ACE: Automated Optimization Towards Iterative Classification in Edge Health Monitors.

Area of Science:

• Biomedical Engineering
• Computer Science
• Edge Computing

Background:

• Wearable devices require efficient real-time biomedical signal processing for health monitoring.
• Limited computational resources on edge devices hinder online processing of complex health algorithms.
• Existing monolithic models lack adaptability and efficiency for diverse edge applications.

Purpose of the Study:

• To propose ACE (Automated optimization towards classification on the Edge), an application-agnostic methodology for optimizing health monitoring algorithms on edge devices.
• To enable real-time processing of biomedical signals with reduced computational load.
• To enhance the deployment of health monitoring applications on resource-constrained wearable technology.

Main Methods:

• ACE decomposes monolithic algorithms into multiple algorithms with varying computational complexities.
• It integrates buffering logic to minimize re-computation of shared, data-intensive features.
• The optimized algorithms are converted to C for edge deployment and executed iteratively based on confidence thresholds.

Main Results:

• ACE achieved significant runtime savings of at least 28.9% for seizure detection and 18.9% for emotional state classification.
• No accuracy loss was observed on a Cortex-A9 edge platform.
• The methodology demonstrated effectiveness across diverse biomedical applications.

Conclusions:

• ACE provides an effective solution for optimizing biomedical signal processing on edge devices.
• The iterative, complexity-adaptive approach enhances efficiency without compromising accuracy.
• ACE empowers designers to deploy sophisticated health monitoring applications on resource-limited wearable systems.