In physical and earth sciences, several processes can be characterized by multi-scale and multi-level physics. They are formulated as strongly-coupled differential equations and simulating them as such, can be computationally challenging. Even in absence of physical equations, building multiple models to separately model the different levels of information leads to increased resource cost. Using a multi-resolution Gaussian processes model, we can accurately forecast the multi-scale physics in a dynamical, chaotic system. We will look at an atmospheric modeling case study to demonstrate this.
Somya is a PhD student in Computer Science working with Vipin Kumar and Snigdhansu (Ansu) Chatterjee. Her research work focuses on uncertainty quantification in Bayesian deep learning and machine learning applications. Her past research work focuses on spatio-temporal modeling, uncertainty quantification, non-parametric inference, approximate Bayesian computation, changepoint detection, climate science and healthcare applications of machine learning.