TransConv-DDPM: Time-Series Data Generation With Diffusion
Published in CS 7300 Course Poster Presentation, 2024
I presented a poster on “TransConv-DDPM: Time-Series Data Generation With Diffusion” as part of the CS 7300 Graduate Research course at Texas State University on December 7, 2024.
The presentation summarized my research published in IEEE PULSE (Vol. 16, No. 1, 2025), co-authored with Anne H. H. Ngu, focusing on synthetic time-series generation for healthcare applications.
The poster highlighted the design of TransConv-DDPM, a diffusion-based generative model that integrates multiscale convolution modules and transformer layers to effectively capture both local and global temporal dynamics in physiological time-series data.
Key results demonstrated that TransConv-DDPM outperformed baseline methods such as TimeGAN and standard DDPM on benchmark datasets including Stick Balancing and SmartFallMM, showing superior performance in metrics like Correlation, DTW, and FID.
Citation:
Kabir, Md Shahriar, and Anne H. Ngu.
“TransConv-DDPM: Time-Series Data Generation With Diffusion.”
IEEE PULSE, 16(1): 29–31, 2025.