Cur Research

Current Focus: Integrating Schrödinger Bridge Models (SBMs) for Multimodal Time Series

Building on AQDiff’s success in single-modality forecasting, our ongoing research extends diffusion models to Schrödinger Bridge Models (SBMs) to address multimodal time series forecasting. SBMs, rooted in optimal transport theory, provide a principled framework to interpolate between complex distributions while maintaining fine-grained control over trajectories. This is particularly valuable in epidemiology, where forecasting requires synthesizing diverse data streams—case counts, genomic surveillance (e.g., viral variant prevalence), mobility trends, and healthcare resource utilization—into a unified probabilistic model.