Scalable Feature Tracking with FTK and Informed Model Fitting for Scientific Data Using the Discrete Legendre Transform

Jiayi Xu
David Lenz

Scalable Feature Tracking with FTK

Feature tracking makes it possible to derive insights during scientific experiments and simulations and to reduce the amount of data to be stored. For large data, however, the scalability of current tracking algorithms is a bottleneck. In this talk, we present distributed parallelism in the Feature Tracking Kit (FTK), highlighting a scalable and asynchronous union-find algorithm. In the FTK pipeline, we first construct and partition a high-dimensional mesh that incorporates both space and time. Then, the trajectories of features are distributed across parallel processes, and trajectory pieces are merged asynchronously using our distributed union-find implementation. Preliminary results demonstrate the scalability of tracking critical points (e.g., locations of local maxima) up to 8,192 processes on Bebop with both experimental data and fusion plasma simulations.  

Jiayi Xu is a Ph.D. student supervised by Prof. Han-Wei Shen at the Ohio State University. His research interests are in data visualization and graph analysis. He spent 14 weeks at Argonne as a summer student. 

Informed Model Fitting for Scientific Data Using the Discrete Legendre Transform

Multivariate Functional Approximation (MFA) is a technique for fitting scientific data sets with a continuous model that allows for better downstream data analysis. An MFA model is specified by several parameters, which in general must be selected and tuned manually to achieve the best performance. In this talk, I will discuss methods for choosing some of these parameters in an informed way so as to minimize the amount of necessary manual parameter tuning. In particular, I will explore the connection between B-Spline fitting and the discrete Legendre transform when the data being modeled are generated by a spectral element solver.

David Lenz is a fifth-year Ph.D. candidate in applied mathematics at the University of California, San Diego, where he is advised by Randy Bank. His doctoral research is in finite element methods on unstructured simplicial meshes, with an emphasis on space-time finite element methods and unstructured 4-dimensional meshes. He is visiting Argonne this summer as a Givens Associate.

Miscellaneous Information: 

This Seminar will be streamed. See details at

Please click here [schedule.ics] to add this event to your calendar.