Computational Imaging/AI/Data Science

Discovery is challenging in the context of extreme chemical heterogeneity when pursued by conventional spectroscopy and imaging methods because complex dynamics within disordered/hybrid materials are collapsed into low-dimensional projections of the truth. Diverse local features that underlie function are obscured because structures of interest are often small, rare, diverse, and transient. Spectroscopic features of a key property, if they are even known, are often too weak to detect relative to background from the surrounding material – one must get “lucky.”
A universal challenge in characterizing disordered materials is discerning properties hidden within our data, e.g., dynamics occurring between measurements or high-order correlations between multiple parameters. To effectively harness our high-dimensional data discussed in the previous sub-aim, we will explore the design of neural network-based representations that can infer structural properties from sparsely or irregularly sampled spectroscopic data