Optimization and Deployment of Deep Neural Networks for PPG-based Blood Pressure Estimation Targeting Low-power Wearables

Summary

This paper from Politecnico di Torino and EPFL addresses the practical bottleneck for PPG-BP wearable deployment: state-of-the-art DNN models are too large for microcontroller (MCU)-class devices. The authors present an end-to-end neural architecture search (NAS) pipeline with structured pruning and int8 quantization specifically tuned for wearable MCUs. The approach explores the accuracy-vs-compute Pareto frontier, identifying models that achieve competitive BP MAE while fitting within the memory and compute budgets of devices like the STM32.