WO2022088541A1 - 一种基于差分进化的联邦学习激励方法和系统 - Google Patents

一种基于差分进化的联邦学习激励方法和系统 Download PDF

Info

Publication number
WO2022088541A1
WO2022088541A1 PCT/CN2021/074276 CN2021074276W WO2022088541A1 WO 2022088541 A1 WO2022088541 A1 WO 2022088541A1 CN 2021074276 W CN2021074276 W CN 2021074276W WO 2022088541 A1 WO2022088541 A1 WO 2022088541A1
Authority
WO
WIPO (PCT)
Prior art keywords
participant
differential evolution
federated learning
cycle
individual
Prior art date
Application number
PCT/CN2021/074276
Other languages
English (en)
French (fr)
Inventor
麦伟杰
沈凤山
危明铸
袁峰
Original Assignee
广州中国科学院软件应用技术研究所
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 广州中国科学院软件应用技术研究所 filed Critical 广州中国科学院软件应用技术研究所
Publication of WO2022088541A1 publication Critical patent/WO2022088541A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G06N20/20—Ensemble learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/12—Computing arrangements based on biological models using genetic models
    • G06N3/126—Evolutionary algorithms, e.g. genetic algorithms or genetic programming

Definitions

  • the embodiments of the present invention relate to the field of information technology, and in particular, to a federated learning incentive method and system based on differential evolution.
  • Federated machine learning refers to a machine learning framework that can effectively help multiple nodes (which can represent individuals or institutions) to jointly train models while meeting the requirements of data privacy protection.
  • the server sends model parameters to multiple nodes, and each node inputs the local training samples into the model for one training. After the training, each node will calculate the gradient based on the training results. . Then, based on the Secure Aggregation (SA, Secure Aggregation) protocol, the server can calculate the sum of the gradients of each node.
  • SA Secure Aggregation
  • the amount of data required for training artificial intelligence application models is very large, but the data information related to major emergencies in the implementation is "small amount of data" and scattered in different institutions or regions, that is This kind of data is either small in scale; or lacks important information such as labels or some feature values; or the data is legally protected private data, a phenomenon called "data silos". Due to the emergence of this phenomenon, the joint participation of all participants is required in the federated learning process to train an accurate and reliable model. However, how to keep the participants involved in federated learning is an important challenge, and the key to achieving this goal is to develop a reward method to share the profits generated by the federation with the participants fairly and equitably.
  • the existing method is the data operator, which is led by industry alliances or key government units. It adopts the project team to develop and share exchange toolsets and platforms, and is responsible for data aggregation and management.
  • the operators and users of the data constitute a complete industrial system. Each unit pays the relevant fees in the process of using the data, and establishes an incentive mechanism in the form of capital returns.
  • the embodiments of the present invention provide a federated learning incentive method and system based on differential evolution, and 3) effectively realize the dynamic adjustment of the total federated learning revenue and the revenue of each participant, maximize sustainable business goals, and minimize the participation of participants. Inequity issues, avoid relying on human intervention.
  • an embodiment of the present invention provides a federated learning incentive method based on differential evolution, including:
  • Step S1 Obtain the expected loss offset of the participant i in the i-th cycle in the federated learning operation cycle T:
  • U i (t) is the revenue of participant i in the t-th cycle
  • B(t) is the total revenue
  • C i (t) is the cost for participant i to contribute data to the federation in the t-th cycle
  • Yi (t) is the difference between the benefits
  • Q i ( t) is the time queue waiting for federal payment
  • Step S3 obtaining C i (t) and Q i (t) of each participant;
  • Step S4 take f(t) as the objective function, and take U i (t), Y i (t), Q i (t), and ⁇ i (t) as constraints, perform differential evolution processing, and obtain the minimum expected loss and waiting time.
  • step S1 Preferably, in the step S1:
  • Y i (t+1) max[(Y i (t)+C i (t)-u i (t),0]
  • the income of each participant is encoded into the form of a population, and the initial fitness f(t)' is obtained by recording, which specifically includes:
  • the number of participants is n, and the income of each participant is encoded into the formation of the population:
  • the obtaining C i (t) and Q i (t) of each participant specifically includes:
  • i represents the participant, that is, the population individual
  • participant i contributes to federated data d i (t)>0, calculate C i (t), Qi ( t);
  • the differential evolution process is performed using f(t) as the objective function and U i (t), Y i (t), Q i (t), and ⁇ i (t) as constraints, specifically including :
  • Step S41 in the contemporary period t, for each individual u i,t , randomly select three individual vectors u r1,t , u r2,t , u r3,t from the current population, where r 1 ⁇ r 2 ⁇ r 3 ⁇ i, and random integers of r 1 , r 2 , r 3 ⁇ ⁇ 1,2,...,n ⁇ ;
  • the mutation operation is performed according to the following formula to produce mutant individuals u i,t :
  • V i,t u r1,t +F ⁇ (u r2,t -u r3,t )
  • Step S42 performing random recombination and crossover on each component of the target vector ui ,t and the variation vector Vi ,t :
  • Step S43 based on the fitness value of the individual, compare the fitness of the experimental vector si,t U i,g and the target vector ui ,t , when the experimental individual Si is better than the target individual ui , select Si . Enter the evolution of the next generation, otherwise, select ui ;
  • it also includes:
  • Step S5 update the value of the participant in round t, and at the same time update the value of Y i (t) and Qi ( t) according to it.
  • it also includes:
  • an embodiment of the present invention provides a differential evolution-based federated learning incentive system, including:
  • the expected loss module is used to obtain the expected loss offset of the participant i in the ith cycle of the federated learning running cycle T:
  • U i (t) is the revenue of participant i in the t-th cycle
  • B(t) is the total revenue
  • C i (t) is the cost for participant i to contribute data to the federation in the t-th cycle
  • Yi (t) is the difference between the benefits
  • Q i ( t) is the time queue waiting for federal payment
  • Participant calculation module to obtain C i (t) and Q i (t) of each participant
  • the differential evolution processing module is used to perform differential evolution processing with f(t) as the objective function and U i (t), Y i (t), Q i (t), and ⁇ i (t) as constraints, and obtain Minimize expectation loss and wait time.
  • an embodiment of the present invention provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor.
  • the processor executes the program, the first embodiment of the present invention is implemented.
  • an embodiment of the present invention provides a non-transitory computer-readable storage medium on which a computer program is stored. Steps of the Federated Learning Incentive Approach.
  • a federated learning incentive method and system based on differential evolution provided by the embodiment of the present invention utilizes the excellent global optimization capability and local detection capability of DE, so that each participant in the federated learning process transfers with time t (assuming a monthly cycle) , the difference (expected loss) between the benefits each participant gets from the federation and the benefits they should get, minimizes the “expected loss and waiting time” between the participants, and automatically balances the actual performance of each participant in federated learning.
  • the expectation of income and the return obtained is poor, which effectively promotes the participants to provide reliable data so that federated learning can be carried out in a long-term and stable manner; effectively realizes the dynamic adjustment of the total income of federated learning and the income of each participant, maximizes sustainable business goals, and at the same time Minimize the unfairness of the participants and avoid relying on human intervention.
  • FIG. 1 is a flowchart of a federated learning incentive method based on differential evolution according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of a server according to an embodiment of the present invention.
  • first and second in the embodiments of the present application are only used for the purpose of description, and cannot be understood as indicating or implying relative importance or implying the number of indicated technical features. Thus, a feature delimited with “first”, “second” may expressly or implicitly include at least one of that feature.
  • the terms “comprising” and “having” and any variations thereof are intended to cover non-exclusive inclusion. For example, a system, product or device comprising a series of components or units is not limited to the listed components or units, but may optionally also include components or units not listed, or Other parts or units inherent in the equipment.
  • "a plurality of” means at least two, such as two, three, etc., unless otherwise expressly and specifically defined.
  • FIG. 1 provides a differential evolution-based federated learning incentive method according to an embodiment of the present invention, including:
  • Step S1 Obtain the expected loss offset of the participant i in the i-th cycle in the federated learning operation cycle T:
  • U i (t) is the revenue of the participant i in the t-th cycle
  • B(t) is the total revenue
  • C i (t) is the data d(t) contributed by the participant i to the t-th cycle in the t-th cycle.
  • Yi (t) is the difference between the benefits
  • Qi (t) represents the time queue waiting for the federation to pay;
  • Step S3 obtaining C i (t) and Q i (t) of each participant;
  • Step S4 take f(t) as the objective function, and take U i (t), Y i (t), Q i (t), and ⁇ i (t) as constraints, perform differential evolution processing, and obtain the minimum expected loss and waiting time.
  • step S1 Preferably, in the step S1:
  • the income of each participant is encoded into the form of a population, and the initial fitness f(t)' is obtained by recording, which specifically includes:
  • the number of participants is n, and the income of each participant is encoded into the formation of the population:
  • the obtaining of C i (t) and Q i (t) of each participant specifically includes:
  • i represents the participant, that is, the population individual
  • participant i contributes to federated data d i (t)>0, calculate C i (t), Qi ( t);
  • the differential evolution is performed with f(t) as the objective function and U i (t), Y i (t), Q i (t), and ⁇ i (t) as constraints. processing, including:
  • Step S41 in the contemporary period t, for each individual u i,t , randomly select three individual vectors u r1,t , u r2,t , u r3,t from the current population, where r 1 ⁇ r 2 ⁇ r 3 ⁇ i, and random integers of r 1 , r 2 , r 3 ⁇ ⁇ 1,2,...,n ⁇ ;
  • the mutation operation is performed according to the following formula to produce mutant individuals u i,t :
  • V i,t u r1,t +F ⁇ (u r2,t -u r3,t ) (6)
  • Step S42 performing random recombination and crossover on each component of the target vector ui ,t and the variation vector Vi ,t :
  • the crossover operation of DE is mainly to improve the potential diversity of the population.
  • each component of the target vector ui ,t and the variation vector Vi ,t is randomly recombined, but it must be ensured that the experimental vector Si ,t has at least one
  • the component is from the variation vector V i,t , the other components are controlled by the parameter CR.
  • the crossover operation is performed according to the following formula (7).
  • Step S43 based on the fitness value of the individual, compare the fitness of the experimental vector si,t U i,g and the target vector ui ,t , when the experimental individual Si is better than the target individual ui , select Si . Enter the evolution of the next generation, otherwise, select ui ;
  • the selection operation of DE is carried out on the basis of the fitness value of the individual (the present invention refers to the participant's gain), which is essentially the experimental vector s i, t U i, g and the target Fitness comparison of vectors u i,t . That is, when the experimental individual Si is better than the target individual ui , Si will be selected into the next generation of evolution, otherwise , ui will be selected.
  • the selection operation is calculated according to the formula (8).
  • Step S5 update the value of the participant in round t, and at the same time update the value of Y i (t) and Qi ( t) according to it.
  • Embodiments of the present invention further provide a differential evolution-based federated learning incentive system, based on the differential evolution-based federated learning incentive methods in the above embodiments, including:
  • the expected loss module is used to obtain the expected loss offset of the participant i in the ith cycle of the federated learning running cycle T:
  • U i (t) is the revenue of participant i in the t-th cycle
  • B(t) is the total revenue
  • C i (t) is the cost for participant i to contribute data to the federation in the t-th cycle
  • Yi (t) is the difference between the benefits
  • Q i ( t) is the time queue waiting for federal payment
  • Participant calculation module to obtain C i (t) and Q i (t) of each participant
  • the differential evolution processing module is used to perform differential evolution processing with f(t) as the objective function and U i (t), Y i (t), Q i (t), and ⁇ i (t) as constraints, and obtain Minimize expectation loss and wait time.
  • an embodiment of the present invention also provides a server.
  • the server may include: a processor (processor) 810, a communication interface (Communications Interface) 820, a memory (memory) 830, and a communication bus 840 , wherein the processor 810 , the communication interface 820 , and the memory 830 communicate with each other through the communication bus 840 .
  • the processor 810 may invoke the logic instructions in the memory 830 to execute the steps of the differential evolution-based federated learning incentive method described in the foregoing embodiments.
  • the above-mentioned logic instructions in the memory 830 can be implemented in the form of software functional units and can be stored in a computer-readable storage medium when sold or used as an independent product.
  • the technical solution of the present invention can be embodied in the form of a software product in essence, or the part that contributes to the prior art or the part of the technical solution.
  • the computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which may be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention.
  • the aforementioned storage medium includes: U disk, mobile hard disk, Read-Only Memory (ROM, Read-Only Memory), Random Access Memory (RAM, Random Access Memory), magnetic disk or optical disk and other media that can store program codes .
  • an embodiment of the present invention also provides a non-transitory computer-readable storage medium, where the computer-readable storage medium stores a computer program, and the computer program includes at least one piece of code, and the at least one piece of code can be executed by a main control device , to control the main control device to implement the steps of the differential evolution-based federated learning incentive method described in the foregoing embodiments.
  • the embodiments of the present application further provide a computer program, which is used to implement the above method embodiments when the computer program is executed by a main control device.
  • the program may be stored in whole or in part on a storage medium packaged with the processor, or may be stored in part or in part in a memory not packaged with the processor.
  • an embodiment of the present application further provides a processor, and the processor is used to implement the above method embodiments.
  • the above-mentioned processor may be a chip.
  • a federated learning incentive method and system based on differential evolution utilizes the excellent global optimization capability and local detection capability of DE, so that each participant in the federated learning process transfers with time t (assuming Monthly cycle), the difference (expected loss) between the benefits each participant gets from the federation and the benefits they should get, minimizes the "expected loss and waiting time" between the participants, and automatically balances the federated learning process.
  • the actual income and the expected return of each participant are poor, which effectively promotes the participants to provide reliable data so that the federated learning can be carried out in a long-term and stable manner; effectively realizes the dynamic adjustment of the total income of the federated learning and the income of each participant, and maximizes sustainable development. business objectives, while minimizing the unfairness of the participants and avoiding relying on manual intervention.
  • the above-mentioned embodiments it may be implemented in whole or in part by software, hardware, firmware or any combination thereof.
  • software it can be implemented in whole or in part in the form of a computer program product.
  • the computer program product includes one or more computer instructions.
  • the computer program instructions when loaded and executed on a computer, produce, in whole or in part, the processes or functions described herein.
  • the computer may be a general purpose computer, special purpose computer, computer network, or other programmable device.
  • the computer instructions may be stored in or transmitted from one computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be downloaded from a website site, computer, server, or data center Transmission to another website site, computer, server, or data center by wire (eg, coaxial cable, optical fiber, digital subscriber line) or wireless (eg, infrared, wireless, microwave, etc.).
  • the computer-readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that includes an integration of one or more available media.
  • the usable media may be magnetic media (eg, floppy disks, hard disks, magnetic tapes), optical media (eg, DVD), or semiconductor media (eg, Solid State Disk), and the like.
  • the process can be completed by instructing the relevant hardware by a computer program, and the program can be stored in a computer-readable storage medium.
  • the program When the program is executed , which may include the processes of the foregoing method embodiments.
  • the aforementioned storage medium includes: ROM or random storage memory RAM, magnetic disk or optical disk and other mediums that can store program codes.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Software Systems (AREA)
  • Biophysics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Mathematical Physics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • General Physics & Mathematics (AREA)
  • Evolutionary Biology (AREA)
  • Computational Linguistics (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Physiology (AREA)
  • Biomedical Technology (AREA)
  • Genetics & Genomics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Medical Informatics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

一种基于差分进化的联邦学习激励方法和系统，利用DE的全局优化能力及局部探测能力，使得联邦学习过程中随时间t转移，每一位参与方从联邦中获得的收益与其应得到的收益之间的差别及等待时间最小化，自动平衡联邦学习中各个参与方在实际收益和所得的回报期望差，有效促进参与方提供可靠的数据使得联邦学习长期、稳定的进行；有效地实现联邦学习总收益与各个参与方收益的动态调整，最大化可持续的经营目标，同时最小化参与方的不公平问题，避免了依靠人工干预。

Description

一种基于差分进化的联邦学习激励方法和系统 技术领域
本发明实施例涉及信息技术领域，尤其涉及一种基于差分进化的联邦学习激励方法和系统。
背景技术
联邦学习(Federated machine learning/Federated Learning)，是指一种机器学习框架，能有效帮助多个节点(可以代表个人或机构)在满足数据隐私保护的要求下，联合训练模型。在联邦学习框架下，服务端下发模型参数给多个节点，每个节点将本地的训练样本输入模型进行一次训练，本次训练结束后，每个节点会基于本次训练结果计算得到的梯度。随后，服务端基于安全聚合(SA，Secure Aggregation)协议，可以计算得到各节点的梯度之和。
一般而言，训练人工智能应用模型所需要的数据量都是非常庞大的，但实现中与重大突发事件相关的数据信息为“小量数据”且零散地分布在不同的机构或区域，即此类数据要么规模较小；要么缺少标签或部分特征数值等重要信息；要么数据是受法律保护的隐私数据，这种现象称之为“数据孤岛”。由于这种现象的出现，导致进行联邦学习过程中需要各个参与方的共同参与才能训练出精确、可靠的模型。然而，如何使得参与方持续地参与到联邦学习中是一项重要的挑战，实现这一目标的关键是制定一种奖励方法，公平公正地与参与方分享联邦产生的利润。现有的方法为数据运营方，由产业联盟或者关键政府单位牵头成立，采用项目团队研发共享交换工具集与平台，负责数据的汇聚和管理，同时向数据的使用方收取一定比例的费用，数据的运营方、使用方构成了完整了产业体系，各单位在使用数据过程中支付相关的费用，并以资金回报的形式建立激励机制。但是，上述的激励方法很难做到随着时间的转移公平、公正、动态将联邦学习的收益合理地分配给各个参与方，并且存在大量的人工干预问题。
发明内容
本发明实施例提供一种基于差分进化的联邦学习激励方法和系统，3)有效地实现联邦学习总收益与各个参与方收益的动态调整，最大化可持续的经营目标，同时最小化参与方的不公平问题，避免了依靠人工干预。
第一方面，本发明实施例提供一种基于差分进化的联邦学习激励方法，包括：
步骤S1、获取联邦学习运行周期T中第i个周期中参与方i的期望损失偏移：
Figure PCTCN2021074276-appb-000001
Figure PCTCN2021074276-appb-000002
其中，U i(t)为参与方i在第t个周期的收益；B(t)为总收益；C i(t)为参与方i在第t个周期将数据贡献给联邦所需的代价；Y i(t)为收益之间的差别；Q i(t)表示等待联邦支付的时间队列；
步骤S2、初始化最大收益轮次T，收益B(t)，Y i(t)＝0，Q i(t)＝0；设定差分进化算法的缩放因子和交叉因子；将各参与方的收益编码成种群形式，记录得到初始的适宜度f(t)’；
步骤S3、获取各参与方的C i(t)和Q i(t)；
步骤S4、以f(t)为目标函数，以U i(t)、Y i(t)、Q i(t)、λ i(t)为约束条件，进行差分进化处理，得到最小化期待损失和等待时间。
作为优选的，所述步骤S1中：
Figure PCTCN2021074276-appb-000003
Y i(t)为一个队列系统：
Y i(t+1)＝max[(Y i(t)+C i(t)-u i(t),0]
Q i(t)为一个时间队列：
Q i(t+1)＝max[(Q i(t)+λ i(t)-u i(t),0]。
作为优选的，所述步骤S2中，设定差分进化算法的缩放因子F＝0.5，交叉因子CR＝0.5；
所述将各参与方的收益编码成种群形式，记录得到初始的适宜度f(t)’， 具体包括：
参与方的数目为n,将各个参与方的收益编码成种群的形成：
Figure PCTCN2021074276-appb-000004
其中，每个参与方拥有的属性维度为D；把t＝0时的Y i(t)、Q i(t)、C i(t)、λ i(t)的值代入期望损失偏移中，记录初始的适宜度f(t)’的值。
作为优选的，所述获取各参与方的C i(t)和Q i(t)，具体包括：
从i＝1开始到n，i表示参与方，即种群个体；
如果参与方i贡献给联邦数据d i(t)>0，则计算C i(t)、Q i(t)；
如果i没有提供任何数据，即C i(t)＝0。
作为优选的，所述以f(t)为目标函数，以U i(t)、Y i(t)、Q i(t)、λ i(t)为约束条件，进行差分进化处理，具体包括：
步骤S41、在当代周期t中，对于每个个体u i,t，从当前种群中随机选择三个个体向量u r1,t，u r2,t，u r3,t，其中r 1≠r 2≠r 3≠i，且r 1，r 2，r 3∈{1,2,···,n}的随机整数；
按照下式进行变异操作生产变异个体u i,t：
V i,t＝u r1,t+F·(u r2,t-u r3,t)
步骤S42、对目标向量u i,t和变异变异向量V i,t的各个分量进行随机重组交叉：
Figure PCTCN2021074276-appb-000005
步骤S43、以个体的适应值为基础，对实验向量s i,t U i,g与目标向量u i,t的适应度比较，当实验个体S i优于目标个体u i时，选中S i进入下一代的进化，否则，选中u i；
Figure PCTCN2021074276-appb-000006
作为优选的，还包括：
步骤S5、更新参与方在t轮次的值，同时根据更新Y i(t)、Q i(t)的值。
作为优选的，还包括：
步骤S6、以目标函数最大的评价次数作为算法的终止条件；若满足条件，则输出最优个体，此值为最佳轮次方案解；否则，令t＝t+1，之后转向步骤S42。
第二方面，本发明实施例提供一种基于差分进化的联邦学习激励系统，包括：
期望损失模块，用于获取联邦学习运行周期T中第i个周期中参与方i的期望损失偏移：
Figure PCTCN2021074276-appb-000007
Figure PCTCN2021074276-appb-000008
其中，U i(t)为参与方i在第t个周期的收益；B(t)为总收益；C i(t)为参与方i在第t个周期将数据贡献给联邦所需的代价；Y i(t)为收益之间的差别；Q i(t)表示等待联邦支付的时间队列；
初始化模块，初始化最大收益轮次T，收益B(t)，Y i(t)＝0，Q i(t)＝0；设定差分进化算法的缩放因子和交叉因子；将各参与方的收益编码成种群形式，记录得到初始的适宜度f(t)’；
参与方计算模块，获取各参与方的C i(t)和Q i(t)；
差分进化处理模块，用于以f(t)为目标函数，以U i(t)、Y i(t)、Q i(t)、λ i(t)为约束条件，进行差分进化处理，得到最小化期待损失和等待时间。
第三方面，本发明实施例提供一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现如本发明第一方面实施例所述基于差分进化的联邦学习激励方法的步骤。
第四方面，本发明实施例提供一种非暂态计算机可读存储介质，其上存储有计算机程序，该计算机程序被处理器执行时实现如本发明第一方面实施例所述基于差分进化的联邦学习激励方法的步骤。
本发明实施例提供的一种基于差分进化的联邦学习激励方法和系统，利用DE优秀的全局优化能力及局部探测能力，使得联邦学习过程中各个参与方随时间t转移(假设以月为周期)，每一位参与方从联邦中获得的收益与其应得到的收益之间的差别(期望损失)，最小化参与方之间“期待损失与等待时间”，自动平衡联邦学习中各个参与方在实际收益和所得的回报期望差，有效促进参与方提供可靠的数据使得联邦学习长期、稳定的进行；有效地实现联邦学习总收益与各个参与方收益的动态调整，最大化可持续的经营目标，同时最小化参与方的不公平问题，避免了依靠人工干预。
附图说明
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作一简单地介绍，显而易见地，下面描述中的附图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的前提下，还可以根据这些附图获得其他的附图。
图1为根据本发明实施例的基于差分进化的联邦学习激励方法流程框图；
图2为根据本发明实施例的服务器示意图。
具体实施方式
为使本发明实施例的目的、技术方案和优点更加清楚，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例是本发明一部分实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有作出创造性劳动前提下所获得的所有其他实施例，都属于本发明保护的范围。
本申请实施例中术语“和/或”，仅仅是一种描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。
本申请实施例中的术语“第一”、“第二”仅用于描述目的，而不能理解为指示或暗示相对重要性或者隐含指明所指示的技术特征的数量。由此，限定有“第一”、“第二”的特征可以明示或者隐含地包括至少一个该特征。本申请的描述中，术语“包括”和“具有”以及它们任何变形，意图在于覆盖不排他的包含。例如包含了一系列部件或单元的系统、产品或设备没有限定于已列出的部件或单元，而是可选地还包括没有列出的部件或单元，或可选地还包括对于这些产品或设备固有的其它部件或单元。本申请的描述中，“多个”的含义是至少两个，例如两个，三个等，除非另有明确具体的限定。
在本文中提及“实施例”意味着，结合实施例描述的特定特征、结构或特性可以包含在本申请的至少一个实施例中。在说明书中的各个位置出现该短语并不一定均是指相同的实施例，也不是与其它实施例互斥的独立的或备选的实施例。本领域技术人员显式地和隐式地理解的是，本文所描述的实施 例可以与其它实施例相结合。
图1为本发明实施例提供一种基于差分进化的联邦学习激励方法，包括：
步骤S1、获取联邦学习运行周期T中第i个周期中参与方i的期望损失偏移：
Figure PCTCN2021074276-appb-000009
Figure PCTCN2021074276-appb-000010
其中，U i(t)为参与方i在第t个周期的收益；B(t)为总收益；C i(t)为参与方i在第t个周期中将数据d(t)贡献给联邦所需的代价，假设已经可用；Y i(t)为收益之间的差别；Q i(t)表示等待联邦支付的时间队列；
步骤S2、初始化最大收益轮次T，收益B(t)，Y i(t)＝0，Q i(t)＝0；设定差分进化算法的缩放因子和交叉因子；将各参与方的收益编码成种群形式，记录得到初始的适宜度f(t)’；
步骤S3、获取各参与方的C i(t)和Q i(t)；
步骤S4、以f(t)为目标函数，以U i(t)、Y i(t)、Q i(t)、λ i(t)为约束条件，进行差分进化处理，得到最小化期待损失和等待时间。
作为优选的，所述步骤S1中：
Figure PCTCN2021074276-appb-000011
Y i(t)为一个队列系统：
Y i(t+1)＝max[(Y i(t)+C i(t)-u i(t),0]            (4)
Q i(t)为一个时间队列：
Q i(t+1)＝max[(Q i(t)+λ i(t)-u i(t),0]           (5)
在上述实施例的基础上，所述步骤S2中，设定差分进化算法的缩放因子F＝0.5，交叉因子CR＝0.5；
所述将各参与方的收益编码成种群形式，记录得到初始的适宜度f(t)’，具体包括：
参与方的数目为n,将各个参与方的收益编码成种群的形成：
Figure PCTCN2021074276-appb-000012
其中，每个参与方拥有的属性维度为D；把t＝0时的Y i(t)、Q i(t)、C i(t)、λ i(t)的值代入式(1)中，记录初始的适宜度f(t)’的值。
在上述实施例的基础上，所述获取各参与方的C i(t)和Q i(t)，具体包括：
从i＝1开始到n，i表示参与方，即种群个体；
如果参与方i贡献给联邦数据d i(t)>0，则计算C i(t)、Q i(t)；
如果i没有提供任何数据，即C i(t)＝0。
在上述实施例的基础上，所述以f(t)为目标函数，以U i(t)、Y i(t)、Q i(t)、λ i(t)为约束条件，进行差分进化处理，具体包括：
步骤S41、在当代周期t中，对于每个个体u i,t，从当前种群中随机选择三个个体向量u r1,t，u r2,t，u r3,t，其中r 1≠r 2≠r 3≠i，且r 1，r 2，r 3∈{1,2,···,n}的随机整数；
按照下式进行变异操作生产变异个体u i,t：
V i,t＝u r1,t+F·(u r2,t-u r3,t)               (6)
步骤S42、对目标向量u i,t和变异变异向量V i,t的各个分量进行随机重组交叉：
Figure PCTCN2021074276-appb-000013
DE的交叉操作主要是为了提高种群潜在的多样性，通常的对目标向量u i,t和变异变异向量V i,t的各个分量进行随机重组实现，但必须确保实验向量S i,t至少有一分量是来自于变异向量V i,t，其它分量由参数CR控制。交叉操作按如下(7)式进行。
步骤S43、以个体的适应值为基础，对实验向量s i,t U i,g与目标向量u i,t的适应度比较，当实验个体S i优于目标个体u i时，选中S i进入下一代的进化，否则，选中u i；
Figure PCTCN2021074276-appb-000014
选择操作：根据“贪心选择”方案，DE的选择操作是以个体的适应值(本发明指参与方获得收益)为基础而进行的，实质上就是实验向量s i,t U i,g与目标向量u i,t的适应度比较。即当实验个体S i优于目标个体u i时，S i会被选中进入下一代的进化，否则，u i会被选中。选择操作按(8)式进行计算。
在上述实施例的基础上，还包括：
步骤S5、更新参与方在t轮次的值，同时根据更新Y i(t)、Q i(t)的值。
在上述实施例的基础上，还包括：
步骤S6、以目标函数最大的评价次数作为算法的终止条件， MAX_FES＝5000*，D为自变量U的维度为D；若满足条件，则输出最优个体，此值为最佳轮次方案解；否则，令t＝t+1，之后转向步骤S42。
本发明实施例还提供一种基于差分进化的联邦学习激励系统，基于上述各实施例中的基于差分进化的联邦学习激励方法，包括：
期望损失模块，用于获取联邦学习运行周期T中第i个周期中参与方i的期望损失偏移：
Figure PCTCN2021074276-appb-000015
Figure PCTCN2021074276-appb-000016
其中，U i(t)为参与方i在第t个周期的收益；B(t)为总收益；C i(t)为参与方i在第t个周期将数据贡献给联邦所需的代价；Y i(t)为收益之间的差别；Q i(t)表示等待联邦支付的时间队列；
初始化模块，初始化最大收益轮次T，收益B(t)，Y i(t)＝0，Q i(t)＝0；设定差分进化算法的缩放因子和交叉因子；将各参与方的收益编码成种群形式，记录得到初始的适宜度f(t)’；
参与方计算模块，获取各参与方的C i(t)和Q i(t)；
差分进化处理模块，用于以f(t)为目标函数，以U i(t)、Y i(t)、Q i(t)、λ i(t)为约束条件，进行差分进化处理，得到最小化期待损失和等待时间。
基于相同的构思，本发明实施例还提供了一种服务器，如图2所示，该服务器可以包括：处理器(processor)810、通信接口(Communications Interface)820、存储器(memory)830和通信总线840，其中，处理器810，通信接口820，存储器830通过通信总线840完成相互间的通信。处理器810可以调用存储器830中的逻辑指令，以执行如上述各实施例所述基于差分进化的联邦学习激励方法的步骤。
此外，上述的存储器830中的逻辑指令可以通过软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本发明的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本发明各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储 器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
基于相同的构思，本发明实施例还提供一种非暂态计算机可读存储介质，该计算机可读存储介质存储有计算机程序，该计算机程序包含至少一段代码，该至少一段代码可由主控设备执行，以控制主控设备用以实现如上述各实施例所述基于差分进化的联邦学习激励方法的步骤。
基于相同的技术构思，本申请实施例还提供一种计算机程序，当该计算机程序被主控设备执行时，用以实现上述方法实施例。
所述程序可以全部或者部分存储在与处理器封装在一起的存储介质上，也可以部分或者全部存储在不与处理器封装在一起的存储器上。
基于相同的技术构思，本申请实施例还提供一种处理器，该处理器用以实现上述方法实施例。上述处理器可以为芯片。
综上所述，本发明实施例提供的一种基于差分进化的联邦学习激励方法和系统，利用DE优秀的全局优化能力及局部探测能力，使得联邦学习过程中各个参与方随时间t转移(假设以月为周期)，每一位参与方从联邦中获得的收益与其应得到的收益之间的差别(期望损失)，最小化参与方之间“期待损失与等待时间”，自动平衡联邦学习中各个参与方在实际收益和所得的回报期望差，有效促进参与方提供可靠的数据使得联邦学习长期、稳定的进行；有效地实现联邦学习总收益与各个参与方收益的动态调整，最大化可持续的经营目标，同时最小化参与方的不公平问题，避免了依靠人工干预。
本发明的各实施方式可以任意进行组合，以实现不同的技术效果。
在上述实施例中，可以全部或部分地通过软件、硬件、固件或者其任意组合来实现。当使用软件实现时，可以全部或部分地以计算机程序产品的形式实现。所述计算机程序产品包括一个或多个计算机指令。在计算机上加载和执行所述计算机程序指令时，全部或部分地产生按照本申请所述的流程或功能。所述计算机可以是通用计算机、专用计算机、计算机网络、或者其他可编程装置。所述计算机指令可以存储在计算机可读存储介质中，或者从一个计算机可读存储介质向另一个计算机可读存储介质传输，例如，所述计算机指令可以从一个网站站点、计算机、服务器或数据中心通过有线(例如同轴电缆、光纤、数字用户线)或无线(例如红外、无线、微波等)方式向另一个网 站站点、计算机、服务器或数据中心进行传输。所述计算机可读存储介质可以是计算机能够存取的任何可用介质或者是包含一个或多个可用介质集成的服务器、数据中心等数据存储设备。所述可用介质可以是磁性介质，(例如，软盘、硬盘、磁带)、光介质(例如，DVD)、或者半导体介质(例如固态硬盘Solid StateDisk)等。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，该流程可以由计算机程序来指令相关的硬件完成，该程序可存储于计算机可读取存储介质中，该程序在执行时，可包括如上述各方法实施例的流程。而前述的存储介质包括：ROM或随机存储记忆体RAM、磁碟或者光盘等各种可存储程序代码的介质。
最后应说明的是：以上实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的精神和范围。

Claims (10)

  1. 一种基于差分进化的联邦学习激励方法，其特征在于，包括：
    步骤S1、获取联邦学习运行周期T中第i个周期中参与方i的期望损失偏移：
    Figure PCTCN2021074276-appb-100001
    Figure PCTCN2021074276-appb-100002
    其中，U i(t)为参与方i在第t个周期的收益；B(t)为总收益；C i(t)为参与方i在第t个周期将数据贡献给联邦所需的代价；Y i(t)为收益之间的差别；Q i(t)表示等待联邦支付的时间队列；
    步骤S2、初始化最大收益轮次T，收益B(t)，Y i(t)＝0，Q i(t)＝0；设定差分进化算法的缩放因子和交叉因子；将各参与方的收益编码成种群形式，记录得到初始的适宜度f(t)’；
    步骤S3、获取各参与方的C i(t)和Q i(t)；
    步骤S4、以f(t)为目标函数，以U i(t)、Y i(t)、Q i(t)、λ i(t)为约束条件，进行差分进化处理，得到最小化期待损失和等待时间。
  2. 根据权利要求1所述的基于差分进化的联邦学习激励方法，其特征在于，所述步骤S1中：
    Figure PCTCN2021074276-appb-100003
    Y i(t)为一个队列系统：
    Y i(t+1)＝max[(Y i(t)+C i(t)-u i(t),0]
    Q i(t)为一个时间队列：
    Q i(t+1)＝max[(Q i(t)+λ i(t)-u i(t),0]。
  3. 根据权利要求2所述的基于差分进化的联邦学习激励方法，其特征在于，所述步骤S2中，设定差分进化算法的缩放因子F＝0.5，交叉因子CR＝0.5；
    所述将各参与方的收益编码成种群形式，记录得到初始的适宜度f(t)’，具体包括：
    参与方的数目为n,将各个参与方的收益编码成种群的形成：
    Figure PCTCN2021074276-appb-100004
    其中，每个参与方拥有的属性维度为D；把t＝0时的Y i(t)、Q i(t)、C i(t)、λ i(t)的值代入期望损失偏移中，记录初始的适宜度f(t)’的值。
  4. 根据权利要求1所述的基于差分进化的联邦学习激励方法，其特征在于，所述获取各参与方的C i(t)和Q i(t)，具体包括：
    从i＝1开始到n，i表示参与方，即种群个体；
    如果参与方i贡献给联邦数据d i(t)>0，则计算C i(t)、Q i(t)；
    如果i没有提供任何数据，即C i(t)＝0。
  5. 根据权利要求3所述的基于差分进化的联邦学习激励方法，其特征在于，所述以f(t)为目标函数，以U i(t)、Y i(t)、Q i(t)、λ i(t)为约束条件，进行差分进化处理，具体包括：
    步骤S41、在当代周期t中，对于每个个体u i,t，从当前种群中随机选择三个个体向量u r1,t，u r2,t，u r3,t，其中r 1≠r 2≠r 3≠i，且r 1，r 2，r 3∈{1,2,···,n}的随机整数；
    按照下式进行变异操作生产变异个体u i,t：
    V i,t＝u r1,t+F·(u r2,t-u r3,t)
    步骤S42、对目标向量u i,t和变异变异向量V i,t的各个分量进行随机重组交叉：
    Figure PCTCN2021074276-appb-100005
    步骤S43、以个体的适应值为基础，对实验向量s i,t U i,g与目标向量u i,t的适应度比较，当实验个体S i优于目标个体u i时，选中S i进入下一代的进化，否则，选中u i；
    Figure PCTCN2021074276-appb-100006
  6. 根据权利要求5所述的基于差分进化的联邦学习激励方法，其特征在于，还包括：
    步骤S5、更新参与方在t轮次的值，同时根据更新Y i(t)、Q i(t)的值。
  7. 根据权利要求6所述的基于差分进化的联邦学习激励方法，其特征在 于，还包括：
    步骤S6、以目标函数最大的评价次数作为算法的终止条件；若满足条件，则输出最优个体，此值为最佳轮次方案解；否则，令t＝t+1，之后转向步骤S42。
  8. 一种基于差分进化的联邦学习激励系统，其特征在于，包括：
    期望损失模块，用于获取联邦学习运行周期T中第i个周期中参与方i的期望损失偏移：
    Figure PCTCN2021074276-appb-100007
    Figure PCTCN2021074276-appb-100008
    其中，U i(t)为参与方i在第t个周期的收益；B(t)为总收益；C i(t)为参与方i在第t个周期将数据贡献给联邦所需的代价；Y i(t)为收益之间的差别；Q i(t)表示等待联邦支付的时间队列；
    初始化模块，初始化最大收益轮次T，收益B(t)，Y i(t)＝0，Q i(t)＝0；设定差分进化算法的缩放因子和交叉因子；将各参与方的收益编码成种群形式，记录得到初始的适宜度f(t)’；
    参与方计算模块，获取各参与方的C i(t)和Q i(t)；
    差分进化处理模块，用于以f(t)为目标函数，以U i(t)、Y i(t)、Q i(t)、λ i(t)为约束条件，进行差分进化处理，得到最小化期待损失和等待时间。
  9. 一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，其特征在于，所述处理器执行所述程序时实现如权利要求1至7任一项所述基于差分进化的联邦学习激励方法的步骤。
  10. 一种非暂态计算机可读存储介质，其上存储有计算机程序，其特征在于，该计算机程序被处理器执行时实现如权利要求1至7任一项所述基于差分进化的联邦学习激励方法的步骤。
PCT/CN2021/074276 2020-10-28 2021-01-29 一种基于差分进化的联邦学习激励方法和系统 WO2022088541A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011170752.4 2020-10-28
CN202011170752.4A CN112329947A (zh) 2020-10-28 2020-10-28 一种基于差分进化的联邦学习激励方法和系统

Publications (1)

Publication Number Publication Date
WO2022088541A1 true WO2022088541A1 (zh) 2022-05-05

Family

ID=74296344

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/074276 WO2022088541A1 (zh) 2020-10-28 2021-01-29 一种基于差分进化的联邦学习激励方法和系统

Country Status (2)

Country Link
CN (1) CN112329947A (zh)
WO (1) WO2022088541A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116384502A (zh) * 2022-09-09 2023-07-04 京信数据科技有限公司 联邦学习中参与方价值贡献计算方法、装置、设备及介质

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113157434B (zh) * 2021-02-26 2024-05-07 西安电子科技大学 一种横向联邦学习系统用户节点的激励方法及系统
CN113656833A (zh) * 2021-08-09 2021-11-16 浙江工业大学 基于进化计算的在垂直联邦架构下的隐私窃取防御方法
CN113837368A (zh) * 2021-09-27 2021-12-24 中国太平洋保险（集团）股份有限公司 一种在联邦学习中评估各参与方数据价值的控制方法及装置
CN114217933A (zh) * 2021-12-27 2022-03-22 北京百度网讯科技有限公司 多任务调度方法、装置、设备以及存储介质
CN115345317B (zh) * 2022-08-05 2023-04-07 北京交通大学 一种基于公平理论的面向联邦学习的公平奖励分配方法

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110363305A (zh) * 2019-07-17 2019-10-22 深圳前海微众银行股份有限公司 联邦学习方法、系统、终端设备及存储介质
CN110490335A (zh) * 2019-08-07 2019-11-22 深圳前海微众银行股份有限公司 一种计算参与者贡献率的方法及装置
CN110910158A (zh) * 2019-10-08 2020-03-24 深圳逻辑汇科技有限公司 联邦学习收入分配方法和系统
CN111222646A (zh) * 2019-12-11 2020-06-02 深圳逻辑汇科技有限公司 联邦学习机制的设计方法、装置和存储介质
US20200242514A1 (en) * 2016-09-26 2020-07-30 Google Llc Communication Efficient Federated Learning

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20200242514A1 (en) * 2016-09-26 2020-07-30 Google Llc Communication Efficient Federated Learning
CN110363305A (zh) * 2019-07-17 2019-10-22 深圳前海微众银行股份有限公司 联邦学习方法、系统、终端设备及存储介质
CN110490335A (zh) * 2019-08-07 2019-11-22 深圳前海微众银行股份有限公司 一种计算参与者贡献率的方法及装置
CN110910158A (zh) * 2019-10-08 2020-03-24 深圳逻辑汇科技有限公司 联邦学习收入分配方法和系统
CN111222646A (zh) * 2019-12-11 2020-06-02 深圳逻辑汇科技有限公司 联邦学习机制的设计方法、装置和存储介质

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
YANG QIWEN, CAI LIANG, XUE YUN-CAN: "A Survey of Differential Evolution Algorithms", PATTERN RECOGNITION AND ARTIFICIAL INTELLIGENCE, vol. 21, no. 4, 31 August 2008 (2008-08-31), pages 506 - 513, XP055925920 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116384502A (zh) * 2022-09-09 2023-07-04 京信数据科技有限公司 联邦学习中参与方价值贡献计算方法、装置、设备及介质
CN116384502B (zh) * 2022-09-09 2024-02-20 京信数据科技有限公司 联邦学习中参与方价值贡献计算方法、装置、设备及介质

Also Published As

Publication number Publication date
CN112329947A (zh) 2021-02-05

Similar Documents

Publication Publication Date Title
WO2022088541A1 (zh) 一种基于差分进化的联邦学习激励方法和系统
Liu et al. Fedcoin: A peer-to-peer payment system for federated learning
Wang et al. Infedge: A blockchain-based incentive mechanism in hierarchical federated learning for end-edge-cloud communications
Zhan et al. Incentive mechanism design for federated learning: Challenges and opportunities
Lotfi et al. Economics of quality sponsored data in non-neutral networks
TW202040484A (zh) 基於區塊鏈的資料處理系統、方法、計算設備及儲存媒體
CN116627970A (zh) 一种基于区块链和联邦学习的数据共享方法及装置
Chen et al. A Mechanism Design Approach for Multi-party Machine Learning
Chen et al. Dim-ds: Dynamic incentive model for data sharing in federated learning based on smart contracts and evolutionary game theory
Wang et al. Smart contract-based caching and data transaction optimization in mobile edge computing
Di et al. The consensus equilibria of mining gap games related to the stability of Blockchain Ecosystems
CN110991789A (zh) 置信区间的确定方法和装置、存储介质及电子装置
EP3696766A1 (en) Program, system and method for determining credit worthiness of a user
Hosseinalipour et al. Options-based sequential auctions for dynamic cloud resource allocation
Ma et al. Monopoly pricing with participation‐dependent social learning about quality of service
WO2023108761A1 (zh) 监控业务带宽分配方法、装置、电子设备及存储介质
CN114116705A (zh) 联合学习中确定参与方贡献值的方法及装置
CN113157434B (zh) 一种横向联邦学习系统用户节点的激励方法及系统
CN113158550B (zh) 一种联邦学习方法、装置、电子设备及存储介质
Jiang et al. A reward response game in the blockchain-powered federated learning system
US11632588B2 (en) Measuring the performance of a peer-managed content distribution network
TWI616107B (zh) 以機器學習為基礎的時依智慧計費架構
CN114116740A (zh) 用于联合学习中确定参与方贡献度的方法及装置
Zeng et al. A hybrid pricing mechanism for data sharing in P2P-based mobile crowdsensing
Ajgaonkar et al. A blockchain approach for exchanging machine learning solutions over smart contracts

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21884285

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21884285

Country of ref document: EP

Kind code of ref document: A1