AISRP 2008 PI Workshop
May 5 – 7, 2008
Adelphi, MD
Abstracts
aisrp.nasa.gov
Monday, May 5, 2008
|
8:00 AM |
Registration |
|
|
9:00 AM |
Joe Bredekamp NASA HQ |
Welcome and AISR Program Update |
|
9:30 AM |
Session 1.1 |
|
|
James Head Brown
University |
ADVISER: Optimizing Science Return |
|
|
Mike Turmon JPL |
Classification of Solar Imagery and Prediction for
Nonlinear Systems |
|
|
Robin Morris USRA-RIACS |
Improving Remote Sensed Data Products Using
Bayesian Methodology for the Analysis of Computer Model Output |
|
|
10:30 AM |
Break |
|
|
10:45 AM |
Session 1.2 |
|
|
Michael Broxton CMU/NASA ARC |
Automated DTM Generation for HiRISE and LROC |
|
|
James Tilton NASA GSFC |
Subdue ing RHSEG... |
|
|
Lutz Hamel University of
Rhode Island |
Exploration of Novel Methods to Visualize Genome
Evolution |
|
|
12 Noon |
Lunch |
|
|
1:00 PM |
Session 1.3 |
|
|
Vinay Kashyap SAO/CXC |
Calibration
Uncertainty and Data Analysis |
|
|
Ashit Talduker JPL |
Global cyclone detection and tracking (GLYDER) |
|
|
Kevin Knuth SUNY Albany |
Bayesian Source Separation for Astrophysical
Spectra |
|
|
Olfa Nasraoui University of
Louisville |
Mining Solar Loops to Support Studies of the
Coronal Heating Problem |
|
|
David Wettergreen Carnegie
Mellon University |
Automated Orbital Mapping: Statistical Data Mining
of Orbital Imagery to Analyze Terrain, Summarize its Characteristics and
Draft Geologic Map |
|
|
Tomasz
Stepinski Lunar and Planetary Institute |
Automated
Identification and Characterization of Landforms on Mars |
|
|
3:00 PM |
Break |
|
|
3:20 PM |
Session 1.4 |
|
|
Ralph
Lorenz JHU APL |
Intelligent
Sensor Network Study of Dust Devils |
|
|
Martin
Weinberg U
MASS |
Enabling
Bayesian Inference for the Astronomy Masses |
|
|
Kiri Wagstaff JPL |
Detecting
Surface Changes via Dynamic Landmarking |
|
|
Timothy Newman Univ of Alabama Huntsville |
Auroral
Phenomenon Localization, Classification, and Temporal Evolution Tracking |
|
|
Adnan Ansar JPL |
Multi-modal
Image Registration and Mapping |
|
|
Jay Johnson PPPL |
Higher-Order
Statistical Methods for Geospace Data |
|
|
Gabor Toth University of Michigan |
Development of
an Adaptive MHD Simulation Tool |
|
|
K. Palaniappan Univ of Missouri-Columbia |
Massively
Parallel Imagery Assimilation Using the 3D Multiscale Multicomponent Modeling
Framework (MMMF) |
|
|
6:00 PM |
Adjourn |
|
Tuesday, May 6, 2008
|
8:30 AM |
Session 21. |
|
|
Kanna Rajan Monterey Bay
Aquarium Research Institute |
Adaptive Control for Underwater Vehicles |
|
|
Simon Krughoff Univ of Washington |
Google Sky |
|
|
Bob Hanisch STScI |
NVO Directions |
|
|
Ani Thakar JHU |
Large Scale on Demand Cross-Matching with Open
SkyQuery |
|
|
Hillol Kargupta Univ. of
Maryland, Baltimore County |
Distributed and Peer-to-Peer Data Mining for
Scalable Analysis of Data from Virtual Observatories |
|
|
10:10 AM |
Break |
|
|
10:30 AM |
Session 2.2 |
|
|
D. Aaron
Roberts NASA GSFC |
Valley of Death: Mid-Level TRLs |
|
|
Thomas McGlynn NASA GSFC |
Presence, Personalization and Persistence: A New
Model for Doing Science in a Collaborative Archive Environment |
|
|
Ed Shaya University of
Maryland |
Automated Data Analysis with Knowledge Ontologies |
|
|
Roy Williams Caltech |
A New Network for Gamma-Ray Bursts and other
Immediate Astronomical Events |
|
|
Pasquale
Tricarico Planetary Science Institute |
A Distributed Computing System Supporting Near
Earth Asteroids Research |
|
|
12:10 PM |
Lunch |
|
|
1:00 PM |
Session 2.3 |
|
|
Patrick Coronado NASA GSFC |
Direct Readout: Saving the Earth in 90 Minutes |
|
|
Tamas Gombosi University of
Michigan |
Development of an Adaptive Non-Ideal MHD Simulation
Tool for Multiple Space Science Applications |
|
|
Peter MacNeice NASA GSFC |
Magnetogram Synthesis - A Vital Data Analysis
Component of A Space Weather Prediction Infrastructure |
|
|
Kevin Olson Drexel
University |
PARAMESH: A parallel, adaptive, grid tool for the
Space Sciences |
|
|
John Houck MIT |
HYDRA: A New Paradigm for Astrophysical Modeling,
Simulation, and Analysis |
|
|
Andrew Ptak Johns Hopkins
University |
On-the-fly and Grid Analysis of Astronomical Images
for the Virtual Observatory |
|
|
3:00PM |
Break |
|
|
3:20 PM |
Session 2.4 |
|
|
Kenneth Mighell NOAO |
Parallel-Processing Astrophysical Image-Analysis
Tools |
|
|
James Schombert Univ of
Oregon |
Hacking for Science |
|
|
R. Daniel Bergeron University
of New Hampshire |
Visualization of multiresolution time series data |
|
|
Jeff Scargle NASA Ames |
Novel Methods for the Analysis of Photon-Limited
Data |
|
|
Robin Morris USRA RIACS |
Event Analysis for GLAST |
|
|
Michael Burl JPL |
Directed Exploration of Complex Systems |
|
|
Simon Krughoff University of
Washington |
Next Generation Data Visualization |
|
|
5:30 PM |
Adjourn |
|
|
6:30 PM |
Group Dinner |
|
Wednesday, May 7, 2008
|
8:30 AM |
Session 3.1 |
|
|
Tsengdar Lee NASA HQ |
Modeling Guru |
|
|
Mark
Giulliano STScI |
Multi-Objective
Optimal Scheduling for Space Science Applications |
|
|
John
Dolan Carnegie Mellon University |
An Analytical
Tool for Robot Mission Reliability Prediction |
|
|
Alan Sussman Univ
of Maryland |
Robust Grid Computing using Peer-to-Peer Services |
|
|
Martin Lo JPL |
MTool Data Analysis and
Visualization |
|
|
10:20 AM |
Break |
|
|
10:40 AM |
Session 3.2 |
|
|
Edward Belbruno Innovative
Orbital Design, Inc. |
Mission Extension Using Sensitive Trajectories and
Autonomous Control |
|
|
Tamal Bose Virginia Tech |
Challenges in On-Board for Future Missions |
|
|
Aravind Dasu Utah State
Univ |
SATH: a simulated annealing to hardware compiler |
|
|
Brian Williams MIT |
Diagnosing Complex Software and
Hardware |
|
|
12 Noon |
Lunch |
|
|
1:00 PM |
Session 3.3 |
|
|
Volodymyr Kindratenko NCSA-UIUC |
Astrophysical Algorithms on Novel HPC Systems |
|
|
Alexander Panasyuk SAO |
Innovative Techniques for Producing Line-of-Sight
Corrected Synoptic Maps |
|
|
Rahul Ramachandran University of
AL Huntsville |
A Distributed Knowledge Extraction Framework Based
on Semantic Web Services |
|
|
Mark Richardson Caltech |
Planetary Atmospheric Data Assimilation System |
|
|
Jeff Jewell JPL |
EPISODE - Software for Trajectory Generation and
Nonlinear Continuous Control in the Presence of Uncertainty |
|
|
Kaichang Di Ohio State
University |
Integration
of Orbital, Descent and Ground Imagery |
|
|
Daniel Sorin Duke
University |
Autonomic
Computer Hardware for Space Missions |
|
|
Nand Lal NASA GSFC |
AISRP Code Repository |
|
|
3:30 PM |
Joe Bredekamp NASA HQ |
Closing Remarks |
James W Head
ADVISER: Optimizing
Science Return
James W Head III, Andrew Forsberg , John Huffman, Samuel Fulcomer, James Dickson
Brown University
ADVISER (Advanced Visualization
in Solar System Exploration and Research) is designed to: 1) advance space
science knowledge, exploration capabilities, teaching and outreach, and 2)
research advanced visualization tools for space science and education through a
problem solving environment (PSE). Geoscientists explore planetary surfaces
with virtual, extended versions of traditional field tools to solve significant
scientific problems. The ADVISER PSE has four basic parts: 1) Geoscientist on
the surface, 2) Importation and visualization of integrated data sets, 3) Field
kit development, and 4) Ancillary virtual field instrument development. We
have: 1) Imported and visualized multiple data sets (MOLA altimetry and HRSC
digital terrain models, HiRISE and CTX ultra-high resolution images, CRISM high
spatial and spectral resolution imaging spectrometer data) and wind vectors and
values for different seasonal atmospheric and climate conditions. 2) Placed 50
geoscientists on the surface of Mars to address scientific problems. 3)
Prototyped important aspects of the field kit and ancillary field instruments
(e.g., virtual photography, virtual GPS, and the PDA field notebook) and
provided geoscientists with these tools in the IVR environment. 4) Continued
integrating ROAM3 rendering system into the IVR environment; built toolkit on
top. 5) Provided support for non-immersive desktop and immersive display
environments. 6) Integrated ArcMap to formalize the initial correlation of data
sets for importing to the IVR environment. 7) Developed tablet PC pen-based UI
to traverse planning application of ADVISER. 8) New science results: Analysis
of: a. Ancient regional glaciation on Mars, b. Tropical mountain glaciers, c.
Valley networks on Mars, d. Lakes on ancient Mars, e. Recent gully formation on
Mars, f. Antarctic Dry Valley gully and slope streak analogs, g. Lobate debris
aprons. Antarctic Dry Valley Field planning: To prepare for deployment for the
6-week field season in the Dry Valleys, we imported digital topography and
IKONOS image data to the IVR environment and planned camp locations and field
traverses. Results from this season are being placed in this environment for
further correlation and analysis. 80 students used the facility to plan the
exploration of scientifically interesting sites on Mars.
1) These efforts fulfill the
fundamental goals and objectives of the NASA AISR program, and address the
major goals of NASA as an agency, including the President s Exploration
Initiative.
2) Relevant technologies:
Interactive 3D terrain visualization with very large data sets (high-resolution
topography, image and related data) and user interface development.
3) Initiation of application to
NASA Astronaut Candidate Training: Used for site selection and traverse
planning for a human reference mission to Mars for broad concept certification
for NASA Astronaut training.
1) Undergraduate university
students filled out forms on their use and analysis of the facility and made
suggestions for improvement.
2) Graduate students have
discussed and described their experiences with students and faculty with all
IVR facility personnel in attendance. More detailed reports are underway.
3) Interested undergraduates are
doing hands-on work to help explore optimal user interface ideas (e.g.,
hand-held PDA device controls and applying the latest video-game-like rapid response controls).
1) Incorporation of the IVR
facility for an Engineering course on mission design (continuing).
2) Integration with navigation
system: how to plan for field work and establish traverse planning highlights
with ArcGIS in order to optimize the planned input to the IVR facility.
3) Develop tool functionality in
desktop version to: a) bring ADVISER capabilities to the Geoscience lab, b)
enable comparison of ADVISER tasks done in Cave (CVV) and at conventional
desktop.
4) Develop initial mechanisms
for tele-robotic viewing within Brown.
5) Develop initial mechanisms
for tele-robotic viewing outside of Brown at NASA and with other users.
6) Develop data management
solutions for out of core data sets.
7) Develop techniques so that
large data sets have low impact on interactivity (high frame rate and low
latency).
8) Continue to develop advanced
visualization and rendering techniques so that large geology data sets can be
viewed in immersive VR (i.e., high frame rate, low latency, without inducing
cybersickness).
9) Continue debriefing by user
systems.
10) Extend to additional science
themes and topics.
Robin D.
Morris
Improving Remote Sensed Data
Products Using Bayesian Methodology for the Analysis of Computer Model Output
Robin D
Morris USRA-RIACS
Athanasios
Kottas UCSC
Roberto
Furfaro, Barry Ganapol U of Arizona
Developed
the theory and implementation for computing the main effects and sensitivity
indices for the Leaf-Canopy-Model Radiative Transfer Model that predicts the
radiation reflected from a forest canopy. This revealed some new effects
regarding which biophysical parameters were most important to determining the
reflection in certain spectral bands. A presentation was made at the AGU Fall
Meeting. A paper has been accepted for publication in IEEE Transactions on
Geoscience and Remote Sensing.
NASA is
responsible for the operation of numerous remote sensing satellites, and for
producing biophysical data products from the returned data. This project will
help determine and quantify the level of uncertainty in these data products,
which has important scientific and public policy implications.
We will
track citations to our publications, and downloads of code once it is made
publicly available.
We are
extending the LCM model to predict Bi-directional reflectance, which is
important for the problem of inverting the LCM model. We will address the
inversion problem, and validate the model with archived satellite and field
data.
Michael J. Broxton
Automated DTM
Generation for HiRISE and LROC
http://ti.arc.nasa.gov/visionworkbench/
Michael J. Broxton CMU/NASA Ames Research Center
Laurence Edwards NASA Ames Research Center
Ross Beyer SETI/NASA Ames Research Center
In this, the first of three years for this proposal, we have
completed the steps necessary for software compatibility between the NASA Ames
Stereo Pipeline (ASP) and the USGS Integrated Software for Imagers and
Spectrometers (ISIS). ISIS contains the authoritative collection of camera
models for NASA imagers on interplanetary spacecraft, and the ASP can now use
any pair of ISIS camera models to build digital terrain models (DTMs).
Significant scalability improvements have also been made in order to support
extremely large imagery for the HiRISE and LROC imagers. Finally, in
preparation for extremely large data processing jobs, new enhancements enable
distributed operation of the Stereo Pipeline on NASA s Columbia supercomputer.
Digital Terrain Models (DTMs) at the meter-scale derived
from HiRISE and other high resolution imagers enable a variety of
scientific investigations that will lead to new discoveries about Mars
and the Moon. Depositional histories, crater morphology, erosion processes, and
other studies of planet morphology rely on 3D data, however the availability of
high resolution DTMs has historically been limited due the expense and manpower
needed to produce them. The ASP is now poised to provide scientists with DTM
generation software that can be run on their own workstations. The new, tighter
integration with ISIS will ease adoption of the software, since ISIS is a image
processing toolchain to which many scientists are accustomed.
The ASP is built on the NASA Vision Workbench, our open
source image processing toolkit. The core components of the stereo image
processing algorithms were released in December of 2007 as part of the Vision
Workbench. We track membership on the Vision Workbench project mailing list,
which now includes over 40 individuals at institutions including NASA Goddard,
NASA JPL, Google, and MIT/Woods Hole.
The first objective of this project in 2009 is to validate
the work done in 2008 via a comparison of HiRISE DTMs of the Candor Chasma
region of Mars. This study will be carried out in cooperation with the USGS.
Once the software has been validated, an open source release of the stereo
pipeline will be prepared for distribution to scientists at Arizona State
University, U. of Arizona, the USGS, and other institutions. Additional
improvements and refinements to the software will be ongoing.
James C Tilton
Subdue ing RHSEG...
James C. Tilton NASA GSFC
Diane J. Cook, Nikhil Ketkar Washington State University
Our project seeks to integrate a graph based knowledge
discovery system called Subdue with image segmentation hierarchies produced by
a hierarchical image segmentation approach called RHSEG. It is expected that
RHSEG segmentations will abstract the image pixel data into region objects from
which Subdue will be able to discover or identify meaningful patterns and
relationships.
A successful combination of RHSEG and Subdue will lead to
more effective data analysis, data mining and knowledge discovery for NASA
data. The product is not yet ready
to be used by others.
We will continue to experiment with interfaces between the
hierarchical segmentations produced by RHSEG and the Subdue system. Once this
interface is made workable, we will experiment with characterizing various
patterns in several types of remotely sensed image data.
Lutz Hamel
Exploration of Novel
Methods to Visualize Genome Evolution
http://bioinformatics.cs.uri.edu/gpx/
Lutz Hamel University of Rhode Island
J. Peter Gogarten University of Connecticut
We have adapted the use of bipartition spectra for
comparative genome analyses. We have applied self-organizing maps, an
artificial neural network approach to unsupervised learning, to genomic quintet
and bipartition data. We have constructed an online analysis tool that
automates some of the analysis steps.
Early life on Earth has left a variety of traces that can be
utilized to reconstruct the history of life, e.g., the fossil and geological
records, and information retained in living organisms. Our research focuses on
how information can be gained from the molecular record, i.e. information about
the history of life that is retained in structure and sequence of
macromolecules found in extant organisms. The interpretation of the molecular
record necessitates its calibration with respect to the geochemical and fossil
records, and needs to consider and incorporate information about biochemical
pathways and evolutionary theory. The analyses of the mosaic nature of genomes
using phylogenetics will be a key ingredient to unravel the life s early
history. Anonymous tracking of the
number of times our online tool is accessed.
Applying self-organizing maps (SOM) and locally linear
embedding (LLE) to larger groups of genomes. This will necessitate to
reformulate the way we represent the genome relationships, we need a more
robust format, in this case quartets rather than bipartition data.
Vinay Kashyap
Calibration
Uncertainty and Data Analysis
http://hea-www.harvard.edu/AstroStat/calerr/
Vinay Kashyap, Hyunsook Lee SAO/CXC
David van Dyk UC Irvine
Rima Izem Harvard
CHASC, et al. CXC/Eureka/Harvard
We have demonstrated that our scheme to incorporate
calibration uncertainties into data analysis via modification of MCMC in the
context of Chandra/ACIS-S effective area uncertainties. We have shown that it
does work and gives results consistent with brute force calculations at a
fraction of the computational cost. We are developing a robust standard that
can be used with any mission, for any instrument, and one that can be
generalized to different schemes of encoding of the uncertainties.
We deal directly with a major source of analysis error,
viz., the uncertainty in instrument calibration and how it affects inference.
This is therefore relevant to all NASA missions.
The code is not public yet, but we intend to make it public
in a year or so, and will make it available via the CHASC web site. We plan to further generalize our method
to account for uncertainties in spectral response and telescope point spread
functions. We also intend to publicise these results amongst astronomers so
that calibration scientists will create the relevant files and other
astronomers will use these products in their analyses. The eventual goal is to
have these methods be available generally in popular analysis software such as
Sherpa and XSPEC.
Ashit Talukder
Global cyclone
detection and tracking (GLYDER)
Data Analysis and Visualization
http://eis.jpl.nasa.gov/~atalukde/
Ashit Talukder, Shen-Shyang Ho, Timothy Liu, Bingham
Jet Propulsion Laboratory
Novel technology development and Capabilities:
1. New transfer learning mechanism to transfer information
between multiple satellite data for remote global cyclone detection and
tracking
2. Ensemble classifiers designed for robust classification
of cyclones from remote QuikSCAT data
3. Improved segmentation and feature extraction techniques
for cyclone region segmentation with minimal false alarms
4. Automated the procedure to pull QuikSCAT datasets for
specified date/time/swath on demand from remote databases (earlier approach
involved manual extraction that was tedious and time-consuming)
5. Prediction of cyclone evolution for constrained, faster
and more efficient cyclone tracking from noisy remote satellite data 6.
Integrated code for cyclone detection and false alarm rejection from QuikSCAT
single sensor data
Significant Science Results
1. Developed and demonstrated capability (offline) to track
hurricanes with a 3-hour temporal resolution by combining multiple satellite
measurements - this is unprecedented
2. Demonstrated GLYDER remote cyclone detection system on
entire 2005 calendar year
a) Demonstrated capability to
successfully detect all 25 documented tropical cyclones for Year 2005
b) Demonstrated detection of four
cyclones 72 hours before National Hurricane Center
c) One cyclone started as
extra-tropical cyclone and evolved later to tropical cyclone (detected first by
GLYDER, reported later by NHC)
d) Detected one unknown tropical
event not reported National Hurricane Center - under science evaluation
e) Numerous undocumented
extra-tropical cyclones in 2005 detected with GLYDER - demo of truly GLOBAL
cyclone detection
• Software IT Products to characterize global cyclone
variability
• Tools that empower climate scientists to study, quantify
spatiotemporal variability of cyclones on a truly global basis
• Tools for Data providers to tag cyclone metadata and
enable content-based searching • Evaluating direct use of GLYDER in
cyclogenesis, to better study the evolution of events before they become
cyclones
• Potential use of GLYDER in mission operational environment
for detecting and tracking global tropical and extra-tropical cyclones in
near-real time (using NRT QuikSCAT and NRT TRMM)
• Working closely with hurricane researchers at JPL/NASA to
evaluate science use of GLYDER in cyclogenesis
• Exploring the possibility of putting the GLYDER software
toolkit on the web (such as the JPL Hurricane Portal)
•Test our implementation over longer time scale in a region
that have multiple cyclone occurrences
•Include TRMM 2B25 swath data to construct a vertical
profile of reflectivity for cyclone detection
•Explore transfer learning for cyclone detection
• Explore Active learning strategies for improved classifier
design
•Explore potential of integration with near real-time data
streams
• Disseminate GLYDER to scientists working with the GLYDER
team
• Test toolkit for cyclogenesis and early evolution of
cyclones
• Integrate GLYDER software technology components to build a
single usable toolkit
• More comprehensively pursue science use of GLYDER by
climatologists and hurricane researchers.
Kevin H Knuth
Bayesian Source
Separation for Astrophysical Spectra
http://knuthlab.rit.albany.edu/ 2 3
Kevin H. Knuth, Deniz Gencaga University at Albany
Duane F. Carbon NASA Ames Research Center
We have developed the ON/OFF constituent model for Bayesian
source separation and applied it to the problem of detecting polycyclic
aromatic hydrocarbons (PAHs) in infrared emissions from star-forming regions.
This technology enables researchers to estimate the quantity of a substance
independent of an estimate of its presence. We have also developed models for
Planck blackbody spectra and implemented them within the nested sampling
framework, so that we can estimate the parameters for multiple simultaneous
blackbodies along the line-of-sight.
We are applying these source separation techniques to the
problem of characterizing polycyclic aromatic hydrocarbons (PAHs) in infrared
emissions from star-forming regions. The ON/OFF methodology enables us to
differentiate between the case where a substance is present, but in small
quantities, from the case where the substance is absent. This methodology will
find use in any source separation application where presence and quantity are
both to be estimated for data. We expect that this computational technology
will be of great use to Raman spectroscopy in Planetary missions.
At this stage, the product is not available for use by
others. In the future, the code will be made available on our website and the
AISRP Code Repository. We will log downloads from our website and ask for users
to volunteer data regarding the usage.
We have applied these techniques to synthetic spectra. We
plan to further refine the algorithm, and to begin testing on real datasets
from the Infrared Space Observatory and the Spitzer Space Telescope. We will
also be porting these algorithms to our Beowulf cluster to improve data
analysis speeds.
Olfa Nasroui
Mining Solar Loops to
Support Studies of the Coronal Heating Problem
Olfa Nasraoui University of Louisville
Joan Schmelz University of Memphis
Background: The
search for interesting images (with coronal loops) is by far the most time
consuming aspect of coronal heating studies. Prior to this project, this
process was performed manually, and was therefore extremely tedious, and
hinders the progress of science in this field. Our project aims to develop an
approach based on data mining to quickly sift through massive data sets
downloaded from the online NASA and ESA solar image databases and automatically
discover the rare but interesting images with solar loops, which are essential
in studies of the Coronal Heating Problem. The solar loop mining scheme will
rely on the following components: (i) Collection and labeling of a sample data
set of images coming from both categories (with and without solar loops), (ii)
An optimal feature selection strategy that will facilitate the retrieval task,
(iii) A classification strategy to classify the transformed image into the
correct class, and (iv) Appropriate measures to validate the effectiveness of
the loop mining process. This project is divided in several phases that target
the image databases collected by two different instruments, EIT aboard the
NASA/European Space Agency spacecraft, SOHO, and NASA s TRACE data sets.
Highlights: The
latest activities involved working on improving the detection rate for the
out-of-disk loops in EIT solar images as well developing good detectors for the
in-disk loops. Our greatest effort has been in addressing three difficult and
continuing challenges: (i) labeling, (ii) better feature construction, (iii)
handling imbalanced data sets. In (i) labeling,
we had to develop an interactive tool for label correction at the level of
individual blocks. This was necessary to correct some of the inconsistencies in
labeling by the experts, and to correct errors when the labels are propagated
to the blocks after automated block extraction. In (ii) better feature construction, we investigated more complex feature
construction techniques, mostly based on: developing curvature features for
imperfect and noisy edges, including spatial information, edge cleaning
techniques to apply before curvature construction, and using a newly developed
stream clustering algorithm to mine the optimal peak locations from huge and
noisy Hough accumulator arrays. Our clustering algorithm is successful, however
the biggest challenge has been in devising the most appropriate features for
loop detection, based on its outputs. In (iii) handling imbalanced data sets, we investigated over-sampling and
under-sampling techniques, compared to adaptive sampling machine learning
techniques such as boosting. In particular, we applied SMOTE-based oversampling,
but concluded that it only helped raise the recall metric slightly, but at an
unacceptable decrease in precision. Boosting on the other hand, yielded the
best results. In addition to the above, we have continued collecting more
labeled examples to improve our classification models, and ran extensive
cross-validation experiments to compare different feature constructions and
different classifiers, and to study the effect of solar cycle information on
the results.
Significance to NASA
Missions: The search for interesting
images for coronal temperature analysis (with coronal loops) amounts to
searching for a needle in a haystack, and therefore hinders the fast progress
of science in this field. The next generation EIT called MAGRITE, on NASA s Solar
Dynamics Observatory, will require state of the art techniques to sift through
the massive wealth of data to support scientific discoveries. Our work supports
the goals of the Applied Information Systems Research (AISR) program, since it
develops novel information technology and computational methods that promise to
increase productivity of the OSS research and public outreach endeavors, and
would benefit the state-of-practice in space science. Our work helps to
increase science and educational return from the data through advanced
knowledge discovery methodologies. Automatic detection of coronal loops can
help in understanding and predicting solar weather which has a significant
impact on space missions, satellites, and the Earth.
Software and
Publications: Our software is currently on the internal project
collaboration platform website:
``http://webmining.spd.louisville.edu/twiki/bin/view/SOLARLoops/ . This includes not only the full-fledged
solar loop mining system with all the expert labeling, training and validation
models, but also the tools for final testing of complete images (for both out
of disk and in-disk EIT data). We have several conference and journal
publications, including some that have been submitted, and still under review.
We also have a completed thesis (by Heba Elgazzar) that describes the
methods details and results from the
early phases of the project.
Upcoming Plans: We plan to continue work on improving our
feature construction and classification techniques, working on improving both
accuracy of the results and scalability to massive data sets. We also plan to
incorporate online learning that will fold new labeled examples into older
models instead of re-training our models from scratch. Our final phase that
targets TRACE data sets, has also just began, and will require not only
borrowing from some of the lessons learned and techniques developed for EIT
data, but also new methods that are specialized for this data set, due to its
distinguishing characteristics. As our cross-validation results reach a more
stable state, we will be more active in submitting publications.
Tomasz F. Stepinski
Automated
Identification and Characterization of Landforms on Mars
Tomasz Stepinski Lunar and Planetary Institute
The project has resulted in a software for an automatic
survay of impact craters on Mars and an automatic survay of valley networks on
Mars. A new catalog of Martian craters was constructed using the software.
Using the new catalog a global maps of craters depth/diameter ratio are
produced. These maps show a striking pattern that can be interpreted by an
existance of subsurface ice at high latidudes. NASA
is focused on studying Mars surface and subsurface. Craters are one of the most
important landform features on Mars and other planets. Cataloging them yields
wealth of geologic information. Developed software can be applied to the Moon
and Mercury once elevation data for these planets become available. The valley networks survey software has been
given to two other scientist who has asked for it. It has also been ported as
the Web service. The crater identification software is available at
cratermatic.sourceforge.net/ where it downloads can be tracked directly. In the last year we are planning to
finish the global catalog of craters and submit it to the Planetary GIS
Webservice. We also planning on finishing a global database of valley networks
and also submit it to the Planetary GIS Webservice.
Ralph D. Lorenz
Intelligent Sensor
Network Study of Dust Devils
Data Management
http://www.lpl.arizona.edu/~rlorenz
Ralph Lorenz JHU
Applied Physics Laboratory
So far (yr 1), demonstrated
potential for inexpensive data acquisition platforms to acquire
spatially-resolved data.
Dust devils are an aviation
hazard on Earth, and a factor in rover operations at Mars. They are an
important dust-raising mechanism on both planets, with implications for climate
and air quality.
Will refine approaches tested
last year, and explore potential of new technologies (specifically an array of
wireless sensor motes for real-time data acquisition and monitoring)
Martin D. Weinberg
Enabling Bayesian
inference for the astronomy masses
http://www.astro.umass.edu/~weinberg/bie
Martin D. Weinberg, Neal Katz, Houjun Mo, Elilot Moss
University of Massachusetts
1) Bayesian tool for inferring galaxy properties from images
shown to be robust even in low S/N regimes, low bias compared to currently used
approaches (e.g. GALFIT). Full posterior simulation allows hypothesis testing
for an ensemble of images.
2) Bayesian semi-analytic modeling tool has revealed
multiple modalities for fit to luminosity function. Current SAM incorporates
physics of many research groups, and will allow hypothesis testing.
3) Alpha-testing persistence subsystem allowing all states
of a Bayesian analysis (e.g. MCMC posterior, data, etc.) to be stored,
commented and reused for future investigations.
1) General efficient parallel platform for Bayesian
computation of astronomical data (any data really . . .).
2) Platform for multiple discipline investigator
collaborations to share infrastructure (e.g. astronomers and statisticians,
extragalactic and planetary research).
3) System allows script-based prototyping and a C++ library
for production suitable for mission data pipelines.
Currently by download recording. When the standalone
packages become available next year, we plan to have follow-up questionnaires
and a wiki in addition to registration and download recording. Core science applications underway in
three areas:
1) bulge to disk correlations with luminosity and
environment for 2MASS/SDSS selected input image catalog.
2) SAM modal analysis and inter-research-group hypothesis
testing underway.
3) LMC metallicity/population/structure star-count analyses
underway.
Papers planned for the upcoming year.
Kiri L. Wagstaff
Detecting Surface
Changes via Dynamic Landmarking
Kiri L. Wagstaff, Adnan Ansar Jet Propulsion Laboratory
Melissa Bunte, Ron Greeley Arizona State University
Norbert Schorghofer University of Hawaii
This project began in November, 2007. We have developed new
methods for automated landmark selection and for surface change detection. The
landmark selection methods are based on statistical measures of local terrain
salience and entropy. We have applied these methods to several Mars surface
images collected from orbit. Using annotations provided by our science
collaborators (dark slope streaks and dust devil tracks), we are able to
quantify the salience of different types of surface features when used as
landmarks. In parallel, we have developed advances in conventional change
detection methods based on pixel registration. Ultimately, we plan to combine
these techniques together to yield highly efficient ways to identify, and
classify, surface changes.
Our landmark selection and change detection methods provide
techniques that can be useful both onboard a remote spacecraft and in ground
processing on the Earth. In an onboard setting, salient landmarks can be
detected and catalogued as they are observed, providing a highly compressed
summary of the region under study. On the ground, more intensive processing can
be used to register images collected at different times and then identify
changes. Further, we plan to apply additional processing to detected landmarks
to classify them into known categories (e.g., crater, ridge, gully, dune, dust
devil track, dark slope streak) as well as to flag any unusual landmark types
that may not previously have been identified. This capability will permit us to
automatically generate semantic annotations for the large body of archived
images, ultimately supporting a content-based search facility that can permit
scientists as well as the general public to quickly find images that contain a
specific feature of interest.
We do not yet have a product that is available for use by
others, but this will change in the coming years of this project.
First, we will conduct a more extensive evaluation of our
landmark selection and change detection methods on a large body of MOC and
THEMIS images. We will develop hybrid methods that leverage the strengths of
landmark-based change detection (obtaining salient regions instead of
individual pixels) and pixel-based change detection (very high detection rates
even for areas that may not be salient). Next, we will train a landmark
classifier to label each selected landmark with its general type. We will also
develop a representation for Regional Landmark Graphs (RLGs) to characterize
the landmarks structure of a given region, and to enable recognition of the
same region when it is observed in subsequent images.
Timothy Newman
Auroral Phenomenon
Localization, Classification, and Temporal Evolution Tracking Methods
T. Newman UAHuntsville Computer Science
G. Germany UAHuntsville CSPAR
J. Spann NASA MSFC / NSSTC
C. C. Hung Southern Polytechnic St. Univ.
New methodologies for localizing aurorae and auroral
phenomena in NASA Polar UVI and IMAGE FUV imagery are described. The methods
are largely driven by exploitation of known auroral shape constraints. The
methods are primarily aimed at enabling web-based retrieval of images of
interest from two large image archives. Preliminary benchmarkings of the
methods effectiveness are also reported.
Our work enables increased exploitation of NASA mission
data. The work is aimed at directly aiding space science, particularly the
study of Sun-Earth systems.
Site hits (counts and region-of-origins) on the legacy web
tool are monitored. When the new tool is activated, we plan to use the same
scheme. The methods developed for
UVI are currently being employed in the upgrade of the capabilities of an
existing web-based search interface. The methods developed for FUV will be used
to build a new web-based tool for retrieval of FUV imagery.
Adnan Ansar
Multi-modal image
registration and mapping
Adnan Ansar, Larry Matthies JPL
We have developed a prototype multi-modal image matcher
capable of automatically matching images across such sensor modalities as
visible, thermal and radar. In this context, imagery refers to the data product
of any sensor capable of generating a 2d pixel representation. Preliminary
tests on Cassini, Mars and terrestrial datasets are very promising
The primary driver for this work is to enable localization
of a Titan aerial vehicle by matching mosaicked aerial data to orbital sensor
data. This enables a critically needed high-precision localization capability
that is unlikely to be achieved in any other way. Our work applies equally well
to any mission, such as small body proximity operations, requiring spacecraft
localization from low orbit using high orbital reference imagery. In addition
to the implications for navigation, the matching result itself is relevant for
landing site selection for future Mars and other lander missions, where image
overlays from multiple sensors are currently generated by hand. Similarly,
there is a potential for automatic matching of ground-based mosaics to orbital
imagery for use in rover localization. Finally, a pixel level automatic
registration of data across sensor modalities will be of considerable benefit
in both space and terrestrial science applications.
Our product is in the early development stage. However, we
have made contacts with and obtained data from Cassini mission scientists, the
Phoenix landing team and terrestrial / atmospheric scientists. We have also
discussed our work with and received guidance from the Titan outer planets
mission study team. We will feed results back to these groups as we proceed and
take guidance from them in developing our tools.
Having demonstrated proof of concept, we are now working on
optimization and improving robustness. Among the areas under development for
this year are incorporation of scale / orientation invariance and image warping
to accommodate inconsistent ortho-rectification of data products. We will also
be studying in detail the feasibility of transitioning this work to an FPGA
implementation for use in flight. Future efforts will include work on structure
from motion to enable ortho-rectification and mosaicking in the presence of 3D
relief, map building, and integration of our methods into a full estimation
framework.
Jay R. Johnson
Higher-Order
Statistical Method for Geospace Data
Jay Johnson Princeton University
Simon Wing The Johns Hopkins University
Development of method for extracting nonlinear dependencies
in geospace data sets using mutual information, cumulants and transfer entropy.
Have identified solar cycle nonlinearity in space data associated with high
speed solar wind streams. Examined issues of causality and information flow in
space systems. Methods may be used
to identify nonlinearity in geospace data sets and access an information
horizon for data flow. Identification of nonlinearity is useful for modeling
efforts. Determination of causality and information flow is important to build
predictive models (for example of hazardous space weather conditions) and to
determine the extent to which predictive models can be improved. Contacts and citations. Plan to upload
product and monitor access of webpage. Investigate
coupling function for solar wind-magnetosphere interactions. Examine
dimensionality of space data and compression of data stream through
dimensional. Investigate causality using entropy based measures with
applications to magnetospheric processes for which the cause and effect remain
unclear.
Gabor Toth
Development of an
Adaptive MHD Simulation Tool
Gabor Toth, Tamas
Gombosi , Darren De Zeeuw, Kenneth
Powell
CSEM, University of
Michigan
Over the last 15 years our group at the University of
Michigan has been developing a general use global MHD code, BATS-R-US, and the
Space Weather Modeling Framework (SWMF) that couples domain models extending
from the Sun to planetary upper atmospheres and ionospheres. BATS-R-US and SWMF
have been extensively used to simulate a broad range of space science
phenomena. Still there are many unmet challenges. There is a need to go beyond
ideal MHD. In the previous AISRP we have developed a new parallel implicit Hall
MHD solver for the 3D block-adaptive grid used in BATS-R-US. In the current
AISR project we went further and developed a two-fluid MHD model with a
separate electron pressure equation and we have developed a general multi-ion
MHD model. These models are fully implemented into BATS-R-US, but they still
require improvements. We have already used the new multi-ion solver to simulate
the magnetosphere of Earth with ion sources coming from the polar wind. The
polar wind code PWOM is coupled to BATS-R-US through the SWMF framework. We are
going to develop an MHD model with a non-isotropic pressure. We will also
implement a new time stepping scheme into BATS-R-US that allows the various
subdomains to evolve at the rate allowed by the local stability conditions.
Having a more accurate physics model and a more efficient
solver allows us to do a better job in modeling the solar corona, the
magnetosphere and the space environment in general. These technologies are
relevant to the following NASA goals: Understand the Sun and its effects on
Earth and the solar system. Advance scientific knowledge of the origin and
history of the solar system, the potential for life elsewhere, and the hazards
and resources present as humans explore space. The
SWMF and BATS-R-US are freely available via registration at the CSEM web site.
The codes are available for runs on request at the Community Coordinated Modeling
Center (CCMC). CCMC maintains a database of all runs and also requires
notification about publications using CCMC runs. Our codes are also used
extensively by the researchers and students at CSEM. Further improve the multi-ion MHD model. Further develop and
apply the two-fluid MHD model. Develop and implement the non-isotropic pressure
model. Designing the algorithm for the local time stepping.
Kanna Rajan
Adaptive Control for
Underwater Vehicles
http://www.mbari.org/autonomy/TREX/index.htm
Kanna Rajan MBARI
We have developed and deployed an onboard Adaptive Control
System that integrates Planning and Probabilistic State Estimation in a hybrid
Executive. The resulting system is a unified representational and computational
framework based on declarative models and constraint-based temporal plans. The
work is motivated by the need to explore the oceans more cost-effectively
through the use of Autonomous Underwater Vehicles (AUV), requiring them to be
goal-directed, perceptive, adaptive and robust in the context of dynamic and
uncertain conditions. The novelty of our approach is in integrating
deliberation and reaction over different temporal and functional scopes within
a single model, and in breaking new ground in oceanography by allowing for
precise sampling within a feature of interest using an autonomous robot. The
system is general-purpose and adaptable to other ocean going and terrestrial
platforms.
Onboard autonomy is critical to NASA s needs for unmanned
deep space exploration. As light time travel increases, spacecraft cannot be
joy-sticked from Earth and will require onboard autonomy to ensure their health
and safety as well as critical mission needs are satisfied. The tools and
techniques we have developed are based on a decades worth of experience at NASA
Ames and flown on NASA s DS1 spacecraft and used on the ground for command and
control of the two MER rovers at JPL. We expect further use of such techniques
as the agency moves towards Lunar and Martian exploration.
Informal collaborations and working within a privately
funded non-profit organization like MBARI, allows us to give away our code. To
date the extensive code-base is being used by LAAS, Toulouse France (in
collaboration with IFREMER, the French national Oceanographic agency), National
Institute of Oceanography, Goa, India and Willow Garage a not-for-profit
Silicon Valley startup in Robotics.
We work closely with MBARI scientists in Microbiology,
Biological and Chemical Oceanography in an inter-disciplinary environment which
is science driven. Our near term commitments are the following:
1. to characterize blooms and ocean Fronts in 4D (space and
time) by sampling and observation using Autonomous Underwater Vehicles (AUVs).
2. to enable mixed-initiative control of AUVs using shore to
vehicle communication and robust onboard commanding to ensure the platforms
meet evolving needs.
3. enable persistent long-duration mission operations of
AUVs by encapsulating fault conditions leading to replan scenarios
automatically and onboard.
Hillol Kargupta
Distributed and
Peer-to-Peer Data Mining for Scalable Analysis of Data from Virtual
Observatories
www.cs.umbc.edu/~hillol/Kargupta/ddmVO/
Hillol Kargupta, Agnik Kamalika Das, Kanishka Bhaduri,
Wesley Griffin UMBC
Kirk Brone George Mason University
Chris Giannella Loyola College
1. Explored the problem of fundamental plane analysis from
distributed astronomy data available from different virtual observatories.
Performed a series scientific data analysis tasks in order to better understand
the problem.
2. Developed communication efficient distributed and
peer-to-peer data mining algorithms for large distributed environments. This project will develop technology
for mining high-througput distributed data Astronomy data sources. The goal is
to develop a collection of scalable local algorithms that will be able to
quickly analyze distributed data repositories and streams in communication
efficient manner. We are currently exploring fundamental plane computation and
outlier detection techniques for distributed data mining. These techniques
should allow Astronomy researchers to identify unusual phenomena and objects of
interest quickly from large volume of distributed data. Research publications are the current products of this
research. The impact of these papers can be tracked using citation indices. We plan to explore several distributed data
mining algorithms. We are also collaborating with Astronomers regarding
practical deployment of these algorithms for solving several science problems. This
involves exploring detailed scientific explorations of the virtual observatory
data. The immediate plan is to identify a collection of scientific tasks that
are likely to be used by Astronomers and develop efficient techniques for
performing those using distributed data mining techniques.
Pasquale Tricarico
Orbit@Home
Computational Methods
P. Tricarico Planetary Science Institute
During the first two months of this project, we have reached
the point where a our first application is available for all major platforms,
and is being debugged with the help of thousands of volunteers from around the
world. Its purpose is to realistically simulate the performance of Near Earth
Asteroids (NEAs) surveys. Once into production, this is expected to help
enhance NEAs discovery rate.
Distributed computing (DC) provides a virtually unlimited
amount of computational power. With orbit@home we try to demonstrate that
moving toward DC is both easy and necessary in order to enable qualitatively
new science, in the form of very large scale numerical simulations or data
analysis. All volunteers are
automatically tracked using a client/server mechanism. We plan to keep all our
products freely accessible. Feedback and network statistics can provide an
estimate of the users base.
We will produce extensive simulations of NEAs surveys, and
then apply the same approach to the real NEAs population to try to improve the
discovery rate.
Tamas I. Gombosi
Multifluid MHD
Simulations of the Magnetosphere
Tamas Gombos, Gabor Toth University
of Michigan, Alex Glocer
University of Michigan
There are several critical research areas where single-fluid
treatment is not an adequate representation of the basic physics. These areas
include space weather, where the ionosphere is a major source of heavy ions
that contribute to the formation of the ring current and thus to magnetospheric
dynamics. Single-fluid description is clearly inadequate to describe many space
weather phenomena. We developed and implement a multifluid MHD model that is
part of a flexible, high-performance, robust, accurate and widely available simulation
tool.
The proposed research is relevant for the following NASA
strategic goals and outcomes:
Strategic Goal 3.2:
Understand the Sun and its effects on Earth and the solar system. NASA Science
Outcomes:
Progress in
understanding the fundamental physical processes of the space environment from
the Sun to Earth, to other planets, and beyond to the interstellar medium.
Progress in
understanding how human society, technological systems, and the habitability of
planets are affected by solar variability and planetary magnetic fields.
Progress in
developing the capability to predict the extreme and dynamic conditions in
space in order to maximize the safety and productivity of human and robotic
explorers.
Strategic Goal 3.3:
Advance scientific knowledge of the origin and history of the solar system, the
potential for life elsewhere, and the hazards and resources present as humans
explore space.
At this stage the new features are tested and they are not
used by others. However, our Space Weather Modeling Framework is widely used by
the space physics community and it is available at the CCMC. We will develop a two-fluid (electron-ion)
Hall MHD capability to simulate CMEs and planetary magnetospheres.
Peter J. MacNeice
Magnetogram Synthesis
Peter MacNeice NASA/GSFC
Joel Allred, Kevin Olson Drexel University
We have only begun and the first funding increment has not
yet arrived at the time of this submission. We have developed reading and
reformating routines for almost all of the primary magnetogram sources and have
implemented the first simple processing widgets for our initial GUI design.
This work is highly relevant to NASA s Space Weather
monitoring and prediction capabilities, and to its solar and heliospheric
research activities. Not yet required. We will extend our development of the
lightweight processing layer and begin studying the most useful algorithms for
minimizing spurious current densities in magnetograms synthesized from multiple
sources.
Kevin Olson
PARAMESH: A parallel,
adaptive, grid tool for the Space Sciences
http://www.physics.drexel.edu/~olson/paramesh-doc/
Kevin Olson Drexel University
C interface Parallel IO Divergence of B control Improved
performance PARAMESH has been integrated into major scientific codes: FLASH
astrophysics code at the University of Chicago and HAHNDOL general relativity
code at NASA/GSFC. Science Projects: 1) Large Scale general relativity
simulations of colliding black holes performed at NASA/AMES under the direction
of Joan Centrella. 2) Large scale simulations of Type Ia supernovae (University
of Chicago FLASH code). Black
hole simulations are directly relevant to the LISA project. Informally via e-mail. Future plans for PARAMESH are tentative. Currently
funding is being sought for further development and support of PARAMESH.
John C. Houck
HYDRA: A New Paradigm
for Astrophysical Modeling, Simulation, and Analysis
John C. Houck , Dan
Dewey, Michael S. Noble, Michael A. Nowak, John E. Davis
MIT
During the past year we have developed software to do
forward folding and comparison of 3D source models with 2D event-based data
sets. These routines build on the volumetric 3D routines developed during our
first year of funding. We have continued working to define a software interface
that can be used to handle more general optimization problems. We have also
continued to explore ways to use parallel computation to speed up various data
analysis tasks. These
technologies support more thorough science analysis of observational data
obtained by NASA spacecraft such as the Chandra X-ray Observatory and also from
ground-based instrumentation. - In adding new capabilities we plan to
focus on ways to make better use of parallel processing.
Andrew Ptak
On-the-fly and Grid
Analysis of Astronomical Images for the Virtual Observatory http://www.xassist.org
Andrew Ptak Johns Hopkins University
Andrew Connolly, Simon Krughoff University
of Washington
Web services for requesting the processing of X-ray data,
retrieving X-ray analysis results, and analyzing more than one optical image
simultaneously (i.e., with one for source detection and another for
photometry). Optical source detection includes automatic correlation with
request astronomical tables. The joint X-ray/optical system has not been made
public yet but will be shortly.
These technologies will lead to better usage of NASA
archival astronomical data, most notably from HST and X-ray missions. Once the services become public and
registered we will track their usage from the access logs.
Publishing the web services, more advanced web services for
X-ray analysis (most notably computing upper limits, which can be critical for
including X-ray data in virtual observatory queries), and formalizing our
approach for combining similar web services for different data (optical and
X-ray) that include wavelength-specific features (e.g., X-ray data include
spectral information rather than just images).
Kenneth J. Mighell
Parallel-Processing
Astrophysical Image-Analysis Tools http://www.noao.edu/staff/mighell/aisr
Kenneth J. Mighell National
Optical Astronomy Observatory
During the past year, the PI has worked closely with Bill
Hoffmann and Bill Glaccum, both members of the Spitzer Space Telescope s
Infrared Array Camera (IRAC) Instrument Team, to demonstrate that his
AISR-funded MATPHOT algorithm for precision stellar photometry and astrometry
yields a significant improvement in photometric precision of IRAC Ch1 stellar
observations over the best results obtained with aperture photometry using the
recommended calibration procedures in the IRAC Data Handbook: the relative
peak-to-peak spread was reduced by a factor of 1.9 from 3.3% to 1.7% and the
relative robust standard deviation decreased by a factor of 1.7 from 0.92% to
0.54%. The PI has developed a new fast parallel-processing image analysis
program called CRBLASTER which does cosmic ray rejection in space-based
astrophysical observations using van Dokkum s L.A.Cosmic algorithm. Processing
a single 800x800 Hubble Space Telescope Wide-Field Planetary Camera 2 (WFPC2)
image takes 1.87 seconds using 4 processors on a 3.0 GHz Apple Xserve; the
efficiency of the program running with the 4 cores is 82%. The code has been
designed to be used as a software framework for the easy development of
parallel-processing image-anlaysis programs using embarrassing parallel
algorithms. The goal of NASA s New Millennium Program Space Technology 8
Dependable Multiprocessor (DM) project is to conduct a comprehensive research
project to investigate and develop for NASA the first supercomputer in space.
The PI, working with the DM Project Team, has recently ported two AISR-funded
parallel-processing applications, QLWFPC2 and CRBLASTER, to the DM-sigma
software-testbed cluster and is now adding fault tolerant features to these
applications so that they should be able to pass fault-injection/radiation
tests which are part of the DM project s Technology Readiness Level 6
Validation effort which should start in June 2008. These technologies will enhance the science return not only
of existing Spitzer Space Telescope IRAC Ch1 and Ch2 observations in the
Spitzer data archive but also those that will be made during during the
possible Spitzer Warm Mission which would start around April 2009 after all of
the cryogen is depleted. Helping the NMP ST-8 Dependable Multiprocessor team
validate NASA s first space-based supercomputer in a prototype demonstration in
a relevant environment demonstrate to NASA s Technology Review Board that the
fault-tolerant middleware techniques used in the DM project are useable by
researchers with parallel-processing scientific analysis applications which
might be suitable for use on future NASA astrophysical missions to be launched
in the next decade.
While this project has no formal tracking mechanism, the PI
is working closely with NASA s Spitzer Space Telescope s Infrared Array Camera
(IRAC) Instrument Team and NASA s New Millenium Program (NMP) Space Technology
8 (ST-8) Dependable Multiprocessor (DM) Project Team.
The PI will continue working with the IRAC Instrument Team
with the goal of developing new calibration and analysis procedures that have
the potential of significantly improving the precision of point-source
photometry. As part of this effort, the PI will customize the MATPHOT code to
do crowded-field photometry in IRAC Ch1 images. The PI will help the DM Project
Team in their TRL-6 Validation efforts. The PI will further develop the
CRBLASTER code and write a paper describing it for submission to the journal
Publications of the Astronomical Society of the Pacific. On June 28, 2008, the
PI will give the oral presentation CRBLASTER: A Fast Parallel-Processing
Program for Cosmic Ray Rejection at the SPIE-Marseille conference on Advanced
Software and Control for Astronomy. On June 23, 2008, the PI, Glaccum, and
Hoffmann will present the poster Improving the Photometric Precision of IRAC
Channel 1 at the SPIE-Marseille conference on Space Telescopes and
Instrumentation I: Optical, Infrared, and Millimeter.
James
Schombert University of Oregon
Hacking for Science
This
project is to develop automatic scripts to monitor/retrieve data from Data
Center websites. The resulting tools have an impact across a variety of science
fields as the use of websites to disseminate data is prevalent. These tools will allow the general user access
to new levels of data analysis.
R. Daniel Bergeron
Visualization of
multiresolution time series data
R. Daniel Bergeron, Andrew Foulks University of
New Hampshire
The principal goal for the previous year was to integrate a
basic multiresolution data support framework into the Visit environment from
the Lawrence Livermore National Laboratory. Visit provides a convenient
interactive environment and a wide range of existing visualization tools. By
integrating our multiresolution data management into Visit, users immediately
gain access to a large existing visualization code base. The software includes
3 components: STARgen, STARgui, and STARvisit plugins. STARgen is a tool for
generating multiresolution data hierarchies of time series data in which each
node in the hierarchy can represent reduction in either spatial resolution or
temporal resolution. STARgui is a friendly interactive program that guides a
user through the process of defining the desired multiresolution hierarchy. The
STARvisit data plugin provides the interface between the multiresolution data
hierarchy and the VisIt system. A user can change data resolutions via a simple
interactive dialog which triggers VisIt to reload data from the appropriate
resolution level. This tool allows a scientist to utilize Visit functionality
for rendering and interactive browsing of data in concert with data resolution
changes. All three components are available for download from the project web
site.
One of the major challenges facing NASA scientists is
effective analysis of the explosion of data that is being generated by current
science technology. This is particularly a problem in the simulation of
time-dependent magnetohydrodynamics (mhd) phenomena that produces enormous data
sets of many hundreds of gigabytes. It is also the case that interactive
visualization of the data is still one of the most effective techniques for
gaining insight into the physical phenomenon. The huge size of the data,
however, severely limits the ability to visualize the data in an effective
interactive environment. Our multiresolution data management framework makes it
easy to develop interactive environments for browsing such data. We have just made a version of the
software available through web download. We do not currently require a
registration in order to download.
The principal goals for the coming year include software
support for adaptive resolution data and better error data generation and
utilization. Our current tools allow a user to create and access data at different
resolutions, but each resolution level uses the same resolution throughout the
spatial and temporal domains. A more flexible data representation scheme allows
a user to define a single data set in such a way that it is represented by data
at different resolutions in different regions -- regions where the data values
are not changing very rapidly can be represented at a lower resolution than
regions in which the data is changing dramatically. Effective decisions about
the resolution level require an effective model for representing the error that
is introduced by using the lower resolution data. We have completed a basic
utility for representing such error, but that tool has to be integrated into
the data generation tool. We have also refined our vision about the kinds of
error that should be available to a scientist as part of the interactive
visualization environment. We will incorporate this new vision of error into
future releases of our software.
Jeffrey Scargle NASA
Ames Research Center
Novel Methods for Analysis of Photon-Limited Data
I have outlined the plan and scope for a Handbook of
Statistics for Event Data, meant to be a useful for scientists analyzing event
or point data. Details will be given for two case studies, one in gamma-ray
astronomy (and targeted at the Gamma Ray Large Area Space Telescope) and the
other in x-ray astronomy:
(a) detection of dispersion (energy
dependent lags) in time- and energy-tagged gamma-ray burst data;
(b) characterization of the
variability of active galactic nuclei from time-tagged x-ray data: the shortest
detectable variability scale for Mkn 421 and NGC 4151.
These studies are meant to exemplify both the general
approach and specific algorithms which will be part of the Handbook.
The Handbook will be of practical use for analysis of a
large fraction of all of the data obtained by NASA space or earth science
missions. (GLAST, Swift, Compton GRO, RXTE and Chandra are specific cases
targeted.)
The Handbook has not been released, even in draft form. For
similar projects in the past, I have tracked the usage in the relatively small
high-energy astrophysics community by personal communication and literature
survey, and in the broader community by web-based searches of the relevant
institutions and archives.
A draft of the complete Handbook will be prepared (with
algorithms presented in MatLab, a high-level data language) and sent to a
number of colleagues for comments and suggestions. Applications to gamma-ray
and x-ray observations will be published in method papers that detail the
numerical methods along side the scientific results (see #1 above).
Robin D. Morris
Event Analysis for
GLAST
Robin D Morris USRA-RIACS
Johann Cohen-Tanugi SLAC
Extended the methodology previously developed to analyse
charged particles incident on the LAT instrument to also analyse incident
gamma-ray photons. Demonstrated a statistical methodology in particle physics
data analysis problems that may help resolve a controversy in the literature
regarding the application of Bayesian statistical methods. An EPO video for
broadcast on PBS in the San Francisco Bay Area (and potentially nationwide) is
in an advanced state.
GLAST is a major NASA mission, with launch expected later
this year. The methodology developed is under consideration by two potential
SMEX missions.
The code is publicly available on the SLAC CVS server, but
we are currently unable to track the downloads by others.
Completion of the gamma-ray analysis code, including
development of the theory and code for model selection to determine the
relative probabilities of the different ways of interpreting the pattern of the
detector response in terms of the physical processes that occured during the
event. Submission of papers describing the methodology and results.
Michael C. Burl
Directed Exploration
of Complex Systems
Michael C. Burl JPL
Brian Enke, William J. Merline Southwest Research Institute
Numerical simulations provide scientists with a valuable
tool for examining massivley complex systems. However, in many simulations long
run-times make a detailed, exhaustive study of the input parameter landscape
infeasible. The key idea we have developed uses support vector machine (SVM)
classification techniques and active learning to cleverly explore the input
parameter space of a simulation. We have successfully applied this approach to
a complex smooth particle hydrodynamic (SPH)/N-body simulation of asteroid
collisions to narrow down plausible initial conditions (impactor size,
velocity, etc.) for generation of an Emma-like asteroid family. The technique
provides a significant reduction (2-fold to 10-fold or more in some cases) in
the time required to explore a simulation; this savings can be parlayed in
different ways depending upon the specific goals of the scientific
investigation: (i) the same final result can be obtained with fewer simulation
trials, (ii) more simulation trials can be conducted in a given amount of time,
(iii) higher-fidelity (e.g., finer spatio-temporal resolution) simulation
trials can be used, or (iv) more trials can be concentrated at the boundary
between interesting and not-interesting regions of input space. In addition to
publication in the scientific literature (e.g. Icarus), this work has been
described in a leading computer science conference (SIAM Data Mining
Conference), NASA Tech Briefs, and a JPL New Technology Report.
Simulations are widely used at NASA and other government
agencies for modeling and examining processes that could not be studied
otherwise. The techniques we are developing are broadly applicable to a variety
of physical systems; however, our current focus has been on the asteroid
collision application which specifically aids in the understanding of how the
Sun s family of planets and minor bodies originated and evolved. There are also
strong connections with ground observations and NASA spacecraft missions. Co-I
Merline is one of the leaders in finding binary asteroid pairs using adaptive
optics. He is also part of a search for small bodies near Mercury in the
currently-flying MESSENGER mission. In a previous seed investigation,
significant speedup in magnetospheric modeling based on data from NASA s IMAGE
spacecraft was obtained.
The algorithms are still in development mode, so currently
the product is only used internally.
The key goal for this year is to mature the software so that
it can easily be used with a variety of simulations. A prototype Bayesian
procedure for using an ensemble of multiple SVMs with different hyperparameters
to model the current state of knowledge will be incorporated. In addition, we
are evaluating new techniques, e.g., based on Gaussian processes, Markov Chain
Monte Carlo (MCMC) sampling, and an information gain criteria, to determine the
strengths and weaknesses of these approaches for different problems.
Alan Sussman
Robust Grid Computing using Peer-to-Peer Services
http://www.cs.umd.edu/projects/hpsl/chaos/Research
Alan Sussman, Pete Keleher, Bobby Bhattacharjee, Derek Richardson,
Dennis Wellnitz
University of Maryland
Work in the second year of the project has concentrated on
algorithms and simulations to create a fully functional peer that both performs
matchmaking well, and scales to large numbers of machines with no serious load
imbalances. Such load imbalances can arise from both problems in the initial
matchmaking process that assigns jobs to machines, and from maintaining the
structured peer-to-peer (P2P) network. Much effort this year has also gone into
the implementation of a usable software system. Based on the results of
simulations of our initial algorithms from year 1, and the new ones from this
year, we have built our first peer implementation. Initial versions of the peer
software have been thoroughly tested, and a large scale evaluation with the
astronomy collaborators on the project is under way. The evaluation involves
running the peer software on multiple clusters and desktop machines in both the
Computer Science and Astronomy departments at Maryland, on a total of well over
100 machines, and running thousands of astronomy simulations (still being
determined by the astronomers) through the complete system. We intend to
analyze many aspects of the system behavior through extensive logging of the
peer behavior during the experiment, including overall scalability and how well
our system simulations match the behavior of the real system.
The system will be used to easily share computational
resources among colleagues at multiple sites, typically within a scientific
discipline. Those resources can be used for large numbers of simulations, as
the astronomy co-Is are doing, for data analysis, or for any other purpose
requiring large numbers of independent computational jobs.
Once the software has been thoroughly tested by the
astronomy co-Is, we will make the software available on the project web site
for others to use. We will log downloads, and keep track of bug fix requests
from users for furthre support of the software.
We are continuing to develop the algorithms for matching
resource requests to available resources, in particular concentrating on
multi-core and multi-processor grid nodes. Effectively utilizing such
resources, which are becoming more widespread and important, is an open
question that has not been addressed at all in the desktop grid community to
date. Another area of ongoing research is in techniques for dynamic load
balancing. Our current algorithms and implementation perform matchmaking once,
when a job is submitted into a grid. That matchmaking is done using the current
(approximate) state of the overall set of peers that exists at that time. We
will investigate methods for determining when to initiate dynamic load
balancing algorithms, to adapt to the ever changing grid environment. The
algorithms for dealing with categorical resource types are being implemented in
the peer software, and will be thoroughly tested and evaluated in the upcoming
year. We are in the process of fully testing and deploying the software within
the project, both in computer science and to the astronomy collaborators. Large
scale testing of the peer in a distributed grid system is under way, and will
consume the majority of the project resources in the coming, final year of the
project. We will evaluate the reliability and scalability of the algorithms and
implementation under real workloads, first from within the project with
astronomy applications run by the astronomy co-Is on the project, and then in a
wider deployment within the wider computational science community at Maryland
(through the Institute for Advanced Computer Studies - UMIACS), and then to
collaborators outside Maryland (especially PI Sussman s space science
collaborators from the space weather modeling community).
Martin W. Lo JPL
MTool Data Analysis and Visualization
1. Application of dynamical systems theory to atmospheric
and ocean data.
2. Application of new 3D reconstruction techniques to
visualize complex data. Using computational differential geometric methods to
work with time varying 3D data.
3. These methods allow the detection and extraction of
spatial temporal coherent structures from the data. Eventually, it will allow
for the computation of flux and diffusion rates to characterize transport in
the atmosphere and ocean from satellite and in situ observations.
1. These techniques will help NASA scientists and engineers
analyze climate data and planetary science data.
2. The same techniques are applicable to the data analysis
of other SMD missions such as those studying solar physics, interstellar media,
wherever there are fluid problems.
3. These techniques are also useful for visualization and
for 3D reconstruction from images and point cloud data.
I plan to deliver MTool 1.0 at the end of FY08. It will be
delivered to several JPL users: the AIRS Instrument Project (data processing),
Titan GCM Project (visualization). I plan to continue working with these users
and to meet with them periodically to track the usage of MTool 1.0. This is a 1-year project. I intend to
submit a new proposal to AISR and other funding opportunities based on the work
accomplished this year.
Edward A. Belbruno Innovative
Orbital Design & Princeton University
Mission Extension
Using Sensitive Trajectories and Autonomous Control
Substantial progress has been obtained on precise
visualization and computation of the weak stability region. This has enabled an
understanding of new types of motions and a step towards understanding how
autonomous use of this region could someday be achieved.
Understanding the WSB region gives a way to understand new
types of low energy transfers and ways to stay in orbit about planets for much
longer periods of time than is traditional. This would extend time for data
mining. Also, fuel saving capability can allow more science to be taken on by
the spacecraft due to utilization of the fuel s mass. This would enhance the
science return.
This is tracked, in part, by new missions NASA is planning
that make direct use of the WSB region and low energy transfers. Currently,
three NASA missions are being planned to use this technology: GRAIL, LADEE,
THEMIS. Science is being enhanced and enabled in all of them by the use of the
WSB region. This is a substantial development. Also, publications and lectures
on this topic by other people is another measure. Use of this region by foreign
space programs, eg ESA is planning to use this methodology in the BepiColombo mission.
1. Hopefully, a nearly complete understanding of the WSB
region in a key situation.
2. New Applications and ways to use low energy transfers
resulting from the theory behind this technology.
3. Trying to give a realistic way to achieve autonomous control
of a spacecraft by making use of the low energy properties of motion in the
WSB.
Tamal Bose
Adaptive Algorithms
for Optimal Classification
Tamal Bose Virginia Tech
Erzsebet Merenyi Rice University
1. Unsupervised clustering: We did benchmark clustering on
our uncompressed hyperspectral test images, and are performing the same on
compressed-decompressed images.
2. We developed the mathematical algorithms for optimizing
the compression algorithms: (a) adaptive filters, (b) predictor models, (c)
adaptive quantizers, and (d) transform coders.
3. We implemented an Adaptive Differential Pulse Code
Modulation (ADPCM) based hyperspectral image coder. The coefficients of the
predictor filter are adapted based on classification metrics.
4. Supervised classification of compressed and decompressed
images yielded excellent results: Classification accuracy of these images
remains within 2% of the classification accuracy on original uncompressed data.
Space science missions that carry spectral imagers can
significantly benefit from the results of this project in scenarios where lossy
data compression is needed at compression ratios as high as can be achieved
without loss of relevant spectral information. Because of our approach to
compression, we will be able to determine the optimal compression ratio in an
adaptive manner. The fast parallel implementation of the compression and
classification algorithms makes the combined system an intelligent, real-time
on-board data understanding, compression and classification machine that learns
on chip and adapts continuously to new circumstances as desired, modifying the
compression scheme to best suit a given environment.
We will create a demo of all our algorithms and post them on
the project web site. All of our compression algorithms, codes, data analysis,
and images will be available on our project web site for interested users. The
number of downloads and hits will be tracked.
1) We will perform classification of hyperspectral images in
the transform and quantized domain. This will include experiments with a
variety of transforms, compression ratios and quantizers.
2) There are two extreme cases of the availability of
training data for classification. One, where there is sufficient data is
available, and the other when we have no labeled training data. There are many
scenarios in between where some knowledge is available on classification. We
will implement and evaluate our algorithms for a representative number of these
cases and accordingly modify our algorithms.
Aravind R. Dasu
SATH: a simulated
annealing to hardware compiler
http://www.usu.edu/rcg/index.php
Dr. Aravind Dasu, Jonathan Phillips Utah State
University
We have designed and developed an electronic system level
design automation tool/compiler called SATH. This is a C to FPGA compiler
specifically customized to auto-design hardware accelerator circuits for
simulated annealing based algorithms.
On-board autonomy requires processing complex event
scheduling tasks among others. These algorithhms are complex and highly
parallelizable. Therefore to take advantage of accelerating these applications
on an on-board FPGA based computer, it is necessary to design complex VLSI
architectures that take several months of engineering. Through our tool (SATH),
scientists and software developers who have no knowledge of hardware circuit
design can feed in ANSI C code for the application and get a circuit
automatically designed and ported onto an FPGA, saving months of design time
and errors. - We have started working on customized fault mitigation
and tolerance circuit designs to protect the auto-generated FPGA circuit from
single and multiple soft errors. Refining this methodology and optimizing the
tool flow for fault protection will be the focus of our effort for the upcoming
year.
Brian C. Williams
Diagnosing Complex
Software and Hardware
Brian Williams, Paul Elliott MIT MERS
This year has been marked by continued progress on the
maturation of the mixed software and hardware monitoring capability in the
context of the EO-1 mission. This capability is built upon the use of
Probabilistic Hierarchical Constraint Automata (PHCA), which describe the
behavior of the mixed software and hardware system to the monitoring
capability. Maturation has focused on improving the expressivity of the models
supported by the capability as well as improving the capacity of the algorithm
to support larger models efficiently through algorithmic improvements. This
year has also shown substantial progress in improving the modeling language,
Reactive Model-Based Programming Language (RMPL), used to model the PHCA. We
have also improved upon the language s compiler to support generating more
complex models. We have been working in collaboration with JPL on development
scenarios relevant to the EO-1, both to focus the EO-1 modeling effort and to
help specify correct software behavior. Using the language improvements, along
with the direction of the scenarios, we ve been able to continue the evaluation
of and improve upon the modeling of the EO-1 mission.
This project directly responds to the Applied Information
Systems Research (AISR) program objectives of NASA s Research Opportunities in
Space and Earth Sciences (ROSES). In particular, our technology is expected to
enhance the science productivity of NASA s space flight missions that are
sponsored by the Science Mission Directorate (SMD). This technology builds upon
the success of the Autonomous Science Experiment (ASE) onboard the Earth
Observing One (EO-1) mission, by providing an onboard capability for monitoring
and diagnosing software and hardware systems, as well as mission goals.
Enhancing the ASE software through the proposed fault management capability
enables extremely high reliability operations, resulting in an increased return
of scientific data. This work also directly responds to NASA s Strategic
National Objective to Study the Earth system from space and develop new
space-based and related capabilities for this purpose. The maturation and
validation of our proposed technology in the context of EO-1 demonstrates its
potential for long term impact on many future NASA missions that are
increasingly relying on complex software and hardware systems.
We are still in the development phase, so most usage is
in-house. We normally remain in detailed email contact with external users,
both to work through issues and to see examples of the models developed for our
tool. We intend to continue to mature
this capability in conjunction with JPL to match the requirements of the EO1
mission.
Volodymyr Kindratenko
Astrophysical
Algorithms on Novel HPC Systems
Robert Brunner UIUC Astronomy Department
Volodymyr Kindratenko UIUC NCSA
The objective of our research is to demonstrate the
practical use of alternative computing technologies, such as those based on
FPGAs and GPUs, for advanced astrophysical algorithms and applications,
particularly those involving very large data sets. In the past year we explored
the use of multi-core CPUs, Field-Programmable Gate Array (FPGA) based
co-processors, and NVIDIA Graphics Processing Units (GPUs) for accelerating
two-point correlation functions which find an extensive use in cosmology
applications. More specifically, i) we extended the two-point angular
correlation function brute force implementation from the previous year to work
on a cluster consisting of multi-core SMP nodes using Message Passing Interface
(MPI), ii) implemented the compute kernel of this cluster application on a
Nallatech H101 FPGA application accelerator board using DIME-C language and
DIMEtalk API and expanding the application to utilize FPGA accelerators
available in the cluster nodes (16 in total), and iii) experimented with the
same compute kernel on the NVIDIA GPU G80 platform using CUDA development environment.
On the Nallatech H101 platform we achieved a 4x-8x per FPGA kernel speedup as
compared to a modern processor core while maintaining 100% accurate results and
running the computations only at fraction of the power budget of the
conventional system. The kernel speedup obtained on the GPU platform was more
substantial: 60x. However, current generation of GPUs only supports 32 bit
arithmetic, thus reducing the useful range of calculations to angular
separations above 1 arcminute. We expect that we will be running full
double-precision calculations on the next generation GPUs to be introduced
later this year.
We achieve significant application speedup while using
accelerator-based systems, which translates into the ability of NASA scientists
to process much larger datasets within the same time frame as current HPC
systems can process smaller datasets. This allows scientists to answer
questions that otherwise can only be answered with the use of very large HPC
resources, or cannot be answered at all with the current technology. Also,
significant cost savings can be achieved. We
are planning to release software in the upcoming year.
There are two main directions that we will continue
exploring novel architectures in the upcoming year: 1) With the introduction of
the double-precision floating-point GPU chips later this year, we will research
and implement the two-point angular correlation kernel on this platform and
will extend our existing cluster application to simultaneously take advantage
of the multi-core chips as well as the Nallatech H101 FPGA accelerators and
NVIDIA GPUs. We will leverage an ongoing effort in NCSA s Innovative Systems
Lab on a novel run-time environment for supporting efficient use of diverse
compute resources. 2) We already extended our base line k-nearest neighbor
kd-tree based implementation of the instance based classification code to work
on a multi-core SMP system via pthreads and tested it with multi-million point
datasets. In the upcoming year we will extend this application via MPI to work
on large cluster systems and will investigate the use of FPGAs and GPUs to
accelerate the kd-tree based range search algorithm used in the k-nearest
neighbor classifier. Our preliminary analysis shows that such an implementation
could provide significant benefits in speeding up the calculations, however it
is not trivial to implement a control flow dominant code on these
architectures.
Alexander V. Panasyuk
Innovative Techniques
for Producing Line-of-Sight Corrected
Synoptic Maps
Alexander V. Panasyuk Harvard-Smithsonian
CfA
Leonard Strachan, John L. Kohl Harvard-Smithsonian
CfA
We present the status for a project which has as its goal to
build a database of localized plasma parameters in the solar corona. The
project uses more than a decade of spectroscopic data from the Ultraviolet
Coronagraph Spectrometer (UVCS) on the SOHO spacecraft. UVCS measurements of
line-of-sight integrated spectral profiles at three wavelengths (H I Ly-alpha,
O VI 103.2 nm and 103.7 nm) are used to produce 3D maps of ion velocity
distributions and bulk outflows. A previously developed algorithm for
reconstructing the emissivities has been extended to include the reconstruction
for the line-widths which are needed to compute accurate outflow velocities.
The specifics that complicate a tomographic reconstruction of the solar corona,
such as coronal dynamics, are discussed, as well as methods to estimate the
uncertainties introduced in the reconstruction.
This effort uses data from the the Solar and Heliospheric
Observatory (SOHO) but the algorithms developed could be used by other missions
that produce imaging or spectroscopic data, even non-solar missions. We plan to set up a Web site that will allow
users to download data and selected softare for visualizing the data. Usage
will be tracked by logging the downloads.
For the upcoming year, we plan to
1) complete the final database of
outflow velocity maps for different activity periods of solar cycle 23;
2) develop and test the data
retrieval tools for selecting the data;
3) document and write a Users
Manual to be supplied with the software; and
4) publish the coronal emissivity,
line-width (temperature), and outflow velocity maps in an archival journal.
Sara Graves
A Distributed
Knowledge Extraction Framework
Sara Graves, Rahul Ramachandran, Helen Conover, Sunil Movva
University of Alabama
Huntsville
Peter Fox NCAR
Deborah McGuiness RPI
Scientific data mining is a very powerful means for
automated knowledge extraction from the ever-increasing volumes of science
observations and model output data available. NASA s Second Data Mining
Workshop found that maturing data mining techniques show potential for
significantly expanding the scientific understanding of NASA s Earth science
data. However, this type of tool has generally been difficult for domain
scientists and students to fully exploit without extended learning curves. And
even data mining specialists may not be familiar with the full range of
components in a mining toolkit, so potentially useful mining strategies may be
ignored. To facilitate exploitation of these promising techniques by the
increasingly IT-sophisticated NASA science community, we have investigated the
use of Semantic Web technologies to build a Smart Assistant for Mining via the
seed funding received through the AISR program. This project has successfully
designed and developed a prototype that demonstrates the value of smart
assistance for mining. The prototype reuses an existing toolkit of data mining
web services designed specifically for the analysis of NASA data in a
web-based, service-oriented architecture. This project has also developed an
initial ontology describing data mining services, with links to data
ontologies. The prototype user interface tool, integrates semantic reasoning
into a traditional workflow composer and allows users to discover available
data and services, assist users in composing mining workflows, and invoke them
to perform the desired analysis.
This prototype has successfully demonstrated that it can
assist researchers in creating data analysis and mining workflows for science
problems. It will also position web and grid services for integration with many
other science data services in the Semantic Web Services context, pointing the
way toward increased science return from NASA data.
Since this project was partially funded primarily to design,
develop and test the prototype for it usefulness, no direct metrics were
tracked. However, the prototype has been presented at several meetings and
conferences The seed funding from AISR has
been fully utilized to develop the prototype. The prototype was instrumental in
successfully competing for a NASA ACCESS grant. Full scale development of the
data mining and data ontologies and the tool will be completed under the NASA
ACCESS program
Kaichang Di
Integration of Orbital, Descent and Ground Imagery
Ron Li, Kaichang Di, Ju Won Hwangbo, Yunhang Chen The Ohio State University
In FY07, we developed methods for rigorous photogrammetric
processing of HiRISE (High Resolution Imaging Science Experiment) stereo
images. Using the developed sensor model and bundle adjustment method, we
processed HiRISE images and derived a digital terrain model and slope map of
the Mars Exploration Rover (MER) Gusev Crater landing site. These mapping
products helped the MER team evaluate potential routes to ¡°Von Braun¡± and
find ¡°Winter Heaven¡±, enabling Spirit rover to survive the local winter.
Orbital ground integration has been researched as well.
The integration of Mars orbital, descent and ground imagery
has the potential to achieve the best possible accuracy for integrated Mars
topographic mapping capability analysis. This research directly contributes to
the AISR program high priority area ¡°to increase science return from data¡± by
integrating three types of imagery and deriving topographic products and rover
localization data that are far superior to those that can be derived from a
single type of imagery. They are critical to MER, MSL and future landed
missions. The developed method and mapping products are currently being used in
MER operations by the science and engineering teams. They are using them to
evaluate rover potential traverses and winter haven sites. For example, we
processed a pair of HiRISE images and derived a digital terrain model and slope
map of the Mars Exploration Rover (MER) Gusev Crater landing site. These
mapping products helped MER operations evaluate potential routes to ¡°Von
Braun¡± and find ¡°Winter Heaven¡±, enabling Spirit rover to survive the local
winter. This high accuracy, integrated mapping capability will be very valuable
for planetary scientists in their studies, particularly in regional geology,
crater mechanics and modeling, cross-site geological processes, etc. It will
greatly aid traverse planning and rover navigation strategy development for
future landed missions. The developed orbital image mapping method can be
directly used in future missions for landing-site selection.
We track the usage of our topographic products though daily
Mars mission operation meetings, peer-reviewed publications, professional
conference presentations, project reports and personal communications. We know
that our products are in MER’s mission briefings to NASA headquarters, team
members¡¯ science papers, daily mission LTP reports and weekly mission
manager’s reports etc.
In FY08, we will develop techniques for extraction, modeling
and matching of landmarks from orbital and ground images. We will also develop
software for integrated bundle adjustment of orbital, descent and ground
imagery.
Daniel Sorin
Autonomic Computer
Hardware for Space Missions
http://www.ee.duke.edu/~sorin/faultfinder/index
Daniel Sorin, Sule Ozev Duke
University
1) Provably comprehensive, low-cost, low-power detection of
errors in microprocessor cores
2) An all-software technique for tolerating permanent faults
in processor cores
3) An all-hardware technique for tolerating permanent faults
in multicore chips
NASA s space missions depend on computer hardware, and our
research enables this hardware to be reliable without user intervention and
without resorting to high-cost, high-power solutions like triple modular
redundancy (TMR).
We do not yet have a product, but we are building a
prototype that is being partly sponsored by Toyota InfoTechnology Center.
Automotive computing has many similarities to space mission computing, in terms
of the need for low-cost reliability. Our work is also being used by other
academic and industrial researchers.
We plan to extend error detection to other parts of the
computer system, including the address translation system and I/O. We are also
developing an interface between the hardware and the operating system (OS) that
enables better scheduling of software threads on cores that are affected by
faults and voltage/frequency scaling.
Brian Doty
Enhancing GrADS for
Earth System Science Research
Brian Doty, Jennifer Adams IGES/COLA
GrADS, GDS, and Greta are a suite of software packages that
support the organization, access, analysis, and visualization of Earth science
data. GrADS, the cornerstone of the software suite, has been in use for 20
years. Recent enhancements increase the software s capabilities to handle
large, multi-member multi-ensemble data sets. The 4-dimensional gridded data
model in GrADS (longitude, latitude, level, and time) has been expanded to
include a 5th dimension that is generally applicable, but intended for use as
an ensemble dimension. GrADS also has a new interface for handling data in the
GRIB2 format, the new standard for much of the ensemble forecasts being
distributed by the international forecasting centers participating in the TIGGE
project.
Updates are intended to keep pace with high-end computing
resources, expanding data volumes, and model output and observational data that
comes in new formats and grid structures. We
have metrics of usage statistics from our public GDS servers. We monitor the
number of users subscribed to the GrADS users listerver. We keep track of the
number of FTP downloads of the software.
We will add support for GIS-compliant output: rasterized
images in the geo-TIFF format, and shapefiles that conform to the ESRI
specification. We will also support the HDF5 format for gridded data. We will
begin the planning and design for the support of quasi-regular swath data.