Loading...

Privacy Preserving Learning with Adjustable Utility Privacy Trade-off

Jamshidi, Mohammad Ali | 2024

8 Viewed
  1. Type of Document: Ph.D. Dissertation
  2. Language: Farsi
  3. Document No: 56820 (05)
  4. University: Sharif University of Technology
  5. Department: Electrical Engineering
  6. Advisor(s): Aref, Mohammad Reza
  7. Abstract:
  8. The rapid evolution of artificial intelligence (AI) technologies has led to the widespread adoption of AI systems in diverse research and industrial fields. Deep neural networks, at the forefront of AI's power, demonstrate high performance by leveraging large volumes of training data. However, acquiring such vast amounts of data requires collaboration among individual data owners, who may have concerns about privacy. To address these concerns, various privacy-preserving methodologies have been proposed. These methodologies share a common goal of striking a balance between preserving privacy and maintaining data utility. This study aims to explore and analyze these privacy protection strategies, while also examining the concepts of privacy, data utility, and the trade-off between the two. As machine learning (ML) plays a crucial role in this field, its characteristics, prerequisites, and tools are also examined. This exploration helps identify and categorize common threats and attacks relevant to the discussion. The focus then shifts to ML-based privacy-preserving methodologies, where the strengths and weaknesses of these approaches are scrutinized. Two autoencoder-based techniques are proposed, leveraging data compression and separating confidential and non-confidential features. These techniques enhance useful non-confidential features while obfuscating the confidential ones, allowing the reconstruction of the obfuscated data from the modified features. In addition to offering superior performance in terms of the privacy-utility trade-off, the proposed structures have several advantages, including reduced complexity and eliminating the need for the data owner to train the model. Remarkably, these methods allow the data owner to adjust the utility-privacy level and are compatible with various data types
  9. Keywords:
  10. Autoencoder ; Deep Learning ; Collaborative Federated Learning ; Virtualization Obfuscation ; Dataset Publishing ; Utility Privacy Trade-off ; Privacy Preserving Learning (PPML)

 Digital Object List

 Bookmark

No TOC