaluation scoring mechanism calculates the mutual information value between the real label and the embedded representation information uploaded by the participants, and the mutual information value is used as part of the data quality contribution score;
嵌入缺失验证打分机制，针对每个参与方，将参与方的嵌入表示信息赋值为全0向量后，将该全0向量与所有其他参与方的嵌入表示信息拼接后，输入至顶部模型得到预测结果，并根据预测结果和真实标签计算参与方的准确度，依据准确度越低赋予越高得分的方式为参与方评分，该评分作为数据质量贡献度评分的另一部分；Embedding missing verification scoring mechanism, for each participant, after assigning the participant's embedded representation information as an all-zero vector, splicing the all-zero vector with the embedding representation information of all other participants, and inputting it to the top model to get the prediction result , and calculate the accuracy of the participants according to the prediction results and the real labels, and score the participants according to the method that the lower the accuracy, the higher the score, which is another part of the data quality contribution score;
两个部分的数据质量贡献度评分之和作为参与方的总评分。The sum of the data quality contribution scores of the two parts is used as the total score of the participants.
在一个实施例中，所述依据准确度越低赋予越高得分的方式为参与方评分，包括：In one embodiment, scoring the participants according to a manner of assigning a higher score with a lower accuracy includes:
对每个参与方的准确度进行排序，然后按照以下公式计算每个参与方的得分：Rank the accuracy of each participant, then calculate each participant's score according to the following formula:
其中，Score2表示得分，x表示参与方的准确度按照由低到高排序的序号。Among them, Score2 represents the score, and x represents the sequence number of the accuracy of the participants sorted from low to high.
在一个实施例中，所述方法还包括：参与方纵向联邦学习得到的顶部模型的验证推理，包括：In one embodiment, the method further includes: the verification reasoning of the top model obtained by the longitudinal federated learning of the participants, including:
在单一模式验证推理中，参与方从新区块中下载顶部模型，将本地数据输入至本地模型，得到嵌入表示信息，然后嵌入表示信息输入至顶部模型中，得到顶部模型的预测结果，根据该预测结果对本地模型和顶部模型验证推理；In the single mode verification inference, the participants download the top model from the new block, input the local data into the local model, get the embedded representation information, and then input the embedded representation information into the top model to obtain the prediction result of the top model, according to the prediction The results validate the inference against the local model and the top model;
在多方模式验证推理中，多个参与方从新区块中同时下载顶部模型，每个参与方将本地数据输入至各自的本地模型，得到嵌入表示信息，然后各自的嵌入表示信息输入至顶部模型中，得到顶部模型的预测结果，根据该预测结果对本地模型和顶部模型验证推理。In the multi-party mode verification inference, multiple participants download the top model from the new block at the same time, each participant inputs local data into their own local models, and obtains embedded representation information, and then their respective embedded representation information is input into the top model. , get the prediction result of the top model, and verify the reasoning on the local model and the top model according to the prediction result.
本发明的技术构思为：1.纵向联邦学习中不同参与方设备算力不同导致排队更新或者可能存在宕机情况使得训练难以进行，纵向联邦学习的同步更新机制难以应对上述问题，该技术方案引入区块链账本存储技术，将参与方的嵌入表示实时上传到区块上，主参与方一旦收集到设定阈值的参与方嵌入就开始执行当前轮次的本地模型更新和顶部模型更新协议，这种异步处理机制减少参与方的排队等待更新时间，并且避免设备宕机造成的训练终止。2.纵向联邦学习完成训练模型后，参与方将本地模型以及主参与方将顶部模型进行上链进行保存，避免模型信息被恶意篡改，开发两种测试模型，即：单一模式和多方模式来便于参与者使用训练完成的模型进行推理主任务标签。3.纵向联邦学习中的激励机制由验证委员会进行评审分配激励值。激励机制结合互信息评测机制和嵌入缺失验证机制来衡量参与方数据的质量。其中互信息评测利用现有的互信息评测机制为参与方的数据质量进行打分。The technical idea of the present invention is as follows: 1. In vertical federated learning, the different computing powers of different participants’ devices lead to queuing for updating or there may be downtime, making it difficult to perform training, and the synchronous updating mechanism of vertical federated learning is difficult to deal with the above problems. This technical solution introduces The blockchain ledger storage technology uploads the participant's embedded representation to the block in real time. Once the main participant collects the participant's embedding of the set threshold, it starts to execute the current round of local model update and top model update protocol. An asynchronous processing mechanism reduces participants' queuing time for updates, and avoids training termination caused by equipment downtime. 2. After the vertical federated learning completes the training model, the participants will save the local model and the main participant will upload the top model to the chain to avoid malicious tampering of model information, and develop two test models, namely: single mode and multi-party mode to facilitate Participants use the trained model to infer the main task labels. 3. The incentive mechanism in vertical federated learning is reviewed and distributed by the verification committee. The incentive mechanism combines the mutual information evaluation mechanism and the embedded missing verification mechanism to measure the quality of the participants' data. The mutual information evaluation uses the existing mutual information evaluation mechanism to score the data quality of the participants.
与现有技术相比，本发明具有的有益效果至少包括：Compared with the prior art, the beneficial effects of the present invention at least include:
本发明提供的一种具有公平激励机制的异步纵向联邦学习方法，在用于广告精准推荐时，通过引入区块链技术的账本存储技术，提出弹性聚合机制使得纵向联邦学习支持异步更新，提高计算运行效率，有效避免设备宕机情况。通过将训练完成的本地模型和顶部模型上链，一方面有效避免模型信息被恶意篡改或丢失，另一方面利于参与者在推理阶段实时调用模型进行测试。为此，本发明开发两种测试模式：单一模式和多方模式。此外，本发明提出互信息评测机制和嵌入缺失验证机制有效评估纵向联邦学习参与者的数据质量，从而促进分配机制的公平性，最后得到的顶部模型在用于广告推荐时，提成了广告投放的精准度。The invention provides an asynchronous vertical federated learning method with a fair incentive mechanism. When it is used for accurate advertisement recommendation, by introducing the ledger storage technology of blockchain technology, an elastic aggregation mechanism is proposed to enable vertical federated learning to support asynchronous update and improve computing efficiency. Operation efficiency, effectively avoid equipment downtime. By uploading the trained local model and the top model on the chain, on the one hand, the model information is effectively prevented from being maliciously tampered with or lost, and on the other hand, it is beneficial for participants to call the model in real time for testing in the inference stage. To this end, the present invention develops two test modes: a single mode and a multi-party mode. In addition, the present invention proposes a mutual information evaluation mechanism and an embedded missing verification mechanism to effectively evaluate the data quality of the vertical federated learning participants, thereby promoting the fairness of the distribution mechanism. precision.
附图说明Description of drawings
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图做简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动前提下，还可以根据这些附图获得其他附图。In order to illustrate the embodiments of the present invention or the technical solutions in the prior art more clearly, the following briefly introduces the accompanying drawings used in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only These are some embodiments of the present invention. For those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative efforts.
图1是本发明实施例提供的具有公平激励机制的异步纵向联邦学习方法的示意图。FIG. 1 is a schematic diagram of an asynchronous vertical federated learning method with a fair incentive mechanism provided by an embodiment of the present invention.
图2是本发明实施例提供的具有公平激励机制的异步纵向联邦学习方法异步更新的示意图。FIG. 2 is a schematic diagram of asynchronous update of an asynchronous vertical federated learning method with a fair incentive mechanism provided by an embodiment of the present invention.
图3是本发明实施例提供的具有公平激励机制的异步纵向联邦学习方法的流程图。FIG. 3 is a flowchart of an asynchronous vertical federated learning method with a fair incentive mechanism provided by an embodiment of the present invention.
具体实施方式Detailed ways
为使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例对本发明进行进一步的详细说明。应当理解，此处所描述的具体实施方式仅仅用以解释本发明，并不限定本发明的保护范围。In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention, and do not limit the protection scope of the present invention.
鉴于纵向联邦学习在实现商业场景中的广泛应用，尤其是现有的广告精准投放业务，然而其缺乏公平有效的激励机制且安全性无法得到保证，实施例提供了一种基于区块链的异步纵向联邦学习公平激励机制方法，促进企业之间参与广告推荐业务的公平性和可靠性，同时提高广告投放的精准度。该方法主要分为3个技术解决方案：1.引入区块链的去中心化技术改变纵向联邦学习的同步训练机制，提出一种弹性聚合机制从而支持纵向联邦学习的异步更新；2.引入区块链使得参与方模型上链，为单一用户在推理阶段随时调用模型提供解决方案；3.激励机制结合互信息评测机制和嵌入缺失验证机制来衡量参与方数据的质量提出一种纵向联邦学习的公平的激励机制，促进纵向联邦学习参与方的分配公平。In view of the wide application of vertical federated learning in the realization of commercial scenarios, especially the existing accurate advertising business, but it lacks a fair and effective incentive mechanism and the security cannot be guaranteed, the embodiment provides a blockchain-based asynchronous The vertical federated learning fair incentive mechanism method promotes the fairness and reliability of enterprises participating in the advertising recommendation business, and at the same time improves the accuracy of advertising. This method is mainly divided into 3 technical solutions: 1. Introducing the decentralization technology of blockchain to change the synchronous training mechanism of vertical federated learning, and proposing an elastic aggregation mechanism to support the asynchronous update of vertical federated learning; 2. Introducing district The blockchain enables the participant model to be on the chain, providing a solution for a single user to call the model at any time in the inference stage; 3. The incentive mechanism combines the mutual information evaluation mechanism and the embedded missing verification mechanism to measure the quality of the participant data and proposes a vertical federated learning system. A fair incentive mechanism to promote fair distribution among vertical federated learning participants.
图1本发明实施例提供的具有公平激励机制的异步纵向联邦学习方法的示意图。图3是本发明实施例提供的具有公平激励机制的异步纵向联邦学习方法的流程图。如图1和图3所示，实施例提供的基于区块链的异步纵向联邦学习公平激励机制方法，包括以下步骤：FIG. 1 is a schematic diagram of an asynchronous vertical federated learning method with a fair incentive mechanism provided by an embodiment of the present invention. FIG. 3 is a flowchart of an asynchronous vertical federated learning method with a fair incentive mechanism provided by an embodiment of the present invention. As shown in Figure 1 and Figure 3, the blockchain-based asynchronous vertical federated learning fair incentive mechanism method provided by the embodiment includes the following steps:
1)初始化阶段。1) Initialization stage.
初始化阶段包括具有广告精准推荐需求的参与方进行注册申请、任务协调者分配公钥和私钥、任务协调者创建本地模型结构参数信息和顶部模型信息。此外，参与方之间还应进行数据对齐和模型部署工作。The initialization phase includes the registration application of the participants who have the requirement of accurate advertisement recommendation, the assignment of public and private keys by the task coordinator, and the creation of local model structure parameter information and top model information by the task coordinator. In addition, data alignment and model deployment should be done between the parties.
1.1)参与方申请注册。1.1) Participants apply for registration.
具有用户信息资源的参与方进行身份注册和身份验证，参与方将自身的数据集大小和算力情况及网络通信速率作为注册信息上传。所有参与方中选择一个作为主参与方，一般以任务发起者作为主参与方，主参与方不仅要进行本地模型的训练，还要进行顶部模型的训练。Participants with user information resources perform identity registration and authentication, and the participants upload their own data set size, computing power, and network communication rate as registration information. One of all participants is selected as the main participant. Generally, the task initiator is the main participant. The main participant must not only train the local model, but also train the top model.
1.2)任务协调者分配签名的公钥和私钥并创建创世块。1.2) The task coordinator assigns the signed public and private keys and creates the genesis block.
区块链中的任务协调者往往由一位公信力的机构担任，例如政府信管部门。任务协调者分配签名的公钥和私钥，此时参与方具有支出激励值的数字签名。任务协调者创建的创世块，包括参与方的本地模型结构flocal(·)和主参与方的顶部模型结构ftop(·)以及其初始化参数信息{θ1,θ2,…,θm,θt}；训练轮回数量E；初始化激励值数值{C1,C2,…,Cm,Ct}；模型学习的学习率η。The task coordinator in the blockchain is often served by a credible institution, such as the government information management department. The task coordinator assigns the signed public and private keys, at which point the participants have a digital signature of the spending incentive value. The genesis block created by the task coordinator includes the participant's local model structure flocal ( ) and the main participant's top model structure ftop ( ) and its initialization parameter information {θ1 ,θ2 ,…,θm , θt }; number of training rounds E; initialized excitation values {C1 , C2 ,..., Cm , Ct }; learning rate η for model learning.
1.3)参与方与计算节点匹配。1.3) Participants are matched with computing nodes.
参与方与附近的计算节点进行匹配通信，计算节点往往对应边缘节点，具有一定的计算资源和通信资源，1个计算节点可以同时匹配多个参与方参与纵向联邦学习训练。Participants communicate with nearby computing nodes. Computing nodes often correspond to edge nodes and have certain computing and communication resources. One computing node can simultaneously match multiple participants to participate in vertical federated learning training.
1.4)参与方部署本地模型。1.4) Participants deploy the local model.
参与方通过计算节点从创世块下载本地模型flocal(·)和初始化本地模型 flocal(·)的参数{θ1,θ2,…,θm}，随着将此本地模型部署到参与方设备，其中参与方的设备具有一定的计算资源和通信资源。The participant downloads the local model flocal ( ) from the genesis block through the computing node and initializes the parameters {θ1 , θ2 ,…, θm } of the local model flocal ( ), and deploys this local model to the participant The device of the participant, in which the device of the participant has certain computing resources and communication resources.
1.5)参与方之间数据秘密对齐匹配。1.5) Data secret alignment and matching between participants.
基于RSA加密技术和哈希算法完成参与方之间的数据秘密对齐。Based on RSA encryption technology and hash algorithm, the data secret alignment between the participants is completed.
2)参与方本地模型训练并上传阶段。2) Participant's local model training and uploading stage.
2.1)参与方本地模型前向传播。2.1) Participant's local model forward propagation.
参与方i利用本地数据Xi在部署完成的模型fi(·)完成1个轮次的前向传播并获得嵌入表示信息Ei，其中Ei＝fi(Xi)。Participant i uses local data Xi to complete 1 round of forward propagation in the deployed model fi (·) and obtains embedded representation information Ei , where Ei= fi( Xi) .
2.2)参与方将嵌入表示Ei、本地资源运行消耗Tcon、数字签名以区块链事务的形式上传给关联的计算节点。2.2) The participant uploads the embedded representation Ei , the local resource running consumption Tcon , and the digital signature to the associated computing node in the form of a blockchain transaction.
3)参与方顶部模型前向传播阶段。3) The forward propagation stage of the top model of the participant.
3.1)计算节点收到带有签名的嵌入表示Ei后，验证签名的合法性并发布事务。3.1) After receiving the embedded representation Ei with the signature, the computing node verifies the validity of the signature and publishes the transaction.
3.2)主参与方收集计算节点发布的事务实现软聚合。3.2) The main participant collects the transactions published by the computing nodes to achieve soft aggregation.
主参与方监听通过计算节点收集参与方发布的事务，获取参与方发布的嵌入表示信息。主参与方监听机制包括监听参与方上传事务的时间戳以及嵌入表示的完整性，当主参与方监听到前n个参与方完成嵌入的上链，聚合拼接n个参与方的嵌入表示，形成初始嵌入表示Einit。与此同时，监听的计数器重置为0，一旦监听到新上链的嵌入计数器加1。The main participant listens to collect the transactions published by the participants through the computing nodes, and obtains the embedded representation information published by the participants. The monitoring mechanism of the main participant includes monitoring the timestamp of the uploading transaction of the participant and the integrity of the embedded representation. When the main participant monitors the uploading of the first n participants to complete the embedding, it aggregates and splices the embedded representations of the n participants to form the initial embedding. Represents Einit . At the same time, the listening counter is reset to 0, and the embedded counter is incremented by 1 once a new up-chain is monitored.
图2是本发明实施例提供的一种具有公平激励机制的异步纵向联邦学习方法异步更新的示意图。如图2所示，由于参与方通过监听参与方上传事务的时间戳以及嵌入表示的完整性，在主参与方监听到前n个参与方完成嵌入的上链，聚合拼接n个参与方的嵌入表示，该过程实现了多个参与方的同一轮次训练的本地模型的嵌入表示的异步聚合更新。FIG. 2 is a schematic diagram of asynchronous update of an asynchronous vertical federated learning method with a fair incentive mechanism provided by an embodiment of the present invention. As shown in Figure 2, since the participants monitor the timestamp of the transaction uploaded by the participants and the integrity of the embedded representation, the main participant monitors the first n participants to complete the embedding on the chain, and aggregates and splices the embeddings of the n participants. Representation, the process enables asynchronous aggregated updates of embedding representations of local models trained in the same round of multiple participants.
3.3)主参与方利用顶部模型完成前向传播。3.3) The main participant uses the top model to complete the forward propagation.
主参与方利用本地聚合产生的初始嵌入表示Einit输入顶部模型ftop(·) 完成前向传播过程，获得输出向量Linit，并经由1个Softmax层获得最终模型预测概率向量Lpredict。其中Linit＝ftop(Einit)，The main participant uses the initial embedding representation Einit generated by local aggregation to input the top model ftop (·) to complete the forward propagation process, obtain the output vector Linit , and obtain the final model prediction probability vector Lpredict through a Softmax layer. where Linit =ftop (Einit ),
其中，zj表示第j个类标模型预测结果。Among them, zj represents the prediction result of the jth class label model.
3.4)主参与方利用顶部模型获得的预测向量Lpredict和真实标签Lright计算交叉熵损失函数Lutility。3.4) The main participant uses the prediction vector Lpredict obtained by the top model and the true label Lright to calculate the cross entropy loss function Lutility .
其中，yi,j为真实标签，y′i,j为预测标签。Among them, yi,j are the real labels, and y′i,j are the predicted labels.
3.5)主参与方在顶部模型运算反向传播算法，更新顶部模型，并生成本地模型的梯度信息，将梯度信息上传形成新区块。3.5) The main participant operates the back-propagation algorithm on the top model, updates the top model, generates gradient information of the local model, and uploads the gradient information to form a new block.
主参与方运行反向传播算法，利用主任务损失函数求顶部模型的梯度 gt。其中，
3.6)验证委员会对新区块的合法性进行验证，并广播通过验证的新区块，同步更新区块链中的账本信息。3.6) The verification committee verifies the legitimacy of the new block, broadcasts the new block that has passed the verification, and updates the ledger information in the blockchain synchronously.
3.7)参与方由计算节点载入事务，更新本地模型。3.7) The participants load the transaction from the computing node and update the local model.
参与方i监听到主参与方更新的事务之后，下载更新梯度信息gt，在本地模型实现反向传播算法获得梯度信息gi并且更新本地模型参数θi。其中
3.8)由此进行，直至模型收敛完成纵向联邦学习的训练。3.8) Continue from this until the model converges and completes the training of vertical federated learning.
通常按照参与者注册阶段默认的训练轮回完成对本地模型和顶部模型的训练。The training of the local model and the top model is usually done according to the default training round of the participant registration phase.
4)纵向联邦学习模型上传阶段。4) Vertical federated learning model upload stage.
4.1)完成训练的参与方上传本地模型到节点形成事务。4.1) Participants who complete the training upload the local model to the node to form a transaction.
主参与方和参与方通过匹配的计算节点上传各自完成训练的模型。发布的事务包括完整的模型结构和模型参数信息。The main participant and the participant upload their respective trained models through matching computing nodes. Published transactions include complete model structure and model parameter information.
4.2)主参与方收集事务并上传顶部模型至区块链，形成新区块。4.2) The main participant collects transactions and uploads the top model to the blockchain to form a new block.
4.3)经过验证委员会对新区块的合法性进行验证，并广播通过验证的新区块，该区块负责记录完成训练的纵向联邦学习模型，同时更新区块链的账本信息。4.3) After the verification committee verifies the legitimacy of the new block, and broadcasts the new block that has passed the verification, the block is responsible for recording the vertical federated learning model that has completed the training, and at the same time updating the ledger information of the blockchain.
5)激励机制。5) Incentive mechanism.
完成纵向联邦学习训练的参与方由验证委员会对参与方的数据质量贡献度进行评分，并根据贡献度的得分为参与方分配激励值。打分机制由两部分组成：互信息评测机制和嵌入缺失验证机制。Participants who have completed the vertical federated learning training are scored by the verification committee on the data quality contribution of the participants, and incentive values are allocated to the participants according to the contribution scores. The scoring mechanism consists of two parts: the mutual information evaluation mechanism and the embedded missing verification mechanism.
5.1)互信息评测机制打分。5.1) Mutual information evaluation mechanism to score.
验证委员会在具有少量样本的基础上，利用互信息测试工具计算真实标签Y和参与方上传嵌入表示Ei之间的互信息数值，形成贡献度评分 Score1。该互信息数值越大表示参与方的嵌入和真实标签对应更加紧密，这表明参与方的数据具有更高的贡献度。On the basis of a small number of samples, the verification committee uses the mutual information testing tool to calculate the mutual information value between the real label Y and the participant uploading embedded representation Ei to form the contribution score Score1. The larger the mutual information value is, the more closely the participant's embedding and the true label correspond, which indicates that the participant's data has a higher contribution.
Score1＝H(Y|Ei)，Score1=H(Y|Ei ),
其中，H(x|y)表示变量x和y的联合交叉熵，Score1表示通过互信息评测机制打分得到的评分。Among them, H(x|y) represents the joint cross-entropy of the variables x and y, and Score1 represents the score obtained by the mutual information evaluation mechanism.
5.2)嵌入缺失验证机制打分。5.2) Embedding missing verification mechanism scoring.
基于随机嵌入缺失的方式组合本地模型和顶部模型预测输出的准确率来判断参与方的数据贡献度。针对每个参与方i，将参与方i的嵌入表示信息赋值为全0向量后，将该全0向量与所有其他参与方的嵌入表示信息拼接后，输入至顶部模型得到预测结果，并根据预测结果和真实标签计算参与方i的准确度，将全0向量的参与方i的ID信息和当前准确度读入字典进行存储。The accuracy of the prediction output of the local model and the top model is combined based on the random embedding missing to judge the data contribution of the participants. For each participant i, after assigning the embedded representation information of participant i as an all-zero vector, after splicing the all-zero vector with the embedded representation information of all other participants, input it to the top model to obtain the prediction result, and according to the prediction The result and the real label are used to calculate the accuracy of the participant i, and the ID information of the participant i and the current accuracy of the all-zero vector are read into the dictionary for storage.
字典按照准确率进行从小到大进行排序，准确度最低的被赋予最高的得分值。The dictionary is sorted from small to large according to the accuracy, and the lowest accuracy is given the highest score value.
其中，Score2表示得分，x表示为参与者在字典中按照准确度由低到高排序的次序号。Among them, Score2 represents the score, and x represents the order number of the participants in the dictionary according to the order of accuracy from low to high.
5.3)综合打分，分配激励值。5.3) Comprehensive scoring and distribution of incentive value.
参与方i的贡献度得分Scorei表示为：Scorei＝Score1+Score2，验证委员会按照该得分分配激励值。The contribution score Scorei of the participant i is expressed as: Scorei =Score1+Score2, and the verification committee allocates the incentive value according to the score.
6)纵向联邦学习的推理阶段。6) The inference stage of longitudinal federated learning.
6.1)选择参与模式：单一模式、多方模式。6.1) Select participation mode: single mode, multi-party mode.
首先根据参与推理标签阶段参与方的数量选择参与模式，当推理阶段参与方的数量仅仅为1时，参与方选择单一模式；当推理阶段参与方的数量大于1时，参与方选择多方模式。Firstly, the participation mode is selected according to the number of participants in the inference tag stage. When the number of participants in the inference stage is only 1, the participant chooses the single mode; when the number of participants in the inference stage is greater than 1, the participant chooses the multi-party mode.
6.2)单一模式中，参与方从新区块下载训练完成的顶部模型，每个参与方将本地数据输入至本地模型，得到嵌入表示信息，然后嵌入表示信息输入至顶部模型中，得到顶部模型的预测结果，根据该预测结果对本地模型和顶部模型验证推理。6.2) In the single mode, the participants download the trained top model from the new block, each participant inputs the local data into the local model, gets the embedded representation information, and then inputs the embedded representation information into the top model to get the prediction of the top model As a result, inferences are validated against the local model and the top model based on this prediction.
多方模式中，多个参与方从新区块中同时下载顶部模型，每个参与方将本地数据输入至各自的本地模型，得到嵌入表示信息，然后各自的嵌入表示信息输入至顶部模型中，得到顶部模型的预测结果，根据该预测结果对本地模型和顶部模型验证推理。In the multi-party mode, multiple participants download the top model from the new block at the same time, and each participant inputs local data into their own local models to obtain embedded representation information, and then input their respective embedded representation information into the top model to get the top model. The prediction of the model, against which the inference is validated against the local model and the top model.
在单一模式下，参与方j具有测试样本特征Xj，从区块链上下载测试模型信息，将测试样本特征输入测试模型获得预测标签Pj。该过程可表示为：In the single mode, the participant j has the test sample feature Xj , downloads the test model information from the blockchain, and inputs the test sample feature into the test model to obtain the predicted label Pj . The process can be expressed as:
Pj＝concat(f1(Xj),…,fm-n(Xj),ft(Xj))，Pj =concat(f1 (Xj ),...,fmn (Xj ),ft (Xj )),
其中，concat(·)表示某一维度进行直接拼接，Xj为参与方的测试样本数据特征，f1,…,fn表示参与方的本地模型，ftop表示主参与方的顶部模型。Among them, concat(·) means direct splicing of a certain dimension, Xj is the test sample data characteristic of the participant, f1 ,...,fn represents the local model of the participant, and ftop represents the top model of the main participant.
在广告推荐业务中，通过上述基于区块链的异步纵向联邦学习公平激励机制方法构建的顶部模型用于广告推荐，广告推荐业务参与方的本地数据集往往是部分广告媒体信息数据集，例如Criteo和Avazu数据集。在用于广告推荐业务的纵向联邦学习系统中，本地模型通常采用5层全联接网络层，每层网络层具有非线性激活函数ReLU，本地模型输入维度对应参与方具有的数据集特征维度，输出的维度通常设置为64；顶部模型通常采用2层非线性全联接网络层，顶部模型的输入维度为64*n，其中n为参与方的数量，输出的维度对应数据集标签的类别数。广告推断业务实际场景的训练过程中，参与方利用本地模型在本地广告媒体数据集完成前向传播输出嵌入特征，主参与方聚合嵌入特征后完成在顶部模型的前向传播，计算损失函数后反向传播并完成模型参数的更新。这个过程中会将模型信息形成事务发布在区块链。广告推断业务实际场景的测试过程中，如上文所述存在两种测试模型：单一模式和多方模式。单一模式是某个参与方进行下载模型进行测试，多方模式是多个参与方共同下载模型进行测试。In the advertising recommendation business, the top model constructed by the above-mentioned blockchain-based asynchronous vertical federated learning fair incentive mechanism method is used for advertising recommendation. The local data sets of advertising recommendation business participants are often partial advertising media information data sets, such as Criteo and the Avazu dataset. In the vertical federated learning system for advertising recommendation business, the local model usually adopts 5 layers of fully connected network layers, each network layer has a nonlinear activation function ReLU, the input dimension of the local model corresponds to the data set feature dimension possessed by the participants, and the output The dimension of is usually set to 64; the top model usually adopts 2 layers of nonlinear fully connected network layers, the input dimension of the top model is 64*n, where n is the number of participants, and the output dimension corresponds to the number of categories of the dataset labels. During the training process of the actual scenario of the advertising inference business, the participants use the local model to complete the forward propagation and output the embedded features in the local advertising media data set. The main participant aggregates the embedded features and completes the forward propagation in the top model. to propagate and complete the update of the model parameters. In this process, the model information will be formed into a transaction and published on the blockchain. In the test process of the actual scenario of advertising inference business, there are two test models as mentioned above: single mode and multi-party mode. In the single mode, a participant downloads the model for testing, and in the multi-party mode, multiple parties jointly download the model for testing.
以上所述的具体实施方式对本发明的技术方案和有益效果进行了详细说明，应理解的是以上所述仅为本发明的最优选实施例，并不用于限制本发明，凡在本发明的原则范围内所做的任何修改、补充和等同替换等，均应包含在本发明的保护范围之内。The above-mentioned specific embodiments describe in detail the technical solutions and beneficial effects of the present invention. It should be understood that the above-mentioned embodiments are only the most preferred embodiments of the present invention, and are not intended to limit the present invention. Any modifications, additions and equivalent substitutions made within the scope shall be included within the protection scope of the present invention.
Claims (10)
Translated fromChinese
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111488605.6ACN114491615A (en) | 2021-12-08 | 2021-12-08 | Blockchain-based Asynchronous Vertical Federated Learning Fair Incentive Mechanism Method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202111488605.6ACN114491615A (en) | 2021-12-08 | 2021-12-08 | Blockchain-based Asynchronous Vertical Federated Learning Fair Incentive Mechanism Method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114491615Atrue CN114491615A (en) | 2022-05-13 |
Family
ID=81492250
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202111488605.6APendingCN114491615A (en) | 2021-12-08 | 2021-12-08 | Blockchain-based Asynchronous Vertical Federated Learning Fair Incentive Mechanism Method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114491615A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114819197A (en)* | 2022-06-27 | 2022-07-29 | 杭州同花顺数据开发有限公司 | Block chain alliance-based federal learning method, system, device and storage medium |
| CN115345317A (en)* | 2022-08-05 | 2022-11-15 | 北京交通大学 | Fair reward distribution method based on fairness theory and oriented to federal learning |
| CN115660114A (en)* | 2022-11-11 | 2023-01-31 | 湖北文理学院 | Blockchain-based asynchronous federated learning architecture system and method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111931242A (en)* | 2020-09-30 | 2020-11-13 | 国网浙江省电力有限公司电力科学研究院 | Data sharing method, computer equipment applying same and readable storage medium |
| CN111966698A (en)* | 2020-07-03 | 2020-11-20 | 华南师范大学 | Credible federal learning method, system, device and medium based on block chain |
| CN111967610A (en)* | 2020-08-18 | 2020-11-20 | 中国银行股份有限公司 | Block chain-based federal learning incentive method, device, equipment and storage medium |
| WO2021004551A1 (en)* | 2019-09-26 | 2021-01-14 | 深圳前海微众银行股份有限公司 | Method, apparatus, and device for optimization of vertically federated learning system, and a readable storage medium |
| CN112328698A (en)* | 2020-11-19 | 2021-02-05 | 杭州趣链科技有限公司 | Longitudinal federal learning incentive training method, electronic device and storage medium |
| CN112686370A (en)* | 2020-12-25 | 2021-04-20 | 深圳前海微众银行股份有限公司 | Network structure search method, device, equipment, storage medium and program product |
| CN113657608A (en)* | 2021-08-05 | 2021-11-16 | 浙江大学 | An Incentive-Driven Federated Learning Approach for Blockchain |
- 2021
- 2021-12-08CNCN202111488605.6Apatent/CN114491615A/enactivePending
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2021004551A1 (en)* | 2019-09-26 | 2021-01-14 | 深圳前海微众银行股份有限公司 | Method, apparatus, and device for optimization of vertically federated learning system, and a readable storage medium |
| CN111966698A (en)* | 2020-07-03 | 2020-11-20 | 华南师范大学 | Credible federal learning method, system, device and medium based on block chain |
| CN111967610A (en)* | 2020-08-18 | 2020-11-20 | 中国银行股份有限公司 | Block chain-based federal learning incentive method, device, equipment and storage medium |
| CN111931242A (en)* | 2020-09-30 | 2020-11-13 | 国网浙江省电力有限公司电力科学研究院 | Data sharing method, computer equipment applying same and readable storage medium |
| CN112328698A (en)* | 2020-11-19 | 2021-02-05 | 杭州趣链科技有限公司 | Longitudinal federal learning incentive training method, electronic device and storage medium |
| CN112686370A (en)* | 2020-12-25 | 2021-04-20 | 深圳前海微众银行股份有限公司 | Network structure search method, device, equipment, storage medium and program product |
| CN113657608A (en)* | 2021-08-05 | 2021-11-16 | 浙江大学 | An Incentive-Driven Federated Learning Approach for Blockchain |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114819197A (en)* | 2022-06-27 | 2022-07-29 | 杭州同花顺数据开发有限公司 | Block chain alliance-based federal learning method, system, device and storage medium |
| CN114819197B (en)* | 2022-06-27 | 2023-07-04 | 杭州同花顺数据开发有限公司 | Federal learning method, system, device and storage medium based on blockchain alliance |
| CN115345317A (en)* | 2022-08-05 | 2022-11-15 | 北京交通大学 | Fair reward distribution method based on fairness theory and oriented to federal learning |
| CN115660114A (en)* | 2022-11-11 | 2023-01-31 | 湖北文理学院 | Blockchain-based asynchronous federated learning architecture system and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN114491615A (en) | Blockchain-based Asynchronous Vertical Federated Learning Fair Incentive Mechanism Method | |
| Lu et al. | Low-latency federated learning and blockchain for edge association in digital twin empowered 6G networks | |
| CN110113388B (en) | Improved clustering algorithm-based block chain system consensus method and device | |
| CN114819197B (en) | Federal learning method, system, device and storage medium based on blockchain alliance | |
| CN115102763B (en) | Multi-domain DDoS attack detection method and device based on trusted federated learning | |
| CN118260788B (en) | Supply chain management method based on artificial intelligence | |
| WO2023029324A1 (en) | Marketing arbitrage underground industry identification method based on dynamic attention graph network | |
| CN114915429B (en) | Communication perception calculation integrated network distributed credible perception method and system | |
| CN116665319B (en) | A multimodal biometric recognition method based on federated learning | |
| CN113228059A (en) | Cross-network-oriented representation learning algorithm | |
| CN113761217A (en) | Artificial intelligence-based question set data processing method and device and computer equipment | |
| CN109858270A (en) | A kind of construction method and system of decentralization digital identity | |
| CN114358907A (en) | Financial risk prediction method and system based on block chain federal learning | |
| CN113902127B (en) | A blockchain-enabled asynchronous federated learning method | |
| CN115564454A (en) | Method and device for construction and application of blockchain-driven enterprise time series knowledge graph | |
| CN115640305A (en) | A fair and credible federated learning method based on blockchain | |
| CN116595094A (en) | Federal learning incentive method, device, equipment and storage medium based on block chain | |
| CN112085051A (en) | Image classification method and system based on weighted voting and electronic equipment | |
| Yan et al. | Reputation consensus-based scheme for information sharing in internet of vehicles | |
| CN116541769A (en) | Node data classification method and system based on federal learning | |
| CN110166415A (en) | Reputation data processing method based on Anonymizing networks and machine learning | |
| CN115374479A (en) | Federal learning privacy protection method under non-independent same distributed data scene | |
| CN115269723A (en) | Block chain green consensus method and system for energy recovery facing federal learning | |
| CN109598289A (en) | cross-platform data processing method, device, equipment and readable storage medium | |
| CN110175283B (en) | A method and device for generating a recommendation model |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |