Integrating ReIMTS with Diffusion Models

Background: The ReIMTS framework uses latent representations derived from IMTS backbone encoders, which can include temporal, variable, or observation representations. Diffusion-based IMTS models often predict noisy latent representations as part of their backbone process.

Question / Future Work: Develop strategies to adapt the ReIMTS framework to function with diffusion-based IMTS backbones, potentially by predicting clean observations during the denoising process or by integrating the recursive fusion mechanism directly within the diffusion model’s denoising backbone.