VA Tech Graduate Student uses Inverse Problems Workshop to Influence Personal Research

Contributed by: Joseph Tanner Slagel, Graduate Student, Dept. of Mathematics, Virginia Polytechnic Institute & State University

As a graduate student at Virginia Tech, my research interests are in large-scale numerical linear algebra. In particular, I have recently been studying stochastic approximation methods for solving very large least square problems.  These are least square problems where the data size is terabytes (or even petabytes!) in size, and thus cannot fit in a computer CPU’s memory all at once.

I looked at the Statistical Inverse Problems Workshop at the Statistical and Applied Mathematical Sciences Institute (SAMSI) as an excellent opportunity to learn about emerging topics and techniques that I could apply to my own personal research.

I attended the Statistical Inverse Problems Workshop from January 26-27, 2017. The friendly and open atmosphere of the SAMSI workshop made it easy for me to make new connections and to discuss topics related to my research. The experience gave me the opportunity to work with a dynamic range of professionals (graduate students, postdocs and junior/senior level faculty), which helped me gain perspective from mathematicians at various stages of their career.

“Attending this SAMSI workshop was a great way for me to connect with other researchers whose interests overlap with my own.”

A distinguishing feature of the workshop was that, in addition to plenary and research talks, there was a lot of time for research discussions and collaborations on current projects – small groups were encouraged to find an open room and work for multiple hours on emerging research questions.

The talks at the workshop gave me an opportunity to learn about important problems in Bayesian inverse problems. Despite the small group, I heard a range of talks that provided an overview of open problems in the field, expounded on main computational and algorithm challenges, and described lots of cool real-life applications. The most beneficial part of the workshop for me was getting to speak to others about my research. I received a lot of helpful input and pointers to resources that have helped me see where my work fits into the larger statistical inversion community.

Ahmed Attia (instructing), a Postdoctoral Fellow at SAMSI, gives a research lecture at the Statistical Inverse Problems Workshop

Attending this SAMSI workshop was a great way for me to connect with other researchers whose interests overlap with my own. I look forward to returning to SAMSI in the future for more collaborations and discussions!

NOTE: To see what subjects were presented at this workshop visit the SAMSI website at: www.samsi.info/opt-inv-prob.

UT Mathematician Discusses Advances & Future of Super-computing

Jack Dongarra, Director of the Center for Information Technology Research and Innovative Computing Laboratory – University of Tennessee (photo courtesy of University of Tennessee)

The Workshop on the Interface of Statistics and Optimization (WISO) at Duke University’s Penn Pavilion wrapped up recently. After we said good bye to all of our participants, I was reminded of an interesting talk given by one of the twelve insightful speakers. The speaker in question was Jack Dongarra.

Jack is the Director of the Center for Information Technology Research and Innovative Computing Laboratory, from the University of Tennessee. He lectured on: “The Road to Exascale and Legacy Software for Dense Linear Algebra (or what we have been doing for the last 43 years).” Jack’s talk at WISO could be appreciated even by those who are not specialists in the esoteric matters of statistics and optimization. True to the title, Jack has been designing high-performance software for over 40 years. His pioneering work has been honored with many awards, including membership in the US National Academy of Engineering.

Twice a year, Jack co-publishes a benchmark of the 500 most powerful supercomputers in the world. The current number one, as of November 2016, is the Sunway TaihuLight machine in Wuxi, China. Built from over 10 million processors, it can execute more than 10¹⁶ operations in a single second. The Titan Cray XK7 at Oak Ridge National Lab, located in Oak Ridge, TN, is number three.

The Sunway TaihuLight machine (above) is the number one super computer in the world. Its more than 10 million processors make it possible to execute millions operations in a single second!

For the many young people in the audience, Jack started his talk with a history of how hardware and software have developed over the years, and how he himself got into the business of benchmarking. A few photos from the 1970’s, featuring Jack and his colleagues in bell bottoms and side burns, proudly posing on a Ford Pinto with the license plate “Linpack” (after their groundbreaking public-domain software package) were well received.

– Jack Dongarra and fellow software developers pose with his trusty Ford Pinto in the 1970’s identified with the license plate “LINPACK.” LINPACK was the group’s groundbreaking public-domain software package.

Jack’s advice for how to build good software: Keep the processors busy with arithmetic, make sure they coordinate their work schedules efficiently rather than sitting idle waiting for results from other processors.

Jack ended with a list of the many challenges for software design, including:

• Efficiency (speed matters),
• Scalability (keep up the efficiency, even in the face of growing work),
• Reliability (with 10 million processors in use, some are bound to fail)
• Portability (the particular hardware platform should not matter).

Jack Dongarra was one of many interesting speakers at the WISO. If you are interested in seeing the other presentations from the 3-day event, visit the WISO Video Page.