PPG-Distill: Efficient Photoplethysmography Signals Analysis via Foundation Model Distillation

2025-09-23

AI Summary

To address the challenge of deploying photoplethysmography (PPG)-based foundation models on resource-constrained wearable devices, this paper proposes PPG-Distill, a novel knowledge distillation framework. Methodologically, it jointly performs morphological distillation—preserving local waveform structures—and rhythmic distillation—capturing inter-segment temporal dependencies—to enable holistic-local collaborative knowledge transfer. Furthermore, it introduces a multi-granularity distillation mechanism operating at the prediction, feature, and segment levels. Evaluated on heart rate estimation and atrial fibrillation detection tasks, the lightweight student model achieves a 21.8% performance gain over the original large teacher model, while accelerating inference by 7× and reducing memory footprint by 19×. These improvements significantly enhance real-time physiological monitoring capabilities at the edge.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Computer Vision: Other Foundations of Computer Vision
• Planning, Routing, and Scheduling: Activity and Plan Recognition

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Photoplethysmography (PPG) is widely used in wearable health monitoring, yet large PPG foundation models remain difficult to deploy on resource-limited devices. We present PPG-Distill, a knowledge distillation framework that transfers both global and local knowledge through prediction-, feature-, and patch-level distillation. PPG-Distill incorporates morphology distillation to preserve local waveform patterns and rhythm distillation to capture inter-patch temporal structures. On heart rate estimation and atrial fibrillation detection, PPG-Distill improves student performance by up to 21.8% while achieving 7X faster inference and reducing memory usage by 19X, enabling efficient PPG analysis on wearables

Problem

• Deploying large PPG foundation models on resource-limited wearable devices
• Improving efficiency of photoplethysmography signal analysis for health monitoring
• Reducing computational requirements while maintaining accurate heart rate and AFib detection

Innovation

• Knowledge distillation framework transfers global and local knowledge
• Morphology distillation preserves local waveform patterns
• Rhythm distillation captures inter-patch temporal structures