In the same way that the Internet has combined with web content and search engines to revolutionize every aspect of our lives, the scientific process is poised to undergo a radical transformation based on the ability to access, analyze, and merge complex data sets.    Scientists will be able to combine their own data with that of other scientists, validating models, interpreting experiments, re-using and re-analyzing data, and making use of sophisticated mathematical analyses and simulations to drive the discovery of relationships across data sets.   This “scientific web” will yield higher quality science, more insights per experiment, a higher impact from major investments in scientific instruments, and an increased democratization of science—allowing people from a wide variety of backgrounds to participate in the science process.  

 

What does this “big science data” view of the world have to do with computer science? Due to the exponential growth rates in detectors, sequencers and other observational technology, data sets across many science disciplines are outstripping the storage, computing, and algorithmic techniques available to individual scientists.  Scientists have always demanded some of the fastest computers for computer simulations, and while this has not abated, there is a new driver for computer performance with the need to analyze large experimental and observational data sets. 

 

In this talk I will describe some examples of how science disciplines such as biology, material science and cosmology are changing in the face of their own data explosions, and how this will lead to a set of open questions for computer scientists due to the scale of the data sets, the data rates, inherent noise and complexity, and the need to “fuse” disparate data sets. 

BIO: Katherine Yelick is a Professor of Electrical Engineering and Computer Sciences at the University of California at Berkeley and the Associate Laboratory Director for Computing Sciences at Lawrence Berkeley National Laboratory.  She is known for her research in parallel languages, compilers, algorithms, libraries, architecture, and runtime systems. She earned her Ph.D. in Electrical Engineering and Computer Science from the Massachusetts Institute of Technology and has been on the faculty at UC Berkeley since 1991 with a joint research appointment at Berkeley Lab since 1996. She was the director of the National Energy Research Scientific Computing Center (NERSC) from 2008 to 2012 and in her current role as Associate Laboratory Director she manages a 300-person organization that includes NERSC, the Energy Science Network (ESNet), and the Computational Research Division. She is an ACM Fellow and recipient of the ACM/IEEE Ken Kennedy Award and the ACM-W Athena award.  She is a member of the National Academies Computer Science and Telecommunications Board (CSTB) and the Computing Community Consortium (CCC), and she previously served on the California Council on Science and Technology. 

 

Finding low dimensional solutions in high dimensional spaces is one of the fundamental problems in data science.  Convex penalty functions, most notably the L1 and nuclear norms, have played a major role in achieving sparse and low rank properties in the past decade. This talk gives an overview of recent development of non-convex penalties and algorithms that improve on L1 methods in several applications such as compressed sensing,  matrix completion, imaging science and machine learning. 

 

BIO: Jack Xin has been Professor of Mathematics at UC Irvine since 2005. He received his Ph.D in applied mathematics at Courant Institute, New York University in 1990. He was a postdoctoral fellow at Berkeley and Princeton in 1991 and 1992. He was assistant and associate professor of mathematics at the University of Arizona from 1991 to 1999. He was professor of mathematics from 1999 to 2005 at the University of Texas at Austin. His research interests include applied analysis, computational methods and their applications in multi-scale problems, sparse optimization, and data science. He authored over hundred journal papers and two Springer books. He is a fellow of the Guggenheim Foundation, and the American Mathematical Society. He is Editor-in-Chief of Society of Industrial and Applied Mathematics (SIAM) Interdisciplinary Journal Multi-scale Modeling & Simulation (MMS).

 

One of the primary objectives of the President’s National Strategic Computing Initiative is “Increasing coherence between the technology base used for modeling and simulation and that used for data analytic computing.“  I will interpret this objective in the context of Large-Scale Data Analytics (LSDA) problems, i.e. problems that require finding meaningful patterns in data sets that are so large as to require leading-edge processing and storage capability.  LSDA problems are increasingly important for government mission work, industrial application, and scientific discovery. Effective solution of some important LSDA problems requires a computational workload that is substantially different from that associated with traditional High Performance Computing (HPC) simulations intended to help understand physical phenomena or to conduct engineering. While traditional HPC application codes exploit structural regularity and data locality to improve performance, many analytics problems lead more naturally to very fine-grained communication between unpredictable sets of processors, resulting in less regular communication patterns that do not map efficiently on to typical HPC systems. In both simulation and analytics domains, however, data movement increasingly dominates the performance, energy usage, and price of computing systems. It is therefore plausible that we could find a more synergistic technology path forward. Even though future machines may continue to be configured differently for the two domains, a more common technological roadmap between them in the form of a degree of convergence in the underlying componentry and design principles to address these common technical challenges could have substantial technical and economic benefits.

Like most other domains of human endeavor, science is being transformed by the progress in computing and information technology, at a pace that is historically unprecedented.  Data volumes and data rates are growing exponentially, following Moore's law.  Even more interesting is the growth of data complexity and the overall information content of the data.  This opens great new opportunities for discovery, but with great challenges that are shared by all data-rich fields, primarily in the arena of knowledge discovery, including machine learning, computational statistics, novel approaches to multi-dimensional data visualization, etc.  As the complexity of data increases, we see an increased reliance on machine intelligence, leading towards a human-computer collaborative discovery.  Common challenges call for common solutions, in the form of a methodology transfer from one field to another.  Thus, the scientific method evolves as well, and "data science" becomes a new universal language of science, akin to the roles historically played by mathematics and statistics.

 

BIO: S. George Djorgovski is a Professor of Astronomy and the founding Director of the Center for Data-Driven Discovery. He has worked in many areas of astronomy and cosmology, and has led several sky surveys, currently the Catalina Real-Time Transient Survey that explores transient and variable objects and phenomena in the universe. He is one of the founders of the Virtual Observatory framework, and of the emerging field of Astroinformatics. His scientific interests are centered on the questions of how is computing and information technology changing the ways we do science, scholarship, and education in the 21st century.

 

Computational Science and Engineering is becoming ubiquitous in science and engineering. Indeed, if scientists and engineers are not using computing, especially high performance computing, they will be at a competitive disadvantage to those who are doing so. High performance computing is becoming necessary to analyze the data from large-scale experiments (e.g. the Large Hadron Collider), study fluid flow, unravel complex chemical and biological processes, understand climate behavior, and analyze and predict many more natural phenomena. Computing is also becoming important for designing and producing complex products from microchips to airplanes and space ships. Computing power has reached the point (~1-100 PetaFlops) that we can make accurate predictions of the behavior of complex systems. We can include all the important effects, use accurate algorithms with adequate resolution, model a complete system (an airplane, not just a wing section), carry out adequate verification and validation, and achieve sufficiently short turn-around times that enough parameter studies can be done so that the results can be both accurate and useful.

 

There are, however, significant and very practical challenges to successfully use this new capability. It’s sort of a Malthusian problem (population growing exponentially while food production grows linearly). While computer power has been growing exponentially (partially because chip design and manufacture has been automated with computers), software is still being developed by people who aren’t getting smarter and more productive nearly as fast. The result is that the software needed to exploit the new computers is lagging. The user paradigm is also changing. More and more scientists and engineers no longer write much code. They are beginning to use codes that someone else wrote. The computer/software have become a “virtual experiment.” The users personally become more productive, but they lose some of the ability to understand the limitations and range of validity of the software. Since the software is only a model, extreme care is needed to ensure that the model accurately predicts the real world.

 

I will discuss these issues by first describing the promise of computing, then the potential pitfalls, and finally what we can (and need) to do to fulfill the enormous promise that computing offers us. I will draw on the “lessons learned” over my career that started in 1967 through my current project. In 2005, I initiated and still lead the DoD Computational Research Engineering Acquisition Tools and Environments (CREATE) Program. The goal is to reduce the costs, time, and risks of successfully developing and fielding complex weapon systems by developing and deploying physics-based HPC software application to predict the performance of military air and ground vehicles, naval vessels, and RF antennas. To succeed we have needed to pay attention to the technical issues of getting the right equations, and solving them correctly and efficiently. But that’s just the beginning. We need to change the paradigm of the way science and engineering are being done. We need to find ways to support “virtual test facilities”, to retain the intellectual property of our codes, to ensure that the codes are accurate, to develop codes that will have a life cycle of many decades, to produce codes that are not only accurate, but usable, maintainable, extensible, portable, and well documented. Good software engineering and agile project management are among the most essential elements.

 

BIO: Dr. Douglass Post established and leads the DoD Computational Research and Engineering Acquisition Tools and Environments (CREATE) Program. He is an IPA from the Carnegie Mellon University Software Engineering Institute. Doug received a Ph.D. in Physics from Stanford University in 1975. He is Associate Editor-in-Chief of the AIP/IEEE publication “Computing in Science and Engineering”. He led the A Division Nuclear Weapons Simulation programs at the Lawrence Livermore National Laboratory and the Los Alamos National Laboratory nuclear weapons code development programs (1998-2005). Doug also led the International Thermonuclear Experimental Reactor (ITER, https://www.iter.org) Physics Team(1988-1990) for which he received the American Nuclear Society (ANS) Outstanding Technical Achievement Award for fusion science and engineering in 1992, and led the ITER In-Vessel Physics Team (1993-1998). He established and led the tokamak modeling group at the Princeton University Plasma Physics Laboratory (1975-1993). He is a Fellow of the American Nuclear Society, the American Physical Society, and the Institute of Electronic and Electrical Engineers (IEEE).  He received the American Society of Naval Engineers 2011 Gold Medal Award for the CREATE Program. He has written over 250 publications with over 7,000 citations.

 

The Hermitian eigenvalue problem is central to many scientific and engineering applications which model increasingly complex phenomena and thus give rise to matrices of huge size.  This poses significant challenges even to high-performance parallel iterative methods.  A related problem is the computation of a few singular values of large matrices, which faces similar challenges because of the explosion of big data applications.  In this talk we will outline the major categories of challenges for iterative methods and describe the state-of-the-art techniques used to address them.  We will also discuss the state-of-the-practice in software.

 

BIO: Andreas Stathopoulos is a Professor of Computer Science at the College of William and Mary in Virginia. He was awarded an NSF CISE Postdoctoral Fellowship after receiving his Ph.D. and M.S. in Computer Science from Vanderbilt University; he also completed a B.S. in Mathematics from the University of Athens in Greece. Dr. Stathopoulos’ research interests include numerical analysis and high performance computing; methods for large eigenvalue problems and linear systems of equations; and related applications from materials science and quantum chromodynamics. He co-developed PRIMME (Preconditioned Iterative MultiMethod Eigensolver), one of the foremost eigenvalue packages, and several other significant software tools and has published numerous journal articles and conference papers in computational sciences and applications.