Lab to Host DOE Announcement of First Companies to Benefit from HPC4Mfg Program
On Wednesday, February 17, David Danielson, Assistant Secretary for Energy Efficiency and Renewable Energy (EERE), and Mark Johnson, Director of EERE’s Advanced Manufacturing Office will announce the first round of industry partners selected for the new High Performance Computing for Manufacturing projects. NERSC will provide computing resources for some of the participating firms.
The project is led by Lawrence Berkeley National Laboratory. The Oak Ridge Leadership Computing Facility is also a partner. The selected firms will receive $3 million in funding through this national lab-run program that will address key challenges facing U.S. manufacturing. These collaborations will apply modeling, simulation, and data analysis to industrial processes and materials to lower production costs and bring new products to market more quickly. »Read the media advisory.
Call for Nominations for 2016 NERSC Scientific Achievement Awards
Nominations are open for the »2016 NERSC Awards for Innovative Use of High Performance Computing and for High Impact Scientific Achievement. NERSC principal investigators, project managers, PI proxies, and DOE program managers may nominate any NERSC user or collaborating group. The deadline for nominations is Friday, March 4, 2016. Winners will be announced at the NERSC Users Group meeting on March 22, 2016 and will be listed on the NERSC web site and highlighted in a NERSC press release. Winners will be chosen by representatives from the NERSC Users Group Executive Committee and NERSC staff.
Selections will be made based on innovations and achievements that are substantially based on work performed using NERSC resources. Those resources could be any combination of computational systems, storage systems, and/or NERSC HPC services.
Details Set for Greg Bell Farewell
From 2 to 3pm on Thursday, Feb. 25, computing sciences will gather in Wang Hall conference room 3101 for refreshments in honor of ESnet’s Greg Bell. After 15 years at Berkeley Lab, including the last six with ESnet, Greg is on to his next adventure as CEO for Broala, a start-up company that deploys the Bro network monitoring software first developed at the lab.
Friday: Cal Innovators Social Mixer
The Cal Innovators Social Mixer, held from 5 to 6:30pm at Berkeley Skydeck (2150 Shattuck Ave., Berkeley), will bring together graduate students from the Schools of Engineering, Computer Science, Information, and Business, along with members of the Lawrence Berkeley National Laboratory and the SkyDeck startup community to meet, learn about each other’s interests, and begin forming relationships that could lead to collaboration opportunities down the road.
This is first and foremost a social environment that will provide participants with an opportunity to discuss personal passions, planned projects, current work or just to learn about new topics. Dress is casual. Drinks and light appetizers will be provided. So come ready to learn, share your interests, have some drinks and form personal and professional connections with the best the Berkeley community has to offer.
This event is in partnership with Skydeck, the Berkeley Entrepreneurship Association and the Lawrence Berkeley National Laboratory. Due to the presence of alcohol, this event is only for those age 21 and older. ID will be checked at the door. For more information see the »FaceBook Event Page. Or »register now at Eventbrite.
Save the Date: BIDS Spring 2016 Data Science Faire
Join the Berkeley Institute for Data Science from 1:30 to 4:30pm, May 3 for their Spring 2015 Data Science Faire. This year’s faire closes out BIDS’ second academic year while celebrating data science at Berkeley.
At this year’s Data Science Faire, we will showcase exciting data-intensive initiatives at BIDS and UC Berkeley, highlighting work from the diverse community of data scientists around campus. Learn more about the exciting open source projects BIDS affiliates are working on and catch up on the work of researchers at some of Berkeley’s top data science centers, including the National Energy Research Scientific Computing Center (NERSC) and Berkeley Research Computing. Details forthcoming.
This Week’s CS Seminars
Alvarez Fellowship Seminar
Design, Implementation, and Optimization of Irregular Graph Algorithms on HPC Platforms
Tuesday, Feb. 16, 10 – 11am, Bldg. 50B, Room 4205
George M. Slota, Penn State University
Graphs are common, complex, and can be very large, which makes them important to study yet computationally challenging to work with. The extreme irregularity of many real-world graphs make the parallelization of algorithms to analyze such graphs difficult. With the massive parallelism of GPUs, Xeon Phis, and distributed systems, the challenges are even harder for algorithm designers to overcome. I will present my efforts that include the development of several techniques for effective multicore, manycore, and distributed parallelization of graph algorithms. This includes research that has resulted in FASCIA, a subgraph enumeration tool in shared and distributed memory that runs several orders-of-magnitude faster than prior work; Multistep, a method for shared-memory graph connectivity on average 2x faster than the most recent state-of-the-art; manycore optimizations, that accelerate graph algorithms on manycore processors such as GPUs and Xeon Phis up to 3x faster than highly optimized CPU code; and PuLP, a low overhead graph partitioner that scales to networks over hundreds of billions of edges. Additionally, I’ll present how using techniques derived from these efforts, a suite of distributed graph analytics was implemented and applied to the largest publicly-available web crawl of 3.5 billion pages and 130 billion links, with end-to-end execution of analysis completing in under 20 minutes on only 256 nodes of the Blue Waters system. At a broad scope, this presentation will discuss my research involving graph algorithm design and implementation in three primary areas: (1) Algorithm design and optimization at the node (shared-memory) level; (2) Algorithm design and optimization at the system (distributed-memory) level; (3) Developing generalized approaches to apply these techniques to real-world problems. My primary research goals are to enable complex analysis of very large graph-structured datasets across a broad range of scientific domains and applications.
CITRIS Research Exchange Seminar
Tackling Urban Challenges with Technology
Wednesday, Feb. 28, 12 – 1pm, 310 Sutardja Dai Hall, Banatao Auditorium – UC Berkeley
Scott Mauvais, Microsoft
No abstract available. The CITRIS Research Exchange Seminar Series is a weekly dialogue highlighting leading voices on societal-scale research issues. Each one-hour seminar starts at 12pm Pacific time and is hosted live at Sutardja Dai Hall on the UC Berkeley campus. »Live webcast available. »Registration recommended.
Prediction of Molecular Properties with Explicitly Correlated Coupled-Cluster Methods
Wednesday, Feb. 17, 1 – 2pm, Bldg. 70, Room 191
Jinmei Zhang, Virginia Tech
During the last few decades, rapid developments in electronic structure theory have allowed the prediction of electronic energies to reach chemical (1 kcal/mol) or even spectroscopic (1 cm−1 ) accuracy. However, such accurate computation requires the combination of high-level wave function models (like coupled-cluster (CC) methods) and larger basis sets, which results in high demand of computational resources. The requirement for large basis sets can be alleviated by using explicitly correlated methods. They include inter-electronic distances (rij ) explicitly in the wave function expansions to account for short-range, two-electron correlations, and thus improve the slow convergence of electron correlation energies with respect to the size of orbital expansions in conventional correlation methods. In our group, we developed the perturbative explicitly correlated CC methods, CC(2)F12, which has the virtue of being technical simple and straightforward to implement. We assessed its performance for accurate studies of various molecular properties including reaction barrier heights, thermochemical properties, and electric dipole and quadrupole moments. We demonstrated that the inclusion of the perturbative F12 correction significantly reduces the basis set error of the correlation energies, and this leads to substantial computational savings.
UCB Applied Math Seminar
Numerical Study of Artificial Microswimmers Propelled by Marangoni Flow
Wednesday, Feb. 17, 3:30 – 4:30pm, 939 Evans Hall – UC Berkeley
Laura Stricker Max-Planck-Gesellschaft
In order to understand the collective behavior of biological microswimmers, such as bacteria and spermatozoa, it is important to know which aspects are governed merely by the physics of the system and which aspects are biology-related. In the present study we address the mechanisms of locomotion of non-biological swimmers. We focus on artificial swimmers consisting in active droplets moving by Marangoni flow, in particular nematic liquid crystal droplets immersed in an aqueous solution containing an ionic surfactant, and self-propelled droplets driven by Belousov-Zhabotinsky (BZ) reactions. We numerically study the flow field outside and inside a single droplet, by means of a level set method, in order to account for both the deformation of the droplet and the interaction with a small number of neighbouring droplets. Further mechanisms that influence the locomotion, such as mass transport, chemical reactions and the coupling between the bulk and the surface concentration of the surfactants, are also specifically addressed.
Alvarez Fellowship Seminar
Machine Learning the Relationship between Structure and Dynamics in Disordered Solids
Friday, Feb. 19, 10 – 11am, Bldg. 50B, Room 4205
Ekin Dogus Cubuk, Harvard University
The structural and mechanical properties of crystals, glasses and biological macromolecules are determined by local interactions between atoms. The relationship between local geometric configurations of atoms and macroscale material properties is often very complex, which can prevent simple analytical models from making useful predictions. I will argue that machine learning representations of materials data, as extracted from unsupervised and supervised frameworks, can be very helpful in elucidating the structural complexities of materials. As an example, I will present recent work in which I constructed predictive models of dynamics in disordered solids; as applied to silica, silicon, and Lennard-Jones liquids. These models can explain the equilibrium and nonequilibrium behavior of strong and fragile glass-formers from a local structural perspective, which has not been possible without using such a data-scientific approach. Finally, I will discuss potential applications of this approach to designing materials with desired mechanical or catalytic properties.
REMOTE ACCESS INSTRUCTIONS
To begin AUDIO conference:
1. Dial Toll-Free Number: 866-740-1260 (U.S. & Canada)
2. International participants dial: Toll Number: 303-248-0285, or Toll-Free: http://www.readytalk.com/intl
3. Enter 7-digit access code: 4867514, followed by “#”
4. Place your phone on “Mute”
To begin WEB conference:
1. Go to:http://www.readytalk.com
2. Enter your access code: 4867514# under “PARTICIPANT-Join A Conference”
BIDS Data Science Lecture
Exascaling Nuclear Innovation
Friday, Feb. 19, 1:10pm to 2:30pm,190 Doe Library – UC Berkeley
Rachel Slaybaugh, UC Berkeley
There is a growing need for innovation in nuclear detection, nuclear security and nonproliferation, and nuclear energy to meet our goals in global security, economic competitiveness, and environmental responsibility. There are incredible opportunities for progress in these important areas enabled by exascale computing and data analysis; however, we need to discover how to effectively capitalize on these opportunities. This talk will touch upon three major challenges: performing computations on exascale architectures effectively and reproducibly, having sufficiently accurate data for the highly resolved solutions we need, and performing analysis on such large datasets efficiently.