Differential Privacy in Geotechnical Engineering

This paper provides a comprehensive review of differential privacy (DP) technology and explores its applications in geotechnical engineering, where data privacy concerns are becoming increasingly critical. As geotechnical engineering often involves the use of sensitive data, such as site-specific information and underground asset assessments, the implementation of privacy-preserving techniques like DP has the potential to safeguard proprietary information while enabling broader data sharing and collaboration. The paper discusses the application of DP techniques, particularly Differentially Private Stochastic Gradient Descent (DP-SGD), to geotechnical data. Specifically, the paper presents two case studies where DP-SGD is applied within a deep learning framework to estimate the vertical scale of fluctuation (SOF) from cone penetration test (CPT) data, and to predict spatially varying soil properties. These studies highlight the trade-offs between privacy guarantees and model accuracy, demonstrating that DP can offer substantial privacy protections with minimal accuracy loss. Finally, this paper emphasizes the need for further research on privacy-aware machine learning models and the development of geotechnical engineering benchmarks for privacy, ensuring that sensitive data is both protected and utilized effectively within the field.