Multi-modal Data Analysis and Multilevel Optimization

The rich multimodal datasets obtained from different acquisition frameworks (e.g., scientific instruments and scientific computers) are vital for complete knowledge of the phenomena of interest. However, knowing what and how to fuse different modalities is a cornerstone to efficiently exploiting the additional diversity offered by multimodalities for scientific discovery. In this talk, I will talk about a couple of optimization-based approaches that attempt to truly enable multimodal-based inverse problem solving and to leverage multilevel and stochastic sampling methods that handle the resulting large-scale optimization problems by solving a hierarchy of scales from both the parameter space and the data space.