Landmark Alternating Diffusion: Efficient Sensor Fusion Through Landmark Diffusion

Abstract

Alternating Diffusion (AD) is a powerful sensor fusion algorithm but suffers from computational limitations. We introduce Landmark Alternating Diffusion (LAD), an efficient alternative inspired by landmark diffusion. LAD uses a landmark set to streamline the fusion process, achieving superior computational efficiency compared to AD. We theoretically analyze LAD and demonstrate its application to automatic sleep stage annotation using EEG data, showing comparable performance with significantly reduced computation time. This is joint work with Xing-Yuan Yeh, Hau-Tieng Wu, and Ronen Talmon.