*Lunch will be provided*

SPEAKER:
Matthew Graham
Computational Scientist
Center for Advanced Computing Research, Caltech

TITLE:
Characterizing the Time Domain

ABSTRACT:
The new generation of synoptic sky surveys promise unprecedented amounts of data and information and automated processing and analysis is a necessity. Light curves, however, can show tremendous variation in their temporal coverage, sampling rates, errors and missing values, etc., which makes comparisons between them difficult and training classifiers even harder. A common approach to tackling this is to characterize a set of light curves via a set of common features and then use this alternate homogeneous representation as the basis for further analysis or training. Many different types of feature are used in the literature to capture information contained in the light curve: moments, flux and shape ratios, variability indices, periodicity measures, model representations. In this talk, we will review characterization features with particular attention to the problem of determining accurate and reliable periods for astrophysical objects.

Jiao Lin
Computational Scientist, Center for Advanced Computing Research (CACR)

Tuesday Sept 20
11AM
100 Powell-Booth

Abstract: Intuitive, responsive, and clean graphical user interface has become more and more important for scientific software applications. Building graphical user interface is tedious, however. Without extreme care, a user interface application can easily become unnecessarily complex and convoluted, and as a result, unmaintainable. Building web-based graphical user interface is harder due to inconsistent implementations of languages among browsers and multiple languages/standards/platforms that could be involved, and that renders management of a web UI project expensive, and sometimes chaotic. With the emergence of cloud computing, we will see many scientific computing packages turning to cloud and demand web or mobile-device user interfaces, while the traditional desktop user interface still has its large user base. A much simplified route of developing desktop/web/mobile-device user interface is needed. This work looks for the most compact set of abstract concepts and principles enough for constructing sophisticated UI. In practical, it intends to reduce the chaos and agony in building user interface applications, to dramatically lower the barrier of creating good user interfaces, and to make it much easier to maintain and evolve them.

Gustavo Kunde Rohde, Assistant Professor, Biomedical Engineering, Carnegie Mellon University

Monday, May 16, 2011
2:00 PM – 3:00 PM
Beckman Institute Auditorium

Novel biomedical imaging techniques have enabled the acquisition of quantitative information from cells, tissues, and organs with unprecedented accuracy and specificity. Combined with the availability of vast computational resources, quantitative biomedical imaging pipelines have the potential to accelerate scientific discovery and improve clinical practice. An important engineering problem in this area relates to extracting quantitative information related to the form (shape and texture) of cells, tissues, and organs. I will describe our recent efforts toward the development of a general purpose segmentation method and present preliminary evidence that a tool capable of high-enough accuracy for quantitative imaging pipelines may one day be available. In addition, recent efforts in developing geometric data analysis tools for mining morphological information from biomedical image data will be described. In particular, I will describe the application of deformation and transportation related metrics, in combination with discriminant analysis techniques, towards understanding the distribution of cellular patterns in cancerous and normal tissues.

Thursday January 6, 2011
2:30 PM
100 Powell-Booth

Abstract

Machine learning and data prediction is crucial to most organizations. Banks predict which loan applicants are likely to default, treasuries forecast tax revenues and medical researchers predict the likelihood of illness from gene sequences.

Crowdsourced data mining can lead to vastly better models. My project, Kaggle, recently hosted a bioinformatics contest, which required participants to pick markers in a series of genetic sequences that predict the progression of HIV. Within a week and a half, the best submission had already outdone the best methods in the scientific literature.

This result neatly illustrates the strength of competitions. Whereas the scientific literature or in-house models tend to evolve slowly (somebody tries something, somebody else tweaks that approach and so on), a competition inspires rapid innovation by introducing the problem to a wide audience. There are an infinite number of approaches that can be applied to any machine learning problem and it is impossible to know at the outset which technique will be most effective.

Bio

Anthony is the Founder and CEO of Kaggle, a global platform for data prediction competitions. In addition to founding Kaggle, Anthony continues to consult to hosts of Kaggle competitions to help them frame prediction tasks, to get the best out of the new platform and help them integrate insights into their day-to-day operations.

Before Kaggle, Anthony was a macroeconomic modeler for the Reserve Bank of Australia and before that the Australian Treasury. In these roles, Anthony built and maintained macroeconomic models of Australia’s economy to improve forecasting and model the economic effect of changes in policy parameters, such as interest rates and fiscal policy.

Anthony graduated with first class honours in econometrics at the University of Melbourne and has published in The Economist magazine and the Australian Economic Review.

A (Hypothetical) Data to Discovery Engine
Mark Stalzer, Caltech – CACR

Description/Abstract:
Moore’s law works for semiconductor-based detectors and there is an increasing flood of data being generated in astronomy, high energy physics, biology, and other sciences. Computation is essential for both (1) making predictions from theory and (2) the analysis of data from experiments. Is there a way to architecturally balance both needs in a high performance computing (HPC) system?

HPC systems are typically constructed from easily available parts, just organized differently and scaled to process extreme workloads. The most power efficient petascale machine as of 2010 is Roadrunner at the Los Alamos National Laboratory. Another interesting machine is the Apple iPad which uses very low power System on a Chip/Package on Package technologies. This talk explores the question of “what happens when you cross Roadrunner with iPads”? The result is a high level of integration between computation and storage on a single server blade, called a Flashblade, with 100x-1,000x performance improvements for some data-centric applications. The Flashblade architecture, expected performance, programming, and scaling with advancing technology are discussed.

(See publication page for PDF link)

Andrew Drake, computational scientist, Caltech Center for Advanced Computing Research

The Catalina Real-time Transient Survey (CRTS) is a Caltech operated optical transient survey that covers most of the Northern and Southern sky in search of transient astrophysical phenomena occurring on timescales of minutes to years. The project uses data from the Catalina Sky Survey NEO search and began real-time discovery and publication of transient events in November 2007. CRTS has found thousands of sources ranging from UV Ceti and dwarf nova outbursts to supernovae and Blazars. I will discuss the survey, the discoveries made to date, and our efforts to provide immediate open access to CRTS discoveries and historical CSS data.

* For further information: contact Gina Armas gina@its.caltech.edu phone: 4671
For the full scoop, see event web page: http://www.astro.caltech.edu/~gma/colloquia.html.
* Sponsored by: Physics, Math and Astronomy

March 12, 2010
2PM
147 Noyes

Abstract

DARPA recently funded a 2 year study of the technical challenges of trying to go from today’s petascale computing to exascale – 1000X – in roughly half the time it took to get from terascale to petascale. This talk will summarize this study, with a particular focus on the most far-reaching of the challenges, namely energy. This will be expanded on by an overview of a recent study of energy consumption within the Linpack algorithm which indicates strongly that we have probably crossed a threshold where the real energy and power problems of the future are in the memory and interconnect – not the processing logic.

Bio

Dr. Peter Kogge currently holds the Ted McCourtney Chair of Computer Engineering at the University of Notre Dame, with research interests in highly scalable computer architectures and nano-technologies. Prior to that he was an IBM Fellow with IBM’s Federal System where among other projects he oversaw the development of arguably the world’s first multi-core chip in 1993 – on a DRAM process. He is the author of 2 books, including the first on the now ubiquitous technique of pipelining, and holds over 30 patents. Applications of his PhD research led to what is now called the Kogge-Stone adder, the fastest known adder constructed out of fixed fanout gates. He was also the chairman of the DARPA working group that developed the Exascale report.

Melaine Stefan
Computational Neurobiology Group
European Bioinformatics Institute (EBI)
Cambridge UK

Aviral Shrivastava
Assistant Professor, Department of Computer Science and Engineering,
School of Computing and Informatics, Arizona State University

Unlike design or manufacturing errors, soft errors are much harder to protect against, since they can happen anywhere and anytime. Consequently, soft error protection mechanisms incur extremely high overheads.  Most software schemes against soft errors are based on re-execution of programs or parts thereof, to detect and correct soft errors, and therefore have very high performance (avoidable by adding more resources) power (unavoidable) overheads. In contrast, our research tries to achieve protection without explicit re-execution. Fundamentally, our compiler changes the way application uses microarchitectural components, so that it uses the already protected components to process the vulnerable data, and uses the unprotected components to process the protected data. Towards this philosophy goal, we have developed several compiler techniques to better use caches, and register files (RFs), two sites which most critically need protection. While I’ll gloss over our solutions for the cache, I will concentrate on the RF protection problem.

BIO:
Aviral Shrivastava is an Assistant Professor in the Department of Computer Science and Engineering, at the Arizona State University, where he has established and heads the Compiler and Microarchitecture Lab (CML).   He received his master’s and doctorate in Information and Computer Science from University of California, Irvine. He received his bachelors in Computer Science and Engineering from Indian Institute of Technology, Delhi. His research interests lie at the intersection of compilers and microarchitectures in particular of embedded systems. He studies microarchitecture and compiler techniques for power, performance, temperature, and most recently reliability.  Dr. Shrivastava is a lifetime member of ACM, and serves on organizing and program committees of several premier embedded system conferences, including CODES+ISSS, CASES and LCTES.