Privacy-Preserved Frameworks for Secured Data Sharing

The aforementioned two parts require diverse data for modeling and algorithm implementation. In practice, however, data acquisition is often constrained by practical bottlenecks, such as incomplete or noisy datasets, fragmented data storage, and data lackness due to privacy concerns. These limitations hinder the reliable development and broad application of advanced operational strategies. To overcome privacy concerns for data sharing, we developed two privacy-preserving frameworks, i.e., a differential-privacy (DP) based framework and a federated learning based framework, tailored to different levels of data holders’ privacy concerns. For data holders willing to exchange their data, we theoretically quantify the trade-off between DP level and downstream solution accuracy in paper link. For data holders unwilling to exchange data directly, we develop a federated shift-invariant dictionary learning framework that collaboratively learns load patterns across different users without centrally storing private timeseries load data paper link.
See more details in my Presentation and Slides .