Abstract
Time series data could be incomplete due to a variety of reasons such as the errors in communications and sensor devices. This paper aims to restore the incomplete time-series medical data using the denoising diffusion probabilistic model (DDPM). A DDPM is applied to restore the missing values based on the model trained using the original data without loss. The proposed diffusion model-based signal restoration approach was successful in restoring the incomplete electrocardiogram signals up to 50%.
Jiwoon Lee
Ph.D. Student, University of Illinois Urbana-Champaign
Jiwoon Lee is a Ph.D. student in Nuclear Engineering at the University of Illinois Urbana-Champaign. His research interests include neuromorphic computing, scientific computing, and computational neuroscience.
Cheolsoo Park
Professor
His research interests include machine learning, adaptive signal processing, computational neuroscience, and wearable technology.