Date: Thursday, May 6, 2010
Time: 11 am — 12:15 pm
Place: Mechanical Engineering, Room 218
Bill Feiereisen
Director of High Performance Computing
DoD for Lockheed Martin Corporation
Abstract:
It has become a cliche to state that the biological sciences have become information sciences.
Vastly increased volumes of experimentally acquired genomic and proteomic data hint at rich
new insights in many areas of the biological sciences, but the demands they place on computing
for their analysis are just as great. This is one of the many reasons why scientists from the more
traditional areas of high performance computing have been attracted into biology. However, the
character of this computing has changed -- away from modeling and simulation, upon which
much of our high performance computing expertise is based, to the extraction of scientific
insight from data analysis.
This talk discusses my journey in moving from traditional computational simulation into high
performance bioinformatics. The motivation occurs through global climate modeling and the
very large contribution that the microbial biology of the ocean has upon the carbon dioxide
budget in ocean models. Current microbial models incorporated into ocean models presume
knowledge of the organisms present and their metabolism. In reality, recent "metagenomic"
ocean surveys have shown that most organisms are not known or understood, nor do we know
about their spatial and temporal distribution. So, how would we use this new information to
evaluate the performance of current models or build new ones?
Metagenomics is the study of microbial communities in situ. Over 99% of microbes in the ocean
cannot be studied in the lab, because they cannot be separated from the symbiosis of their
community and survive. Their genomes must be acquired together and teased apart with new
computational algorithms. I will discuss work in sequence based and similarity based algorithms
to categorize the mixed fragments of DNA for assembly into complete genomes. Comparison of
these genome fragments and complete genomes can be performed through multiple alignment
algorithms. Both of these algorithmic tasks are now overwhelming our high performance
computing capability and point the way to fertile new fields for algorithm developers.
Bio:
Bill Feiereisen is the Director
of High Performance Computing, DoD for Lockheed Martin Corporation. He was formerly
the laboratory Chief Technologist and Division Director of the Computer and
Computational Sciences Division at Los Alamos and before that the head of the
NASA Advanced Supercomputing Facility at Ames Research Center.
He is active in the broader computer science community, serving on the editorial board of IEEE
Computers in Science and Engineering, as the former chairman of the Advisory Committee for the Open
Grid Forum, on the council of the NSF Computing Community Consortium and as a founding member of
the New Mexico Computing Applications Center. He is a member of numerous review boards and
advisory committees.
Bill's original interests in high performance computing come from computational fluid dynamics and
range from turbulence modeling to rarefied gas dynamics with applications to achinery and hypersonic flows. However the computer science of high performance computing that underlies
computational science has been a motivator for his work since the nineties. His recent computational
interests are in the field of bioinformatics and high performance computing.
He holds a Doctorate and Masters in Mechanical Engineering from Stanford University and a Bachelors
Degree from the University of Wisconsin.
In his copious free time he is a wannabe bicycle racer and can usually be found running last in club races
in New Mexico.
Date: Tuesday, May 4, 2010
Time: 11 am — 12:15 pm
Place: Mechanical Engineering, Room 218
Daniel Dennett
Center for Cognitive Studies
University Professor
Austin B. Fletcher Professor of Philosophy
Tufts University
Bio:
Daniel Dennett is currently Miller Fellow at SFI, and University Prof.
and Co-Director, Center for Cognitive Studies, at Tufts University.
He is the author of CONSCIOUSNESS EXPLAINED (1991), DARWIN'S DANGEROUS
IDEA (1995), and various articles on robotics, AI, computers and the
mind, technology, etc. His most recent book is BREAKING THE SPELL:
RELIGION AS A NATURAL PHENOMENON (2006).
Date: Tuesday, April 27, 2010
Time: 11 am — 12:15 pm
Place: Mechanical Engineering, Room 218
Kiri Wagstaff
Senior Researcher in artificial intelligence and machine learning
Jet Propulsion Laboratory
Abstract:
Imagine a machine learning agent deployed at each station in a sensor
network, so that it can analyze incoming data and determine when
something interesting happens. Traditionally, this analysis would be
done independently at each station. But what if each agent could talk
to its neighbors and find out what they're seeing? We've developed a
learning system that enables collaboration so that the agents can
autonomously (without human input) improve their performance. Each
agent can ask its neighbors for their opinions, then use them to
refine its own results. When each agent is given the task of
clustering the observed data, the opinions are expressed in the form
of pairwise clustering constraints. We evaluated several heuristics
for selecting which items an agent should query and found that the
best strategy was to select one item close to its assigned cluster and
one item at the boundary between two clusters. We applied this
technique to seismic and infrasonic data collected by the Mount Erebus
Volcano Observatory, in which the goal was to separate eruptions from
non-eruptions. Collaborative clustering achieved a 150% improvement
over regular, non-collaborative clustering. This is joint work with
Jillian Green (California State Univ., Los Angeles), Rich Aster and
Hunter Knox (New Mexico Institute of Mining and Technology), Terran
Lane (Univ. of NM), and Umaa Rebbapragada (Tufts Univ.), funded by the
NSF.
Bio:
Kiri Wagstaff is a senior researcher in artificial intelligence and
machine learning at the Jet Propulsion Laboratory. Her focus is on
developing new machine learning and data analysis methods,
particularly those that can be used for in situ analysis onboard
spacecraft (orbiters, landers, etc.). She has developed several
classifiers and detectors for data collected by instruments on the
EO-1 Earth orbiter, Mars Pathfinder, and Mars Odyssey. The
applications range from detecting dust storms on Mars to predicting
crop yield on Earth. She holds a Ph.D. in Computer Science from
Cornell University (2002) and an M.S. in Geological Sciences from the
University of Southern California (2008).
Date: Thursday, April 22, 2010
Time: 2 pm (Panel begins at 3 pm )
Place: Mechanical Engineering, Room 218
Herbert S. Lin
Chief Scientist, CSTB
National Academies
Abstract:
Much has been written about the possibility that terrorists or hostile
nations might conduct cyberattacks against critical sectors of the
U.S. economy. However, the possibility that the United States might
conduct its own cyberattacks -- defensively or otherwise -- has
received almost no public discussion. Recently, the US National
Academies performed a comprehensive unclassified study of the
technical, legal, ethical, and policy issues surrounding cyberattack
as an instrument of U.S. policy. This talk will provide a framework
for understanding this emerging topic and the critical issues that
surround it.
Bio:
Herbert S. Lin is chief scientist for the National Research Council's
Computer Science and Telecommunications Board where he directs major
study projects at the intersection of public policy and information
technology. Relevant for this talk, he was study director of the 2009
Academy study "Technology, Policy, Law, and Ethics Regarding U.S.
Acquisition and Use of Cyberattack Capabilities." He previously
served as staff scientist in defense policy and arms control for the
House Armed Services Committee. Lin holds a doctorate in physics from
the Massachusetts Institute of Technology.
Panel:
David Ackley (Associate Professor, UNM Dept. of Computer Science;External Professor at the Santa Fe Institute)
Daniel Dennett (University Professor and Austin B. Fletcher Professor of Philosophy, Tufts University; Miller Scholar, Santa Fe Institute)
Robert Hutchinson (Senior Manager for Computer Science and Information Operations; Sandia National Laboratories)
Herbert Lin (Chief Scientist at the Computer Science and Telecommunications Board, National Research Council of the National
Academies)
Andrew Ross (Director, UNM Center for Science, Technology and Policy; Professor, UNM Dept. of Political Science)
The report is availble here
and as a PDF.
Questions about this event can be directed to crandall at cs.unm.edu
Date: Monday, April 12, 2010
Time: 2 pm — 3 pm
Place: George Pearl Hall, Room 101
Abstract:
Tor is a free-software anonymizing network that helps people around
the world use the Internet in safety. Tor's 1600 volunteer relays
carry traffic for several hundred thousand users including ordinary
citizens who want protection from identity theft and prying
corporations, corporations who want to look at a competitor's website
in private, and soldiers and aid workers in the Middle East who need
to contact their home servers without fear of physical harm.
Tor was originally designed as a civil liberties tool for people in
the West. But if governments can block connections *to* the Tor
network, who cares that it provides great anonymity? A few years ago
we started adapting Tor to be more robust in countries like China. We
streamlined its network communications to look more like ordinary SSL,
and we introduced "bridge relays" that are harder for an attacker to
find and block than Tor's public relays.
In the aftermath of the Iranian elections in June, and then the late
September blockings in China, we've learned a lot about how
circumvention tools work in reality for activists in tough situations.
I'll give an overview of the Tor architecture, and summarize the
variety of people who use it and what security it provides. Then we'll
focus on the use of tools like Tor in countries like Iran and China:
why anonymity is important for circumvention, why transparency in
design and operation is critical for trust, the role of popular media
in helping -- and harming -- the effectiveness of the tools, and
tradeoffs between usability and security. After describing Tor's
strategy for secure circumvention (what we *thought* would work), I'll
talk about how the arms race actually seems to be going in practice.
Bio:
Roger Dingledine is project leader for The Tor Project, a US
non-profit working on anonymity research and development for such
diverse organizations as the US Navy, the Electronic Frontier
Foundation, and Voice of America. In addition to all the hats he wears
for Tor, Roger organizes academic conferences on anonymity, speaks at
a wide variety of industry and hacker conferences, and also does tutorials on
anonymity for national and foreign law enforcement.
Date: Thursday, April 8, 2010
Time: 11 am — 12:15 pm
Place: CEC auditorium (Not the normal place at Mechanical Engineering, Room 218)
Melanie Mitchell
Portland State University
Santa Fe Institute
Abstract:
Enabling computers to understand images remains one of the hardest
open problems in artificial intelligence. No machine vision system
comes close to matching human ability at identifying the contents of
images or visual scenes or at recognizing similarity between different
scenes, even though such abilities pervade human cognition. In this
talk I will describe research---currently in early stages---on
bridging the gap between low-level perception and higher-level image
understanding by integrating a cognitive model of perceptual
organization and analogy-making with a neural model of the visual
cortex.
Bio:
Melanie Mitchell is Professor of Computer Science at Portland
State University and External Professor at the Santa Fe Institute.
She attended Brown University, where she majored in mathematics and
did research in astronomy, and the University of Michigan, where she
received a Ph.D. in computer science, working with her advisor Douglas
Hofstadter on the Copycat project, a computer program that makes
analogies. She is the author or editor of five books and over 70
scholarly papers in in the fields of artificial intelligence,
cognitive science, and complex systems. Her most recent book,
"Complexity: A Guided Tour", published in 2009 by Oxford University
Press, was named by Amazon.com as one of the ten best science books of
2009.
Date: Thursday, April 1st, 2010
Time: 11 am — 12:15 pm
Place: Mechanical Engineering, Room 218
Alexandros Stamatakis
Department of Computer Science
Technical University of Munich
Abstract:
The reconstruction of phylogenetic (evolutionary) trees from molecular
sequence data under the Maximum Likelihood model represents a compute and
memory intensive task.
In this talk I will address how to orchestrate the phylogenetic likelihood
function on a large variety of parallel architectures ranging from FPGAs, over
multi-core processors, to the IBM BlueGene architecture. I will also address
the challenges of maintaining a production level sequential and parallel
open-source code for phylogeny reconstruction and discuss some interesting
bugs. I will conclude with future challenges in the field.
Date: Thursday, March 11, 2010
Time: 11 am — 12:15 pm
Place: Mechanical Engineering, Room 218
Joao Dias
Department of Electrical Engineering and Computer Science
Tufts University
Abstract:
In programming languages, a new idea may help programmers express
algorithms more easily or may guarantee that programs behave better.
To evaluate such an idea, you need to see how programmers use it in
practice, so you need an implementation which is good enough that
programmers will actually use it. Historically, while programmers may
be forgiving at first, eventually they demand compilers that generate
native machine code of high quality. Such compilers are difficult
to build. The goal of my research is to develop new ways of building
high-quality compilers cheaply.
In building a high-quality compiler, one of the big costs is finding
the services of a person who is expert in *both* the compiler *and*
the target machine. The main consequence of my work is that such
an expert is no longer needed: from a formal description of the
semantics of a target machine, I can *generate* a translator that
chooses target-machine instructions. Generating translators for
such machines as x86, PowerPC, and ARM takes just minutes. These
results rest on three technical contributions:
- I proved that the problem is undecidable in general, so any
attack must involve heuristic search.
- I developed a new search algorithm that, unlike prior work,
explores *only* computations that can be implemented on the
target machine.
- I developed a new pruning heuristic that enables my algorithm
to explore long sequences of instructions without allowing
search times to explode.
Date: Tuesday, March 9, 2010
Time: 11 am — 12:15 pm
Place: Mechanical Engineering, Room 218
Lydia Tapia
Institute for Computational Engineering and Sciences (ICES)
The University of Texas at Austin
Abstract:
At first glance, robots and proteins have little in common. Robots
are commonly thought of as tools that perform tasks such as vacuuming
the floor, while proteins play essential roles in many biochemical
processes. However, the functionality of both robots and proteins is
highly dependent on their motions. In the case of robots, complex
spaces and many specialized planning methods can make finding feasible
motions an expert task. In the case of protein molecules, several
diseases such as Alzheimer's, Parkinson's, and Mad Cow Disease are
associated with protein misfolding and aggregation. Understanding of
molecular motion is still very limited because it is difficult to
observe experimentally. Therefore, intelligent computational tools
are essential to enable researchers to plan and understand motions.
In this talk, we draw from our unique perspective from robotics to
present a novel computational approach to approximate complex motions
of proteins and robots. Our technique builds a roadmap, or graph, to
capture the moveable object's behavior. This roadmap-based approach
has also proven successful in domains such as animation and RNA
folding. With this roadmap, we can find likely motions (e.g., roadmap
paths). For proteins, we demonstrate new learning-based map analysis
techniques that allow us to study critical folding events such as the
ordered formation of structural features and the time-based population
of roadmap conformers. We will show results that capture biological
findings for several proteins including Protein G and its structurally
similar mutants, NuG1 and NuG2, that demonstrate different folding
behaviors. For robots, we demonstrate new learning-based map
construction techniques that allow us to intelligently decide where
and when to apply specialized planning methods. We will show results
that demonstrate automated planning in complex spaces with little to
no overhead.
Bio:
Lydia Tapia is a Computing Innovation Post Doctoral Fellow in the
Institute for Computational Engineering and Sciences at the University
of Texas at Austin working with Prof. Ron Elber. She received a
Ph.D. in 2009 from Texas A&M University after working with Prof. Nancy
Amato. At A&M she participated as a fellow in the Molecular
Biophysics Training and GAANN programs and was awarded a Sloan
Scholarship and a P.E.O. Scholars Award. Lydia also attended Tulane
University where she received a BS in Computer Science with academic
and research honors. Prior to graduate school, she worked as a member
of technical research staff as part of the Virtual Reality Laboratory
at Sandia National Laboratories. More information about Lydia Tapia's
research and publications can be found at http://parasol.tamu.edu/~ltapia
Date: Thursday, March 4, 2010
Time: 11 am — 12:15 pm
Place: Mechanical Engineering, Room 218
Jacob Sorber
Dept. of Computer Science
University of Massachusetts-Amherst
Abstract:
Recent advances in low-power electronics, energy harvesting, and sensor technologies
are poised to revolutionize the field of mobile computing, by enabling mobile systems
that are long-lived, energy-aware, and self-managing. When realized,
this new generation of perpetual systems will have a transformational impact,
improving ability to observe natural phenomena, providing network
services to remote communities, and enabling many ubiquitous computing applications
for which regular maintenance is not feasible. Unfortunately, energy and mobility
make building even the simplest systems challenging. Energy harvesting is highly
variable, battery storage is limited, and mobility introduces sparse connectivity.
Instead of rising to meet these challenges, current mobile applications,
operating systems, and languages have evolved very little from their desktop computing origins.
In this talk, I will describe challenges, results, and lessons learned from developing self-tuning mobile sensing systems in the context of two ongoing wildlife studies, focused on endangered tortoises and invasive mongooses. Specifically, I will describe language, runtime, and network techniques that simplify programming energy-aware systems and provide energy efficiency and fairness in energy-constrained networks.
Bio:
Jacob Sorber is a PhD candidate in
Computer Science at the University of Massachusetts-Amherst, graduating Summer 2010.
His research focuses on mobile systems, pervasive computing, and sensor networks,
with an emphasis on making mobile computing systems more flexible, energy-aware, and self-managing.
Date: Tuesday, March 2, 2010
Time: 11 am — 12:15 pm
Place: Mechanical Engineering, Room 218
Todd Mytkowicz
Dept. of Computer Science
University of Colorado
Abstract:
To evaluate an innovation in computer systems a performance analyst
measures execution time or other metrics using one or more standard
workloads. In short, the analyst runs an experiment. To ensure the
experiment is free from error, s/he carefully minimizes the amount of
instrumentation, controls the environment in which the measurement
takes place, repeats the measurement multiple times, and uses
statistical techniques to characterize her/his data.
Unfortunately, even with such a responsible approach, the analyst's
experiment may still be misleading because of bias. A biased
experiment occurs when one experimental setup---or the environment in
which we carry out our measurements---inadvertently favors a
particular outcome over others.
In this talk, I demonstrate that bias is large enough to mislead
systems experiments and common enough that it cannot be ignored by the
systems community. I describe tools and methodologies that my
co-authors and I developed to mitigate the impact of bias on our
experiments. Finally, I conclude with my future plans for
research---tools that aid performance analysts in understanding the
complex behavior of their systems.
Bio:
Todd Mytkowicz recently defended his Ph.D. in Computer Science at the
University of Colorado, advised by Amer Diwan and co-advised by
Elizabeth Bradley. During his graduate tenure he was lucky enough to
intern at both Xerox' PARC and IBM's T.J. Watson research lab. He was
also a visiting scholar at the University of Lugano, Switzerland. His
research interests focus on performance analysis of computer
system---specifically, he develops tools that aid programmers in
understanding and optimizing their systems.
Date: Thursday, February 25th, 2010
Time: 11 am — 12:15 pm
Place: Mechanical Engineering, Room 218
Westley Weimer
Assistant Professor
Dept. of Computer Science
University of Virginia
Abstract:
Automatic program repair has been a longstanding goal in software engineering,
yet debugging remains a largely manual process. We introduce a fully automated method
for locating and repairing bugs in software. The approach works on off-the-shelf legacy
applications and does not require formal specifications, program annotations or special coding practices.
Once a program fault is discovered, an extended form of genetic programing is used to evolve program variants
until one is found that both retains required functionality and also avoids the defect in question.
Standard test cases are used to exercise the fault and to encode program requirements.
After a successful repair has been discovered, it is minimized using structural differencing
algorithms and delta debugging. We describe the proposed method and report experimental results
demonstrating that it can successfully and rapidly repair multiple types of defects from many different programs.
Date: Thursday, February 18th, 2010
Time: 11 am — 12:15 pm
Place: Mechanical Engineering 218
Jesse Davis
Department of Computer Science and Engineering
University of Washingoton
Abstract:
Machine learning has become an essential tool for analyzing biological and clinical data, but significant
technical hurdles prevent it from fulfilling its promise. Standard algorithms make three key assumptions:
the training data consist of independent examples, each example is described by a pre-defined set of attributes,
and the training and test instances come from the same distribution. Biomedical domains consist of complex,
inter-related, structured data, such as patient clinical histories, molecular structures and protein-protein
interaction information. The representation chosen to store the data often does not explicitly
encode all the necessary features and relations for building an accurate model.
For example, when analyzing a mammogram, a radiologist records many properties of each abnormality,
but does not explicitly encode how quickly a mass grows, which is a crucial indicator of malignancy.
In the first part of this talk, I will focus on the concrete task of predicting whether an abnormality
on a mammogram is malignant. I will describe an approach I developed for automatically discovering
unseen features and relations from data, which has advanced the state-of-the-art for machine
classification of abnormalities on a mammogram. It achieves superior performance
compared to both previous machine learning approaches and radiologists.
Bio:
Jesse Davis is a post-doctoral researcher at the University of Washington.
He received his Ph.D in computer science at the University of Madison in 2007 and a B.A.
in computer science from Williams College in 2002. His research interests include machine learning,
statistical relational learning, transfer learning, inductive logic programming and data mining for biomedical domains.
Date: Tuesday, January 26th, 2010
Time: 11 am — 12:15 pm
Place: Mechanical Engineering 218
Majeed M. Hayat
Professor
Electrical and Computer Engineering
University of New Mexico
Abstract:
The ability to model and optimize reliability and task-execution speed is central in designing survivable distributed
computing systems (DCSs) where servers are prone to fail, possibly permanently and in a spatially correlated manner.
Correlated component failures in networks have been receiving attention in recent years from government agencies
due to their association with damage from weapons of mass destruction. In this talk we discuss the problem of
modeling service reliability and task-execution speed of a DSC in uncertain topologies as well as the problem
of load balancing in such environments. Service reliability and the mean task-execution time are analytically
characterized by means of a novel regeneration-based probabilistic technique. The analysis takes into account
the stochastic failure times of servers, the heterogeneity and uncertainty in service times and communication delays,
as well as arbitrary task-reallocation policies. Two models are presented: the first one assumes Markovian
(exponentially distributed) communication and service random times, and the second relaxes this assumption.
The theory is utilized to optimize certain load-balancing policies for maximal service reliability or
minimal task-execution time; the optimization is carried out by means of an algorithm that scales
linearly with the number of nodes in the system. The analytical model is validated using both Monte-Carlo simulations and experiments.
Bio:
Majeed M. Hayat was born in Kuwait, in 1963.
He received the B.S. degree (summa cum laude) in electrical engineering from the
University of the Pacific, Stockton, CA, in 1985, and the M.S. and Ph.D. degrees in
electrical and computer engineering from the University of Wisconsin-Madison, Madison, in 1988 and 1992.
He is currently a Professor of Electrical and Computer Engineering and a member of the
Center for High Technology Materials at the University of New Mexico, Albuquerque.
His research contributions cover a broad range of topics in signal/image processing and applied probability.
His current areas of interest include image processing and noise reduction in thermal images,
algorithms for infrared spectral sensing and recognition, queuing models and strategies for
resilient distributed systems and networks, modeling of noise and stochastic carrier dynamics in avalanche photodiodes,
and performance characterization of optical receivers and photon counters.
News & Events | People | Academics | Research | Jobs | Companies | Visitors | Computer Facilities | Home
