Time Series Discord Discovery Under Multi-party Privacy Preserving

Time series discords are subsequences of a longer time series that are maximally different to all the rest of the time series subsequences. Recent advancements in sensor technology have made it possible to collect enormous amounts of data in real time, sometimes, the unilateral data can't support the data mining task, you must consider the multi-party situation, combining multi-party data into a complete dataset. However, every party has their privacy data which they don't want other party to knows, how to make the privacy data safety and guarantee the accuracy of time series discord discover is the key problem we should concerns. In this paper we propose a construction to discover the time series discord under the multi-party privacy preserving. Our construction is based on the homomorphic encryption and it allows the inputs to be encrypted under different independent public keys. The correctness of our solution is validated by extensive experimental evaluation.