Privacy via Maintaining Small Similitude Data for Big Data Statistical Representation

Despite its attractiveness, Big Data oftentimes is hard, slow and expensive to handle due to its size. Moreover, as the amount of collected data grows, individual privacy raises more and more concerns: “what do they know about me?” Different algorithms were suggested to enable privacy-preserving data release with the current de-facto standard differential privacy. However, the processing time of keeping the data private is inhibiting and currently not practical for every day use. Combined with the continuously growing data collection, the solution is not seen on a horizon. In this research, we suggest replacing the Big Data with a much smaller similitude model. The model “resembles” the data with respect to a class of query. The user defines the maximum acceptable error and privacy requirements ahead of the query execution. Those requirements define the minimal size of the similitude model. The suggested method is demonstrated by using a wavelet transform and then by pruning the tree according to both the data reduction and the privacy requirements. We propose methods of combining the noise required for privacy preservation with noise of similitude model, that allow us to decrease the amount of added noise thus, improving the utilization of the method.