Privacy-preserving and reputation system in distributed computing with untrusted parties

In today's society, people are impossible to avoid using electronic device and accessing network. Just like safety of food and water, the security of personal information should be pay attention by every person in case of malicious intention. The personal data information can be used to help medical institute and network company to optimize their research result and industrial structure. However, some sensitive personal data cannot be shared due to the human factors of untrusted parities. In this dissertation, I study privacy issues in data mining and provide a reputation system for wireless network with untrusted parities.;Data mining with recent technical advances has become more important in Big Data Era. However, as the data continuously growing, people will care their privacy increasing more. So how to protect the privacy is the most important part when the data involving people's sensitive information such as genome sequence. The challenge in privacy-preserving data mining is avoiding the invasion of personal data privacy. Secure computation provides a solution to this problem.;In this thesis, first I present a privacy-preserving method for training Restricted Boltzmann Machine. The method guarantees the privacy-protection in semi-honest model. Second, I design and analyze a symmetric-key based privacy-preserving scheme for mining support counts. Third, to protect data privacy in clustering, I study the method of privacy-preserving clustering using representatives over arbitrarily partitioned data. Fourth, I introduce Scorpio, a simple and convenient add-in for privacy preserving logrank test.;Finally, I propose a incentive consideration to the study of secure computation by presenting a reputation system in wireless network. The system can provide an incentive for the misbehaving nodes to behave honestly. Experiments have shown that our reputation system is very efficient in detecting misbehaving nodes and increasing average throughput in the whole network.