InTheLoop | 08.20.2012
August 20, 2012
Michael Wehner Discusses Mega-Droughts and Climate Change
The five-year drought in the American West that began in the last decade was the worst in 800 years. Could devastating mega-droughts—some lasting decades—be the new normal in big parts of the United States? Michael Wehner, a climate scientist in Berkeley Lab’s Computational Research Division, was one of three panelists who discussed this possibility last week on the NPR program On Point. Listen here.
And in a YouTube video, Wehner invites viewers to submit questions about extreme weather and climate change, which he will answer this week in a follow-up video.
David Leinweber Comments on Runaway Computer Trading
A Los Angeles Times article on runaway computer trading quotes David Leinweber, who heads the Center for Innovative Financial Technology in the Computational Research Division at Berkeley Lab, on the need for intensive simulations to test algorithmic trading programs—something that has not yet been done. Read more.
Berkeley Lab Is Accepting Applications for Alvarez Fellowship
Applications are now being accepted for the Luis W. Alvarez Fellowship in Computing Sciences, sponsored by Berkeley Lab’s Computing Sciences Directorate. Researchers in computer science, applied mathematics, or any computational science discipline who have received their Ph.D. within the last three years are encouraged to apply. The successful applicant will receive a competitive salary and excellent benefits.
This Week’s Computing Sciences Seminars
Python Boot Camp
Monday–Wednesday, August 20–22, 8:30 am–5:00 pm, 1 LeConte Hall, UC Berkeley
Josh Bloom, UC Berkeley
The purpose of the Boot Camp is to get those familiar with other computing languages (like C, Java, FORTRAN, and Lisp) ramped on the basics of the Python language. This is not a computer science class; understanding of basic CS concepts (like looping, recursion, etc.) are presupposed. The Boot Camp itself is a mixture of formal lectures, in-class demos, coding breakout sessions for participants, and homework projects. The Camp is open to anyone within the UC Berkeley community.
An Algebraic Multifrontal Preconditioner That Exploits a Low-Rank Property
Monday, August 20, 3:00–4:00 pm, 50F-1647
Artem Napov, LBNL/CRD
The solution of large sparse linear systems is a critical component for many of today's scientific and engineering codes. We consider preconditioned iterative solution methods and present an algebraic preconditioner based on an approximate factorization of the original system. The factorization is of a multifrontal type and amounts to a (partial) factorization of a sequence of dense matrices. For systems arising in the context of partial differential equations (PDEs), these matrices are known to exhibit a low-rank property. We explain how this property may be exploited in an algebraic (or black box) solution algorithm to reduce memory use and operations count. Numerical experiments are presented to demonstrate the potentialities of the approach.
AMP Camp Big Data Bootcamp
Tuesday and Wednesday, August 21–22, 9:30 am–5:30 pm, Sutardja Dai Hall, Banatao Auditorium, UC Berkeley
The first UC Berkeley AMP Camp will be hosted in Berkeley August 21–22, 2012. AMPLab researchers work at the intersection of machine learning, cloud computing, and crowdsourcing. We are integrating next-generation Algorithms, Machines, and People (AMP) to make sense of Big Data, and want to share our expertise with you.
AMP Camp attendees will learn to solve Big Data problems using components of the Berkeley Data Analytics System (BDAS) and cutting edge machine learning algorithms.
Specific sessions include:
- Hands-on lessons teaching developers to use Spark and Shark, which are capable of running iterative jobs up to 30x faster than Hadoop MapReduce
- State-of-the-art scalable machine learning algorithms
- Techniques for using the crowd to answer questions that can't be answered by computers alone
- Case-studies presented by active BDAS users
Joint Contour Nets: Theory and Applications
Wednesday, August 22, 10:00–11:00 am, 50F-1647
Hamish Carr, Visualization and Virtual Reality Group, School of Computing, University of Leeds
As scientific data sets increase in size and complexity, scientific visualization increasingly depends on formal analysis of the data. One of the most successful forms of analysis uses computational topology to analyse properties such as minima, maxima, thresholds, ridges and flow. To date, however, these methods have been applied to univariate (scalar) fields and to vector fields, but not to the more general case of multivariate fields.
In particular, Contour Trees and Reeb Graphs are often used for analysing univariate (scalar) fields. We generalize this analysis to multivariate fields with a data structure called the Joint Contour Net that quantizes the variation of multiple variables simultaneously. We report the first algorithm for constructing the Joint Contour Net, and demonstrate some of its fundamental properties. Based on this, we also show some preliminary results on its use for visualization by applying it to a problem from nuclear fission analysis, in which the topological insight provided aided scientists in understanding a physical phenomenon.
(Joint work with David Duke (University of Leeds), Aaron Knoll (Argonne National Laboratory), Nicolas Schunck (Lawrence Livermore National Laboratory), Hai-Ah Nam (Oak Ridge National Laboratory) and Andrzej Staszczak (University Marie Curie-Skłodowska, Lublin, Poland).
Asymptotic Performance Modeling of Application-Architecture Co-Design
Thursday, August 23, 11:00 am–12:00 pm, 50F-1647
Vladimir Getov, University of Westminster, U.K.
Building on previous results, this work proposes an abstract performance model for characterizing the couple application-architecture and a methodology for their co-design. Two groups of parameters are first introduced based on typical high-end computer architecture. A set of suitability functions are then defined using those parameters. These functions could be used for both optimizing the application-architecture co-design as well as for scalability and comparative performance analysis.
Functional Approaches to Visualization
Friday, August 24, 10:00–11:00 am, 50F-1647
David Duke, University of Leeds, United Kingdom
Functional approaches to computation have started to appear in mainstream visualization. Two recent example are Bostock's D3 toolkit which makes extensive use of functional abstractions, and the Diderot domain-specific language (DSL) for parallel graphics from Kindlmann's group at U.Chicago. And as a result of progress in efficient compilers and libraries, even pure functional languages such as Haskell have been investigated as platforms for more computationally intensive tasks in scientific visualization. This talk will describe the speaker's experience gained through over six years of working with functional programming in visualization, identifying both the opportunities and hard challenges in using functional programming and in particular Haskell. Successes include work on volumetric streaming where the Haskell code appears to perform at least as well as imperative code many times more complicated, and how by using embedded DSLs a small team was able to develop a credible solution to the 2008 Visualization Design contest in under 4500 lines of code.
One attraction of purely functional programming is the potential for simpler exploitation of parallelism. The talk will report on the difficulties encountered in trying to develop an efficient, parallel implementation of Carr's contour tree algorithm, and the lessons learnt. The work also illustrates hard questions about matching language and task: Are some technologies (like C/C++) simply a better match to the computational demands of scientific visualization, or will ongoing work on DSLs for GPU programming and multi-core parallelism yet become mainstream? The talk will conclude by describing recent work that looks at the relationship from the opposite angle—how visualization can be employed to understand the performance of Haskell programs, whose run-time behaviour makes conventional approaches to performance analysis impractical.
Link of the Week: Mom Puts Kids to Bed with Math
The U.S. ranks 25th out of 34 countries when it comes to kids' math proficiency. One New Jersey parent wants to change that by overhauling the culture of math. An astrophysics graduate and mother of three kids, she started a ritual when each child was two years old: a little bedtime mathematical problem-solving that soon became a beloved routine. Parent friends began to bug her to send them kid-friendly math problems, too. Now Bedtime Math is gaining fans among children and math-shy parents around the country. Read the NPR interview or visit the Bedtime Math website.
About Computing Sciences at Berkeley Lab
The Lawrence Berkeley National Laboratory (Berkeley Lab) Computing Sciences organization provides the computing and networking resources and expertise critical to advancing the Department of Energy's research missions: developing new energy sources, improving energy efficiency, developing new materials and increasing our understanding of ourselves, our world and our universe.
ESnet, the Energy Sciences Network, provides the high-bandwidth, reliable connections that link scientists at 40 DOE research sites to each other and to experimental facilities and supercomputing centers around the country. The National Energy Research Scientific Computing Center (NERSC) powers the discoveries of 7,000-plus scientists at national laboratories and universities, including those at Berkeley Lab's Computational Research Division (CRD). CRD conducts research and development in mathematical modeling and simulation, algorithm design, data storage, management and analysis, computer system architecture and high-performance software implementation. NERSC and ESnet are Department of Energy Office of Science User Facilities.
Lawrence Berkeley National Laboratory addresses the world's most urgent scientific challenges by advancing sustainable energy, protecting human health, creating new materials, and revealing the origin and fate of the universe. Founded in 1931, Berkeley Lab's scientific expertise has been recognized with 13 Nobel prizes. The University of California manages Berkeley Lab for the DOE’s Office of Science.
DOE’s Office of Science is the single largest supporter of basic research in the physical sciences in the United States, and is working to address some of the most pressing challenges of our time. For more information, please visit science.energy.gov.