Towards Characterizing Knowledge Distillation of PPG Heart Rate Estimation Models

TL;DR

This paper investigates how large pre-trained PPG-based heart rate estimation models can be distilled into smaller, efficient models suitable for real-time wearable device deployment, analyzing different strategies and their performance scaling.

Contribution

It systematically evaluates four knowledge distillation strategies for PPG heart rate models and characterizes their performance scaling laws for edge deployment.

Findings

Decoupled knowledge distillation (DKD) outperforms other methods.

Performance scales predictably with model size.

Guidelines for building efficient edge models are proposed.

Abstract

Heart rate estimation from photoplethysmography (PPG) signals generated by wearable devices such as smartwatches and fitness trackers has significant implications for the health and well-being of individuals. Although prior work has demonstrated deep learning models with strong performance in the heart rate estimation task, in order to deploy these models on wearable devices, these models must also adhere to strict memory and latency constraints. In this work, we explore and characterize how large pre-trained PPG models may be distilled to smaller models appropriate for real-time inference on the edge. We evaluate four distillation strategies through comprehensive sweeps of teacher and student model capacities: (1) hard distillation, (2) soft distillation, (3) decoupled knowledge distillation (DKD), and (4) feature distillation. We present a characterization of the resulting scaling laws describing the relationship between model size and performance. This early investigation lays the groundwork for practical and predictable methods for building edge-deployable models for physiological sensing.