云计算环境下安全的极限学习机外包机制

Secure outsourcing of extreme learning machine in cloud computing

应用程序中涉及到的数据日益扩大且结构日益复杂,使得在大规模数据上运行极限学习机ELM成为一个具有挑战性的任务。为了应对这一挑战,提出了一个在云计算环境下安全和实用的ELM外包机制。该机制将ELM显式地分为私有部分和公有部分,可以有效地减少训练时间,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和一些简单的矩阵计算;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计算Moore-Penrose广义逆的操作。该广义逆也作为证据以验证结果的正确性和可靠性。我们从理论上对该外包机制的安全性进行了分析。在CIFAR-10数据集上的实验结果表明,我们所提出的机制可以有效地减少用户的计算量。

Duo to the enlarging volume and increasingly complex structure of data involved in applications, running the extreme learning machine （ELM） over large-scale data becomes a challenging task. In order to reduce the training time while assuring the confidentiality of ELM＇s input and output, we present a secure and practical outsourcing mechanism for ELM in cloud computing. In this mechanism, we explicitly divide the ELM into two parts： public part and private part. The latter is executed locally to generate random parameters and do some simple matrix computation while the former part is outsourced by cloud computing that is mainly responsible for calculating the Moore-Penrose generalized inverse, the heaviest computational operation. The inverse also serves as the correctness and soundness proof in result verification. We analyze the confidentiality theoretically and the experimental results demonstrate that the proposed mechanism can effectively release customers from heavy computation.