Profile-aided distillation framework for personalized sleep analysis with compact models using LLM-guided synthetic...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence
• Wearable Technology

Background:

• Personalized sleep analysis on edge devices is vital for real-time health insights but hindered by limited data and hardware constraints.
• Existing methods struggle with data scarcity and the computational demands of edge deployment.

Purpose of the Study:

• To create a framework for personalized sleep analysis on edge devices.
• To address challenges of limited physiological data and restricted model capacity.

Main Methods:

• Developed a Physiologically-Constrained Adaptive Hierarchical Copula approach for synthesizing realistic physiological signals.
• Introduced Profile-Aided Distillation of Expert Inference with MoE LoRA for enhanced personalized inference on resource-limited models.

Main Results:

• Distilled models achieved performance comparable to state-of-the-art large language models.
• Models operated efficiently within edge device computational and memory constraints.
• Demonstrated effectiveness on both public and in-house datasets.

Conclusions:

• The proposed framework provides a practical solution for on-device personalized sleep analysis.
• It bridges the gap between high-performance modeling and real-time healthcare applications in resource-constrained environments.