2P3FL:A Novel Approach for Privacy Preserving in Financial Sectors Using Flower Federated Learning

The increasing data pool in finance sectors forces machine learning(ML)to step into new complications.Banking data has significant financial implications and is confidential.Combining users data from several organizations for various banking services may result in various intrusions and privacy leakages.As a result,this study employs federated learning(FL)using a flower paradigm to preserve each organization’s privacy while collaborating to build a robust shared global model.However,diverse data distributions in the collaborative training process might result in inadequate model learning and a lack of privacy.To address this issue,the present paper proposes the imple-mentation of Federated Averaging(FedAvg)and Federated Proximal(FedProx)methods in the flower framework,which take advantage of the data locality while training and guaranteeing global convergence.Resultantly improves the privacy of the local models.This analysis used the credit card and Canadian Institute for Cybersecurity Intrusion Detection Evaluation(CICIDS)datasets.Precision,recall,and accuracy as performance indicators to show the efficacy of the proposed strategy using FedAvg and FedProx.The experimental findings suggest that the proposed approach helps to safely use banking data from diverse sources to enhance customer banking services by obtaining accuracy of 99.55%and 83.72%for FedAvg and 99.57%,and 84.63%for FedProx.

基于FedAvg的山林火灾扑救行动兵力预测模型构建方法

以夏秋季极端高温山林火灾扑救行动为研究对象,以分布式机器学习为理论基础对任务中气温及机动兵力的建模和预测进行研究。首先提出一种基于联邦平均算法(Federal Average Algorithm,FedAvg)的模型构建方法,从更贴近任务实际、更加精细的角度对各任务方向的最高气温及机动兵力数量进行定量预测;其次通过引接政府公共资源平台及作战数据库中多区域气温和机动兵力,在各数据客户端不互传数据的情况下,通过聚合不同客户端参数共同训练全局模型达到预测目的,为各数据源无法共享环境下分析数据、使用数据提供理论支撑。

异步联邦学习中隔代模型泄露攻击及防治方法

联邦学习已成为数据孤岛背景下知识共享的成功方案。随着梯度逆向推理等新式攻击手段的问世,联邦学习的安全性再度面临新挑战。针对联邦学习可能存在参与者恶意窃取其它客户端梯度信息的风险,提出一种异步联邦学习框架下的隔代模型泄露攻击方式:利用中心服务器“接收则聚合”的特点,多名恶意客户端可按照特定更新顺序,通过隔代版本的全局模型差异逆向计算其他客户端的模型更新数据,从而窃取对方的模型。针对此问题,提出基于α-滑动平均的随机聚合算法。首先,中心服务器每次收到客户端的模型更新后,将其与从最近α次聚合中随机选出的全局模型进行聚合,打乱客户端的更新顺序;其次,随着全局迭代次数增加,中心服务器对最近α次聚合的全局模型进行滑动平均,计算出最终全局模型。实验结果表明,与异步联邦学习方法相比,FedAlpha方法有效降低隔代模型泄露攻击的可能性。

基于联邦平均的异常检测拜占庭容错算法

传统联邦学习训练模型时假定所有参与方可信,但实际场景存在恶意参与方或恶意攻击模型,现有的联邦学习算法面对投毒攻击时,存在模型性能严重下降的问题。针对模型投毒问题,本文提出一种基于联邦平均(federated averaging,Fedavg)与异常检测的联邦检测算法——FedavgCof,该算法考虑到所有参与方之间的差异对比,在中心服务器和本地模型之间添加异常检测层,通过基于聚类的本地异常检测因子(cluster-based local outlier factor,COF)异常检测算法剔除影响模型性能的异常参数,提升模型鲁棒性。实验结果表明,虽然新型投毒方式攻击性更强,但是FedavgCof能够有效防御投毒攻击,降低模型性能损失,提高模型抗投毒攻击能力,相较于Median和模型清洗算法平均提升精度达到10%以上,大幅提升了模型的安全性。

EMFedAvg——基于EMD距离的联邦平均算法

信息技术给人们生活带来便利的同时也会泄露个人隐私.联邦学习是一种可以保护数据隐私的机器学习技术,不同于现有的机器学习方法,联邦学习中数据不出参与方本地,通常面临着数据非独立同分布的问题(non-identically Independently Distributions, non-IID),因而现有的机器学习方法在联邦学习non-IID问题上效果大大降低.文章针对联邦学习中的non-IID问题,在联邦平均算法的基础上进行改进,对MNIST数据集进行non-IID划分并分发到各参与方,计算各参与方数据的EMD(Earth Mover’s Distance, EMD)距离,以四分位距为上界,主动去掉EMD距离过大的参与方以保证联邦整体的效果.实验结果表明,文章采用的方法比联邦平均算法提高了约5%的准确率,减少了联邦学习训练过程的通信开销,提高了整体效率,引入EMD距离还可以为衡量各参与方的贡献值提供度量依据.

基于一种新的联邦优化算法的信用风险预测方法

ResNet深度神经网络模型与联邦学习融合建模,实现信用风险预测并保证用户数据的隐私安全。模型准确率和精准度可达89.67%和94.22%。首次采用联邦优化算法-局部全局联合平均算法(LG-FEDAVG)优化训练过程,对贷款用户的信用风险建模。研究表明,LG-FEDAVG算法在保证信用风险预测效果的情况下产生的通信成本占通信总参数的3.22%。

基于Bi-SRNN的联邦学习区域电力短期负荷预测模型

随着配电网终端需求多样化和清洁能源的大规模接入,对区域电力负荷的准确预测变得至关重要。在电力市场化改革背景下,客户端倾向于用电信息保存在本地以确保隐私安全。利用天气数据和历史负荷数据,提出面向区域客户端隐私保护的联邦学习双向叠加循环神经网络负荷预测框架。根据短期电力负荷长序列数据之间的强关联性建立基于双向叠加循环神经网络的负荷预测模型。利用联邦平均算法构建基于联邦学习的区域负荷预测框架,将多个利用不同区域客户端负荷数据训练得到的双向叠加循环神经网络的模型进行融合,反复迭代获得全局模型。采用某市96组实时区域电力负荷公开的数据集,对该模型在区域客户端不共享负荷数据条件下的训练效果进行测试,结果表明,所构建模型具有较低的训练耗时和较高的预测精度。

利用ARIMA(自回归移动平均)模型对跑道侵入事件的分析及预测

综合运用具有相当精度的时间序列分析方法,建立美国民航运输安全中的机场跑道侵入的AR IMA(自回归移动平均)模型,克服了样本空间总是有限带来的不足,揭示出民航跑道侵入的动态变化规律,并对未来美国民航跑道侵入事故发生次数进行较准确的预测,为我国民航部门科学地制定飞行计划、人员培训、提高安全管理水平,提供可靠的依据。

基于时间序列分析的自适应联邦滤波算法

以信息源的视角审视联邦滤波信息分配系数的问题,提出基于时间序列分析的自适应联邦滤波算法。通过引入导航传感器历史数据作为信息基础,以自回归滑动平均模型对其进行分析,使信息分配系数与导航传感器工作状态和环境状态相适应,提高系统性能。将此算法应用于组合导航系统,通过仿真实验对比,新的算法比经典联邦滤波算法定位精度提高40%,比其他的自适应联邦滤波算法位置误差标准差、速度误差标准差分别平均提高19.74%和30.67%,说明该算法具有良好性能。

黄金价格影响因素的实证分析

以研究现阶段黄金价格主要影响因素为目的,先结合黄金价格与影响变量以往走势对影响因素进行定性分析,然后基于相关理论筛选出现阶段影响黄金价格的主要因素。选取2001年1月—2016年3月的月度数据,建立向量自回归模型进行实证分析。研究发现:该阶段通货膨胀率与黄金价格呈不相关关系;美元指数和美国联邦基金利率与黄金价格呈负相关关系,其中美元指数对黄金价格影响最为显著;道琼斯工业指数和原油价格与黄金价格呈正相关关系。

基于联邦学习的无线任务:数据非IID一定影响性能?

作为一种分布式训练框架,联邦学习在无线通信领域有着广阔的应用前景,也面临着多方面的技术挑战,其中之一源于参与训练用户数据集的非独立同分布(Independent and identically distributed,IID)。不少文献提出了解决方法,以减轻户数据集非IID造成的联邦学习性能损失。本文以平均信道增益预测、正交幅度调制信号的解调这两个无线任务以及两个图像分类任务为例,分析用户数据集非IID对联邦学习性能的影响,通过神经网络损失函数的可视化和对模型参数的偏移量进行分析,尝试解释非IID数据集对不同任务影响程度不同的原因。分析结果表明,用户数据集非IID未必导致联邦学习性能的下降。在不同数据集上通过联邦平均算法训练得到的模型参数偏移程度和损失函数形状有很大的差异,二者共同导致了不同任务受数据非IID影响程度的不同;在同一个回归问题中,数据集非IID是否影响联邦学习的性能与引起数据非IID的具体因素有关。

改进的联邦加权平均算法

针对基于层次分析改进的联邦平均算法在计算其数据质量时存在主观因素的影响,提出改进的联邦加权平均算法,从数据质量的角度来处理多源数据。首先,将训练样本划分为预训练样本与预测试样本;然后,使用初始全局模型在预训练数据上的精度作为该数据源的质量权重;最后,将质量权重引入到联邦平均算法中,重新进行全局模型中权重更新。仿真结果表明,在均等分割的数据集与非均等分割的数据集上,改进的联邦加权平均算法训练的模型与传统联邦平均算法训练的模型相比,准确率最高分别提升了1.59%和1.24%;改进的联邦加权平均算法训练的模型与传统整合多方数据再训练的模型相比,虽然准确率略有下降,但数据与模型的安全性有所提升。

基于层析分析改进的联邦平均算法

联邦平均(Fedavg)算法采用权重更新来更新全局模型,该算法在权重更新时仅考虑每个客户端数据量的大小,未考虑数据质量对模型的影响。针对该问题,文中提出了基于层次分析改进的联邦平均算法,首次从数据质量的角度来处理多源数据。首先采用熵权法计算数据中各属性的重要度,并将其作为层次分析中准则层的数值,计算每个客户端数据的质量,然后结合客户端数据量的大小,重新计算全局模型中的权重。仿真实验的结果表明,对于中小型数据集而言,使用支持向量机训练的模型准确度最高,达到了85.7152%;对于大型数据集而言,采用随机森林训练的模型准确率最高,达到了91.9321%。与传统联邦平均方法相比,所提方法在中小数据集上准确率提升了3.5%,在大数据集上提升了1.3%,能够在提升模型准确率的同时提高数据与模型的安全性。

A Weighted Average Consensus Approach for Decentralized Federated Learning

Federated learning(FedL)is a machine learning(ML)technique utilized to train deep neural networks(DeepNNs)in a distributed way without the need to share data among the federated training clients.FedL was proposed for edge computing and Internet of things(IoT)tasks in which a centralized server was responsible for coordinating and governing the training process.To remove the design limitation implied by the centralized entity,this work proposes two different solutions to decentralize existing FedL algorithms,enabling the application of FedL on networks with arbitrary communication topologies,and thus extending the domain of application of FedL to more complex scenarios and new tasks.Of the two proposed algorithms,one,called FedLCon,is developed based on results from discrete-time weighted average consensus theory and is able to reconstruct the performances of the standard centralized FedL solutions,as also shown by the reported validation tests.

Assessment of the Demarcation Method for Federal Riverine and Accreted Lands: Case Study of the Rio De Janeiro State Section of the South Paraíba River

This study analyzes the demarcation method of riverine and accreted land of the Brazilian Federal Heritage Department and proposes the incorporation of the flow rate corresponding to the recurrence interval of two years, as recommended by the State Environmental Institute of the state of Rio de Janeiro. The case study of the Rio de Janeiro section of the Paraiba do Sul River was investigated, and the results indicate that the Federal Heritage Department’s method does not consider the ongoing anthropization of the river, caused mainly by the construction and operation of hydroelectric plants. In addition, it was observed that the limnimetric scales of the studied gauging stations are influenced by constant changes in the riverbed and by riverbank occupation, making it difficult to estimate the ordinary flood level. The study concludes by suggesting the adoption of a flow rate with a recurrence interval of two years and the simulation of the runoff conditions for demarcation of the average ordinary flood line.

The Essence of Optimism of The Economists and Investors with Regard to the Economy of Russia

The gross domestic product of Russia,expressed in US dollars,indicates problems in the Russian economy associated with the decline in oil prices on the world energy market and the consequences of the sanctions of the United States and the European Union against Russia.The crisis situation of the Russian economy has a negative impact on the income of the population of country,represented mainly by wages.However,an economist or investor may be optimistic about Russian economic development in the medium term.This optimism is related to the economic policy of the United States.The expansion of the United States economy within the global space,based on economic growth,requires maintaining inflation within the target level and weakening the US dollar.These tasks are solved with the help of soft monetary policy of the US Federal Reserve System.The reduction of interest rates by the US Federal Reserve System against the background of inflation of the target level and the devaluation of the US dollar will contribute to economic growth in the United States,because it will lead to the depreciation of public debt,lower consumption of imports,increase in exports and trade balance,growth of production,income,consumption.The economic policy of the United States,which contributes to the devaluation of the US dollar,will also reduce the US dollar against the ruble.The optimistic view of investors-economists on the Russian economy is due to a significant strengthening of the ruble against the US dollar.Consequently,in the medium term,the gross domestic product and wages of citizens of Russia,expressed in US dollars,will significantly increase,and the purchasing power of the national currency of the country will also increase.This growth may continue until the next election of a new President of the United States in november 2020.After the election of the new President of the United States,there is a high probability of sanctions against Russia and of decline in oil prices in the world energy market in accordance with the future economic policy of the United States–two main reasons for the sharp strengthening of the US dollar against the ruble,which will cause a deeper economic crisis in Russia in the medium and long term.