Prototype-based Self-Supervised Multimodal Learning for PPG and Accelerometry Signals
The paper introduces Prototype-based Self-Supervised Multimodal Learning for PPG and Accelerometry Signals, addressing the limitation that existing AI methods often lose unique sensor signatures by relying on simple alignment. The proposed ProtoMM uses a shared prototype dictionary to capture both collective patterns and individual characteristics, demonstrating robust performance across various tasks. This represents a fundamental shift toward more nuanced understanding of complex physiological data for personalized health monitoring.