Scalable Computation of Extremum Graphs

Abhijath Ande

Abstract

Extremum graphs have become increasingly important in the field of topology, as they allow for dimensionality reduction and exploratory analysis of high-dimensional scalar fields, while also preserving local geometric structure. As the interesting features of a scalar field are mostly associated with its extrema, this makes extremum graphs an appealing choice for high-dimensional scalar field analysis. But various works define their own proper but different notions of extremum graphs and how to compute them. This causes a lot of research time to be spent on designing and implementing common boilerplate code to compute extremum graphs for their studies. It also causes difficulties in transferring and comparing results across different studies. To solve these issues, we have created scalable library to compute extremum graphs utilizing GPU parallelism.

[PDF]