PEARL framework enhances privacy-utility control in human-centric CPS

Researchers have developed PEARL, a novel framework for controlling privacy and utility in human-centric Cyber-Physical Systems (CPS) that utilize deep reinforcement learning. PEARL employs a dual-path Early-Exit Deep Q-Network to structurally limit the descriptive power of shared actions without injecting noise. This is achieved by using Utility Confidence Labels (UCL) and Privacy Confidence Labels (PCL) to ensure action quality and control privacy leakage. The framework has been validated on smart-home HVAC and VR classroom systems, demonstrating a significant reduction in adversarial state-inference accuracy with a controlled utility cost. AI

IMPACT PEARL offers a new method for managing privacy-utility tradeoffs in edge AI applications, potentially improving user trust and data security in smart environments.

RANK_REASON The item is a research paper detailing a novel framework for privacy-utility control in cyber-physical systems using deep reinforcement learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• Cyber-Physical Systems
• deep reinforcement learning
• Early-Exit Deep Q-Network
• PEARL
• Personalized Early-exit Adaptive Reinforcement Learning
• Privacy Confidence Labels
• Salma Elmalaki
• Utility Confidence Labels

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →