High Performance Computing:  Enabling Complex Decisions Making in the Information Era

Ananth Grama
Director, Computational Science and Engineering
Associate Director, Center for Science of Information
Professor of Computer Science
Purdue University.


High performance computing has emerged as a critical enabling technology for simulation,
design, and prototyping. More recently, large scale computing platforms have found extensive
use in data analytics and decision support. Purdue has significant strengths in all of
these areas. This talk summarizes research at Purdue on various aspects of high performance
computing. It also highlights closely related efforts on data analytics and a comprehensive
graduate program -- the Computational Science and Engineering specialization, which is one
of the top-ranked programs, worldwide, with over 150 affiliated faculty and 160 graduate
students. The talk will focus on Purdue's presence in the following areas:

High Performance Computing: Methods, Software, Environments, and Platforms. Research
at Purdue has resulted in highly scalable numerical techniques for linear and non-linear
system solvers, eigenvalue problems, multi-body simulations (from atomistic simulations
to PPM/ SPH/ MPM methods) and their applications. These methods provide the basis for
complete simulation environments integrating uncertainty quantification techniques in
support of predictive design. These methods have been validated in the context of applications
ranging from nanoscale devices to large wind turbines. A number of public domain software
libraries have been developed, along with web-enabled services that enable remote
access to these software frameworks on large parallel computers at ITaP.

Large-Scale Data Analytics: Algorithms, Statistical Techniques, Systems Software.
Purdue researchers have worked on topics including scalable distributed formulations of
a variety of Machine Learning and Data Mining kernels and associated statistical frameworks.
In addition, they have developed systems software, ranging from programming models and
compilers to runtime systems for distributed platforms. These systems have been validated
on a variety of applications ranging from supervised/ unsupervised learning to information
retrieval and recommender systems.

Related Centers and Efforts: A number of Centers at Purdue investigate closely related
topics. This includes the Center for Science of Information, a Science and Technology
Center of the National Science Foundation. This center explores quantitative approaches
to data storage, representation, communication, analysis, valuation, and use in various
contexts.

Education Programs: The Computational Science and Engineering program is a multidisciplinary
program aimed at training graduate students from different disciplines in the use of
computational methods. Core courses in the curricula include parallel algorithms, programming,
numerical methods, and optimization techniques. Optional course from various departments
specialize the skill-set to specific disciplines. In addition to training students, this
program provides critical expertise to a number of large projects on campus, including
the PRISM Center, Center for Science of Information, and the Network for Computational
Nanotechnology. A new Data Science graduate specialization will be initiated in 2014-15.