Ce Feng and Parv Venkitasubramaniam propose RQP-SGD, a privacy-preserving approach that combines differential privacy and randomized quantization for low-memory edge devices in their paper titled RQP-SGD: Differential Private Machine Learning through Noisy SGD and Randomized Quantization.

• Applicability: Ideal for deploying ML models on IoT devices that handle sensitive data.
• Utility Convergence: Studies show its effectiveness over deterministic quantization for convex objectives.
• Enhanced Privacy: Provides a measured privacy guarantee in machine learning with quantized model parameters.

The paper emphasizes the feasibility and practicality of RQP-SGD, showing its potential in managing the delicate balance between model accuracy and privacy protection on the edge of the computational spectrum.