Research On Secure Outsourcing Schemes Of Two Scientific Computing Problems

Cloud computing is an internet-based delivery model, and can provide various on-demand services(e.g. storage, and outsourcing computation services) for the enterprises and individuals. As one of fundamental services of cloud computing, outsourcing computation can provide the resource-constrained clients with massive computational power. In the outsourcing computation paradigm, the clients can outsource their expensive computational workloads to the cloud server, and then the server can return the computing results. Thus, the clients can avoid their huge capital in human resources and computing devices deployment.Despite of the enormous benefits, outsourcing computation inevitably suffers from some new security challenges. The outsourcing data always contain some sensitive privacy information of the clients, such as: purchase history, research data, etc. But the untrusted cloud server would like to attempt to collect some sensitive information. Thus privacy protection is one of security challenges for outsourcing computation. Moreover, for cost savings, the cloud server may reduce their computing time and then return a result which is low in accuracy or invalid to the client. Hence, the verifiability of the result returned is another security challenge. In short, it is significant to design an outsourcing computation protocol that can not only protect the sensitive information, but also verify the validity of the result returned.Non-negative matrix factorization and convex quadratic programming are two kinds of significant scientific computing problems and have been widely applied in the field of image processing, text document clustering and data mining, etc. The resource-constrained clients need to consume huge computational cost to solve a large scale problem. It is an economical solution for the clients to outsource these problems to the powerful cloud server.This thesis investigates secure outsourcing of non-negative matrix factorization and convex quadratic programming. The main contributions of this paper are as follows:1. For the large-scale non-negative matrix factorization problem, we propose a secure and verifiable outsourcing computation protocol. In our protocol, all the input matrix and the output result after factorization are regard as the privacy of the client. In order to protect privacy of the input and output data, the permutation technique is employed to transform an original problem into a new and random one, which is sent to the cloud server. The cloud server receives new problem and then return the computing results to the client, and the client can obtain the results of the original problem through some reverse transformations on the results returned. In the phase of verification, to reduce the verification cost, the matrix 1-norm technique is utilized to verify the result returned from the cloud server. Theoretical analyses and experimental results show that our scheme can bring great computation savings for the resource-constrained client.2. For the large-scale convex quadratic programming problem, we first analysis the advantage and disadvantage of Zhang's protocol, and then present a new outsourcing computation protocol. In the new protocol, the client utilizes the permutation technique to transform the original problem into a new and random optimization problem. The cloud server receives and solves it, and the client verifies the results returned. Security analysis indicates that the proposed protocol can protect the privacy of input/output data, and detect the misbehavior of the cloud server with probability 1(optimal) under the malicious model. Experimental results show that our protocol has a comparative advantage of the transformation and verification efficiency than Zhang's protocol.