Computational Inverse Problems Can Drive a Big Data Revolution

Abstract

The attendees at Simula’s Challenges in Computing conference were privileged to receive a double dose of geophysical science. First, Carsten Burstedde was named a co-winner of the Springer Computational Science and Engineering (CSE) Award for his work on mantle convection simulation. In addition, his mentor, Omar Ghattas of the University of Texas, was one of the eight invited speakers at the meeting. While Burstedde lectured on adaptive mesh refinement in simulations of the Earth’s mantle flow, Ghattas cast his net more broadly and outlined five ‘grand challenges’ in scientific computing. Geoscience is currently undergoing a ‘perfect storm’, as Ghattas described it: a convergence of immense amounts of sensor data, new supercomputers to analyse it, and improved mathematical models to plug the data into. All of these converging streams have to funnel through a bottleneck known as inverse problems. Without fundamental improvements in this essentially computational problem, Ghattas argued, we will lose much of the opportunity for extracting geophysical knowledge from the data deluge.