云计算环境下安全的极限学习机外包优化部署机制
|
摘要
为应对大数据提出的挑战,在对海量数据进行处理分析时,需要大量计算资源与高效的机器学习算法。论文对极限学习机(Extreme Learning Machine,ELM)的云计算外包机制进行研究,提出了云计算环境下安全的ELM外包优化部署机制。该优化部署机制将ELM显式地分为私有部分和公有部分,可有效减少训练时间,并确保算法输入与输出的安全性。私有部分主要负责随机参数的生成以及隐层输出矩阵、输出权重矩阵与中间矩阵的求解。公有部分外包到云计算服务器中,由云计算服务商负责ELM算法中计算量最大的中间矩阵求逆操作。该中间矩阵的逆亦可作为证据以验证结果的正确性和可靠性。在CIFAR-10数据集上的实验结果表明,所提出的优化部署机制可以有效地减轻用户的计算量。
To address the challenge of big data,abundant computing resource and efficient machine learning algorithm are necessary for processing and analyze big data. This paper focuses on outsourcing Extreme Learning Machine(ELM)in cloud,andpresents an optimization mechanism in order to improve training speed of ELM,as well as ensure the confidentiality and privacy ofinput and output. The proposed optimization mechanism explicitly divides ELM into two parts,a public part and a private part. Theprivate part is executed locally,consisting of generation of random parameters,calculation of output matrix of hidden layer,interme-diate matrix,and output weight matrix. The public part is executed in cloud that is mainly responsible for calculating the inverse ofthe intermediate matrix,which is the heaviest operation computationally. The inverse also serves as the correctness and soundnessproof in result verification. We analyze the confidentiality and communication cost theoretically and the experimental results demonstrate that the proposed mechanisms can effectively release customers from heavy computations.
To address the challenge of big data,abundant computing resource and efficient machine learning algorithm are necessary for processing and analyze big data. This paper focuses on outsourcing Extreme Learning Machine(ELM)in cloud,andpresents an optimization mechanism in order to improve training speed of ELM,as well as ensure the confidentiality and privacy ofinput and output. The proposed optimization mechanism explicitly divides ELM into two parts,a public part and a private part. Theprivate part is executed locally,consisting of generation of random parameters,calculation of output matrix of hidden layer,interme-diate matrix,and output weight matrix. The public part is executed in cloud that is mainly responsible for calculating the inverse ofthe intermediate matrix,which is the heaviest operation computationally. The inverse also serves as the correctness and soundnessproof in result verification. We analyze the confidentiality and communication cost theoretically and the experimental results demonstrate that the proposed mechanisms can effectively release customers from heavy computations.
引文
[1]C. Wang,K. Ren,and J. Wang. Secure and practical out-sourcing of linear programming in cloud computing[C]//Proceedings of IEEE INFOCOM 2011. 2011:820-828.
[2]G.-B. Huang,Q.-Y. Zhu,and C.-K. Siew. Extreme learn-ing machine:A new learning scheme of feedforward neu-ral networks[C]//Proceedings of International Joint Confer-ence on Neural Networks(IJCNN2004),vol. 2. Buda-pest,Hungary:2004:985-990.
[3]G.-B. Huang,Q.-Y. Zhu,and C.-K. Siew. Extreme learning machine:Theory and applications[J]. Neurocomput-ing,2006,70(1):489-501.
[4]G.-B. Huang,H. Zhou,X. Ding,et al. Extreme learningmachine for regression and multiclass classification[J].IEEE Transactions on Systems,Man,and Cybernetics,Part B:Cybernetics,2012,42(2):513-529.
[5]Jiarun Lin,Jianping Yin,Zhiping Cai,et al. A Secureand Practical Mechanism for Outsourcing ELMs in CloudComputing[C]//IEEE Intelligent Systems,2013,28(6):35-38.
[6]Q. He,T. Shang,F. Zhuang,et al. Parallel extreme learn-ing machine for regression based on MapReduce[J]. Neu-rocomputing,2013,102(2):52-58.
[7]M. van Heeswijk,Y. Miche,E. Oja,et al. GPU-acceler-ated and parallelized ELM ensembles for large-scale re-gression[J]. Neurocomputing,2011,74(16):2430-2437.
[8]D. Serre. Matrices:Theory and applications[M]. Spring-er,2010.
[9]Y. Cheng,Z. Wang,J. Ma,et al. Efficient revocation inciphertext-policy attribute-based encryption based crypto-graphic cloud storage[J]. Journal of Zhejiang Universi-ty-Science C(Computers&Electronics),2013,14:85-97.
[10]P. Shi,H. Wang,H. Yin,et al. Dependable Deploy-ment Method for Multiple Applications in Cloud ServicesDelivery Network[J]. China Communications,2011,8:65-75.
[11]A. Krizhevsky and G. Hinton. Learning multiple layers offeatures from tiny images[M]. Master's thesis,Depart-ment of Computer Science,University of Toronto,2009.
[2]G.-B. Huang,Q.-Y. Zhu,and C.-K. Siew. Extreme learn-ing machine:A new learning scheme of feedforward neu-ral networks[C]//Proceedings of International Joint Confer-ence on Neural Networks(IJCNN2004),vol. 2. Buda-pest,Hungary:2004:985-990.
[3]G.-B. Huang,Q.-Y. Zhu,and C.-K. Siew. Extreme learning machine:Theory and applications[J]. Neurocomput-ing,2006,70(1):489-501.
[4]G.-B. Huang,H. Zhou,X. Ding,et al. Extreme learningmachine for regression and multiclass classification[J].IEEE Transactions on Systems,Man,and Cybernetics,Part B:Cybernetics,2012,42(2):513-529.
[5]Jiarun Lin,Jianping Yin,Zhiping Cai,et al. A Secureand Practical Mechanism for Outsourcing ELMs in CloudComputing[C]//IEEE Intelligent Systems,2013,28(6):35-38.
[6]Q. He,T. Shang,F. Zhuang,et al. Parallel extreme learn-ing machine for regression based on MapReduce[J]. Neu-rocomputing,2013,102(2):52-58.
[7]M. van Heeswijk,Y. Miche,E. Oja,et al. GPU-acceler-ated and parallelized ELM ensembles for large-scale re-gression[J]. Neurocomputing,2011,74(16):2430-2437.
[8]D. Serre. Matrices:Theory and applications[M]. Spring-er,2010.
[9]Y. Cheng,Z. Wang,J. Ma,et al. Efficient revocation inciphertext-policy attribute-based encryption based crypto-graphic cloud storage[J]. Journal of Zhejiang Universi-ty-Science C(Computers&Electronics),2013,14:85-97.
[10]P. Shi,H. Wang,H. Yin,et al. Dependable Deploy-ment Method for Multiple Applications in Cloud ServicesDelivery Network[J]. China Communications,2011,8:65-75.
[11]A. Krizhevsky and G. Hinton. Learning multiple layers offeatures from tiny images[M]. Master's thesis,Depart-ment of Computer Science,University of Toronto,2009.