A-Z Index | Phone Book | Careers

InTheLoop 06.13.2016

June 13, 2016

Noon Vigil for Orlando Shooting Victims Today

The Lambda Alliance, a Lesbian, Gay, Bisexual, Transgender and Questioning/Queer (LGBTQ) organization at the lab, and the lab's Diversity Council are joining the rest of the UC community in mourning the victims of this weekend's shooting in Orlando. The "Cal Stands with Orlando" takes place from 12:10-1 p.m. today at Sproul Plaza, Savio Steps on the campus of UC Berkeley.

New Mathematics Accurately Captures Liquids and Surfaces Moving in Synergy

Gas bubbles in a glass of champagne, thin films rupturing into tiny liquid droplets, blood flowing through a pumping heart and crashing ocean waves—although seemingly unrelated, these phenomena have something in common: they can all be mathematically modeled as interface dynamics coupled to the Navier-Stokes equations, a set of equations that predict how fluids flow.

Today, these equations are used everywhere from special effects in movies to industrial research and the frontiers of engineering. However, many computational methods for solving these complex equations cannot accurately resolve the often-intricate fluid dynamics taking place next to moving boundaries and surfaces, or how these tiny structures influence the motion of the surfaces and the surrounding environment.

This is where a new mathematical framework developed by Robert Saye, Lawrence Berkeley National Laboratory's (Berkeley Lab's) 2014 Luis Alvarez Fellow in Computing Sciences, comes in. By reformulating the incompressible Navier-Stokes equations to make them more amenable to numerical computation, the new algorithms are able to capture the small-scale features near evolving interfaces with unprecedented detail, as well as the impact that these tiny structures have on dynamics far away from the interface. A paper describing his work was published in the June 10 issue of Science Advances.

Models Pinpoint Better Material for Nuclear
Fuel Recycling

Researchers are investigating a new material that might help in nuclear fuel recycling and waste reduction by capturing certain gases released during reprocessing. Conventional technologies to remove these radioactive gases operate at extremely low, energy-intensive temperatures. By working at ambient temperature, the new material has the potential to save energy, make reprocessing cleaner and less expensive. The reclaimed materials can also be reused commercially.

Appearing in Nature Communications, the work is a collaboration between experimentalists and computer modelers exploring the characteristics of materials known as metal-organic frameworks. These materials have tiny pores inside, so small that often only a single molecule can fit inside each pore. When one gas species has a higher affinity for the pore walls than other gas species, metal-organic frameworks can be used to separate gaseous mixtures by selectively adsorbing.

To find the best MOF for xenon and krypton separation, computational chemists led by Berkeley Lab's Maciej Haranczyk and Berend Smit screened 125,000 possible MOFs for their ability to trap the gases. Although these gases can come in radioactive varieties, they are part of a group of chemically inert elements called "noble gases." The team used computing resources at NERSC, the National Energy Research Scientific Computing Center, a DOE Office of Science User Facility at Berkeley Lab.

"Identifying the optimal material for a given process, out of thousands of possible structures, is a challenge due to the sheer number of materials. Given that the characterization of each material can take up to a few hours of simulations, the entire screening process may fill a supercomputer for weeks," said Haranczyk. "Instead, we developed an approach to assess the performance of materials based on their easily computable characteristics. In this case, seven different characteristics were necessary for predicting how the materials behaved, and our team’s grad student Cory Simon’s application of machine learning techniques greatly sped up the material discovery process by eliminating those that didn’t meet the criteria."

Materials Project Releases Data Trove, New Tools

The Materials Project, a Google-like database of material properties powered by NERSC, has released an enormous data trove along with new visualization and analysis tools to the public. The release gives energy materials researchers in general, and battery researchers in particular, access to new avenues of exploration.

Co-founded and directed by Lawrence Berkeley National Laboratory (Berkeley Lab) scientist Kristin Persson, the Materials Project uses supercomputers to calculate the properties of materials based on first-principles quantum-mechanical frameworks. It was launched in 2011 by the U.S. Department of Energy’s (DOE) Office of Science. The project has been used by scientists researching everything from

Last month's release included nearly 1,500 compounds investigated for a promising new battery technology along with more than 21,000 organic molecules relevant for liquid electrolytes as well as a host of other research applications. Other researchers make similar data openly available, but they don't give scientists the tools to explore, visualize and analyze those data sets through a web interface, as does the Materials Project. Its recent release expands that toolset with two new applications—the Molecules Explorer and the Redox Flow Battery Dashboard—plus an add-on to the Battery Explorer web app enabling researchers to work with other ions in addition to lithium. “Not only do we give the data freely, we also give algorithms and software to interpret or search over the data,” said Persson, who is also a UC Berkeley professor.

Profiles in LGBTQ Pride

To mark LGBTQ Pride Month, the lab recently wrote profiles of three LGBTQ staff. We don't like to brag (okay, maybe just a little), but two of the three profiles were our own Deb Agarwal and Elijah Goodfriend, both of the Computational Research Division.

LGBTQ Pride Month is celebrated each year in the month of June to honor the 1969 Stonewall riots in Manhattan. The Stonewall riots were a tipping point for the Gay Liberation Movement in the United States. Since then, the lesbian and gay communities have united with transgender, intergender, queer and questioning people and their allies to defend and celebrate the diversity of human sexuality and gender identity. This year at the lab, the Lambda Alliance, Lamda Employee Resource Group and the Diversity and Inclusion office are hosting a variety of events throughout the month to celebrate. Some things to expect: two free screenings of the Oscar-winning film The Danish Girl (11:30 a.m., Thursday, June 14 in bldg. 50 auditorium & 11:30 a.m., in bldg. 66 auditorium), a resource fair, a Pride-themed Giants baseball game and networking event. Please check the Lambda Alliance site for more events and details.

Canning Gives Invited Talk at PRE’16 Workshop 

Andrew Canning of CRD’s Computational Chemistry, Materials & Climate Group was an invited speaker at last week’s International Workshop on Photoluminescence in Rare Earths: Photonic Materials and Devices (PRE’16) held at Clemson University.

Canning’s June 8 talk on “First Principles Theoretical Studies of Luminescent Properties of Rare Earth Based Scintillator Materials” described large-scale computations for optical properties of detector materials.

Bethel Presents Paper on In Situ Methods and Infrastructure at Eurovis 2016 

Wes Bethel, who leads the Computational Research Division’s (CRD’s) Data Analytics and Visualization Group, presented a "State-of-the-Art Report" paper "In Situ Methods, Infrastructures and Applications on High Performance Computing Platforms" at the Eurovis 2016 conference last week. The conference was held June 6-10 at the University of Groningen in the Netherlands.
 
The paper was essentially a discussion among researchers, developers and practitioners using in situ methods and infrastructures in extreme-scale high performance computing (HPC). It included a review of existing methods in the field, system infrastructures, as well as a range of computational science and engineering applications currently using in situ analysis and visualization. The paper will be published in an upcoming issue of the journal Computer Graphics Forum.
 
This work is part of the “Scalable Analysis Methods and In Situ Infrastructure for Extreme Scale Knowledge Discovery” project, which is led by Bethel and supported by the Department of Energy’s (DOE’s) Office of Advanced Scientific Computing Research. This project focuses on new in situ algorithms and infrastructures for a wide-range of science areas, which will allow researchers do data visualization and analysis without first writing that data to disk. The combination of algorithmic and infrastructure work is grounded in direct interactions with specific application code teams, all of which are engaged in their own R&D aimed at evolving to the exascale systems.
 
In addition to Bethel, other co-authors on the paper include: Andrew Bauer, Berk Geveci and Patrick O’Leary (Kitware); Hasan Abbasi and Scott Klasky (ORNL); Hank Childs (University of Oregon); James Ahrens (LANL); Kenneth Moreland (Sandia National Lab); Venkatram Vishwanath (ANL); and Brad Whitlock (Intelligent Light).

Apply Now: SC16 Early Career Program

SC16, the supercomputing conference, is accepting applications through July 31 for the SC16 Early Career Program for Professionals.

The Early Career Program is designed to equip participants for significant contributions in their new careers by building skills related to obtaining funding, identifying publishing venues, establishing long-term mentor relationships, and effectively managing their time. The program is intended for people in their first five years of a permanent position (assistant professors, and members of the technical staff.

Sessions will be held in a workshop format on Monday, Nov.  14 before SC16 starts to allow participants to enjoy the full technical program. The program includes formal presentations, time for peer mentoring, a speed mentoring session to meet 4 to 5 potential long-term senior mentors and an organized lunch for informal networking.

SC16 Women in Networking Program Extends Application Deadline to June 22

The Women in IT Networking at SC program (WINS) is extending the deadline for applications to Wednesday, June 22. The WINS program funds U.S. women in their early to mid-careers to help build SCinet, a terabit-scale research network for the SC conference. The program focuses on mentoring and hands-on training in networking, infrastructure, systems and security in the weeks prior to and the duration of the SC conference.  SC16 will be held Nov. 13-18 in Salt Lake City, Utah.

The program funds selected participants to travel for the staging (if applicable), setup and attendance of the SC conference and SCinet. Visit the WINS site for more information.

This Week's CS Seminars

»CS Seminars Calendar

Wednesday, June 15

Opportunities and Challenges in Accelerated Image Processing
11:30 a.m. – 12:30 p.m., Wang Hall - Bldg. 59 Room 4102
Youssef Nashed, Argonne National Laboratory

Recent advances in imaging hardware, specifically higher spatial resolution and data acquisition rates, require faster and more robust image processing algorithms. There is a gap between generated data volumes and their corresponding management, analysis, and visualization methods in the scientific domain. In the last few years, industry has been successful narrowing this gap, boosted by rapid advancements in parallel processing hardware and machine learning techniques, namely Graphics Processing Units (GPUs) and deep learning. On the other hand, the research community has yet to catch up with industry and capitalize on this success.

In this talk, I will present several efforts aimed at extracting useful information from images and time varying datasets through developing GPU-based open source software libraries. I will first discuss the viability of bio-inspired metaheuristics parallel implementations for mathematical optimization. Those implementations were abstracted and grouped in an open source function optimization library, libCudaOptimize. This library was shown effective in various applications. Ranging from finding histological structures in medical images to understanding traffic signs from autonomous vehicle videos.

Additionally, I will describe a novel parallel model of the human neocortex, that is able to detect, classify, and predict patterns in time-series data, in an unsupervised way. Furthermore, ongoing work with the Advanced Photon Source (APS) will be overviewed.

This work includes Ptycholib, a parallel library for real-time image reconstruction of ptychograpic datasets on high performance computing systems. Ptycholib uses a hybrid parallel strategy to divide the computation between multiple GPUs, achieving a final reconstruction by merging sub-dataset results into a single complex phase and amplitude image. I will also demonstrate how rapid reconstructions enabled beamline scientists push the boundaries of their data acquisition schemes, empowering multimodal and fly-scan experiments.

Thursday, June 16

Gaining Insight into Image-based Data Collected from Experimental Science Projects
10–11 a.m.,  Bldg. 50B Room 4205
Talita Perciano, Berkeley Lab

Image-based data obtained from experiments is the basis of research across different science domains. However, there is a lack of image analysis software tools capable of extracting valuable and hidden information in digital images. These tools are essential to help scientists to explore and understand complex data, providing accurate and deep understanding through, for example measurements, for decision-making. The amount of data generated every day by advanced instruments is increasing rapidly, along with the broad variety of sensors. Additionally, specific characteristics of the image data such as the presence of heterogeneous structures in multiple scales and their complex organization architecture pose algorithmic challenges when applied to large volumes of data. My research contributes to this endeavor by focusing on designing software applications that deal with complex large datasets using pattern recognition and machine learning techniques allied with advanced computer architectures. I will give a general overview about the different projects I am involved with, and present details about a recent published work using a Markov Random Field framework for image segmentation. Perciano is a member of CRD's Data Analytics and Visualization Group.