|
内容記述 |
A multi-party computation (MPC) allows η parties to compute an agreed-upon func- <br />tion of their inputs and every party learns the correct function output. To solve a <br />multi-party computation problem (MPCP), the participants may need to share their <br />private data (inputs) between one another, resulting in data privacy loss. The key <br />research issue that has been addressed in this thesis is - how to solve multi-party <br />computation problems without disclosing anyone's private data to others. <br /><br /> Firstly, by studying and analyzing the traditional computational models, we have <br />devised a privacy loss model for multi-party computation problems and proposed a <br />novel metric, called the Min privacy metric, for quantitatively measuring the amount <br />of data privacy loss in solving the MPCPs. Then, we have presented a mobile agent- <br />based scheduling algorithm that applies pseudonymization technique to reduce data <br />privacy loss. Finally, we have proposed the security system design, including security <br />policies and security architecture, of an agent server platform for enhancing data <br />privacy protection while solving the MPCPs.<br /><br /> The privacy loss model has identified three factors affecting the amount of privacy <br />loss in solving the MPCPs: (1) the fraction ofprivate data which is shared with others, <br />(2) the probability of associating the shared private data with the data subject, and <br />(3) the probability of disclosing the shared private data to unauthorized parties.<br />Privacy loss can be reduced by any mechanisms which reduces the values of any <br />of the three factors. The proposed Min privacy metric accounts for the number of <br />participants that lose their private data and the amount of private data disclosed to <br />unauthorized parties, regardless of how many parties they are revealed to. <br /><br /> Existing scheduling algorithms aim for a global objective function. As a result,<br />they incur performance penalties in computational complexity and data privacy. This <br />thesis describes a mobile agent-based scheduling scheme called Efiicient and Privacy-<br />aware Meeting Scheduling (EPMS), which results in a tradeoff arnong complexity,<br />privacy, and global utility for scheduling multiple events concurrently. We have intro- <br />duced multiple criteria for evaluating privacy in the meeting scheduling problem. A <br />common computational space has been utilized in EPMS for reducing the complexity <br />and pseudonymization technique has been applied to reduce the privacy loss in the <br />scheduling problem. The analytical results show that EPMS has a polynomial time <br />computational complexity. In addition, simulation results show that the obtained <br />global utility for scheduling multiple meetings with EPMS is close to the optimal <br />level and the resulting privacy loss is less than for those in extsting algorithms. <br /> Cryptography-based aJgorithms for MPCPs are either too complex to be used <br />practically or applicable only to the specific applications for which they have been <br />developed. In addition, traditional (non-cryptography-based) algorithms do not pro- <br />vide good privacy protection for MPCPs. We have proposed a novel privacy pro- <br />tection mechanism in which MPCPs are solved by mobile agents using traditional <br />algorithms at an agent server platform, called isolated Closed-door One-way Plat- <br />form (iCOP). The participating mobile agents are trapped into iCOP where they <br />are allowed to share their private information to solve the problem using traditional <br />algorithms. However, they are protected from disclosing the shared private infor- <br />mation to the outside world. The enforcement of the security policies protects the <br />participating agents from sending anything other than the computational result to <br />the users. The security and privacy analysis illustrates that the proposed mechanism <br />provides very good privacy protection if the participants solve the problem with dis- <br />tributed algorithms and can provide complete privacy protection if the participants <br />exchange inputs within the iCOP and each of them solve the problem with centralized <br />algorithms. Finally, experimental evaluation shows that the proposed agent platform <br />security system significantly enhances privacy protection while solving many MPCPs <br />with traditional algorithms.<br /> |