Landmark Alternating Diffusion

Material

• Paper in revision
• Slides
• Poster in MATLAB & Simulink Conference

Introduction

Alternating Diffusion (AD) [2] is a commonly applied diffusion-based sensor fusion algorithm. We propose Landmark AD (LAD), a variation of Alternating Diffusion, inspired by ROSELAND [1], to enhance computational efficiency while maintaining the benefits of AD. Theoretical analyses of LAD are provided, and its effectiveness is demonstrated in automatic sleep stage annotation using two EEG channels.

References

1. C. Shen and H.-T. Wu, (2022), Scalability and robustness of spectral embedding: landmark diffusion is all you need, IMAIAI.
2. R. Talmon and H.-T. Wu, (2019), Latent common manifold learning with alternating diffusion: Analysis and applications, ACHA.