End-to-end Automated Deep Neural Network Optimization for PPG-based Blood Pressure Estimation on Wearables

Summary

This recent (April 2026) arXiv paper extends the neural architecture search (NAS) paradigm for PPG-based blood pressure estimation with an end-to-end automated pipeline covering architecture search, structural pruning, and quantization jointly optimised for wearable hardware constraints. Unlike prior NAS approaches that optimise stages sequentially, this end-to-end approach allows the search to consider the full deployment footprint simultaneously. The result is models that are 10–100× smaller than manual designs while maintaining competitive BP MAE, enabling real-time inference on MCU-class devices.