Privacy-Utility Trade-Off for Time-Series with Application to Smart-Meter Data

We consider the online setting where a user would like to continuously release a time-series of data that is correlated with his private data, to a service provider in the hope of deriving some utility. Due to correlations, the continual observation of the released time-series puts the user at risk of inference of his private data by an adversary. To protect the user from inference attacks on his private data, the time-series is randomized prior to its release according to a probabilistic privacy mapping. The privacy mapping should be designed in a way that balances privacy and utility requirements over time. Our contributions are threefold. First, we formalize the framework for the design of utility-aware privacy mappings for time-series data, under both online and batch models. We provide a sequential scheme that allows to design online privacy mappings at scale, that account for privacy risk from the history of released data and future releases to come. Second, we prove the equivalence of the optimal mappings under the batch and the online models, in the case where the time-series samples are independent across time. We further show that there exists a gap between optimal batch and online privacy mappings when certain conditions are not satisfied. Finally, we evaluate the performance of the framework over synthetic and real-world time-series data. In particular, we show that smart-meter data can be randomized for privacy purposes to prevent disaggregation of per-device energy consumption, while preserving the utility.