About the book
Many recent innovations are directed to robust anonymization of data with specified reidentification risk. Differential privacy (DP) minimizes privacy impact in shareable microdata by injecting noise beyond sampling or by randomly generating synthetic data that is distributionally close to the actual dataset. Additionally, there is great interest in achieving privacy-preserving versions of statistical and machine learning algorithms. With increased emphasis on privacy, DP has growing prevalence in data engineering and analysis. However, many challenges remain. For example, DP operating on real-world data may not preserve enough information utility after DP filtering and augmentation. There is a need to strike a reasoned balance between privacy and precision of data. Some critical questions for statisticians, data scientists, and engineers include:
● Can varying levels of privacy assurance be built into DP models, to accommodate varying willingness to share personal information than others, especially if this results in more utility for them.
● How is differential privacy able to address evolving statistical distributions of model features in streaming data use-cases?
● How shall we statistically account for excess randomness introduced by privacy-preserving methods?
This book aims to provide an account of the current state-of-the-art in DP and developments in this strategically important area.