Department Status for CSE at Georgia Tech
The Georgia Institute of Technology inaugurated its School of Computational Science and Engineering at a convocation in February. The result of an initiative that began in 2005, the new school is located within the College of Computing, where it joins the School of Computer Science and the School of Interactive Computing. Georgia Tech is one of the rare institutions to have a separate department of CSE with its own faculty.
By creating the School of Computational Science and Engineering, says founding chair Richard Fujimoto, Georgia Tech has taken the position that CSE is a discipline in its own right, one that deserves an academic home for its faculty and students. “CSE derives much of its richness from collaborations with disciplines such as mathematics, science, and engineering,” he continues. “This is a key principle underlying the formation of the School of CSE.”
Founding members, in addition to Fujimoto, are Haesun Park and David Bader, and seven new members have joined the school since 2005. Through several joint and adjunct appointments, the school has established links with other disciplines and, in some cases, other institutions, including Oak Ridge National Laboratory.
Defining CSE as “a discipline devoted to the systematic study, creation and application of computer-based models to understand, analyze, and/or design natural and engineered systems,” Georgia Tech elected to focus on the following subfields: high-performance computing, computational data analytics, modeling and simulation, numerical computing, and computational algorithms. These areas form the core of the CSE graduate program curriculum, which currently enrolls approximately 60 students; they are also central to an undergraduate minor program currently under development. Both the undergraduate and the graduate programs emphasize the development of expertise in a science or engineering domain in conjunction with knowledge of computation. A noteworthy aspect of Georgia Tech’s new school is its bringing together of modeling and simulation with data analytics, sometimes called the fourth paradigm of science, in one department.
The February convocation, held on the Georgia Tech campus in Atlanta, included a panel discussion of future directions for the discipline of CSE and a keynote address, “The Exascale: Why and How,” by David Keyes (King Abdullah University of Science and Technology). “In many fields the cost of experiments has risen past the ability of any one nation to perform, while the cost of simulation continues to plummet,” Keyes remarked. “These trends favor an increasing number of computational scientists to work among seven million publishing scientific researchers in the world today. Georgia Tech continues to take leadership steps in this campaign.”
Haesun Park, a member of the panel, discussed the need to establish a body of knowledge to help curriculum development efforts. “Deriving knowledge and insight from massive, complex data,” she said, “has come to the forefront as a key technological challenge that must be addressed in order to enable discovery and innovation in many science and engineering fields.” Discussing challenges in high-performance computing, Jeffrey Vetter, who holds a joint appointment at Georgia Tech’s School of CSE and Oak Ridge National Laboratory, identified two of the most important: designing new architectures that can improve energy efficiency by two orders of magnitude, and designing innovative programming systems that allow users to build applications that use billions of threads, and that are efficient and reliable. Jeffrey Skolnick, a professor of biology at Georgia Tech, considered multi-scale computational methods and critical challenges in the modeling of cells on a molecular level. Such simulations are of interest not only because they can provide fundamental understanding, he said; they could also be used to model the transformation of normal cells to cancerous ones. Georgia Tech chemistry professor David Sherrill highlighted such challenges as scalability and fault tolerance, noting that 50% of the quantum chemistry codes he executes fail on large supercomputers today because of node failures. He cited the need for extensive domain knowledge in order to determine acceptable approximations for exascale codes in quantum chemistry. Both Sherrill and Skolnick hold joint appointments with the new school.
Additional information can be found at http://www.cse.gatech.edu/about