A compendium of obstacles and Ideas for how to move
from doing machine learning _on_ healthcare data to
using machine learning _for_ healthcare, to achieve real
impact.
We developed a method to calibrate classification predictions
by leveraging similarity in feature space. This approach can
particularly benefit subpopulations within the data set.
Characterizing False Positives in CRISM VNIR Frost
Detections.
Pranav Rupireddy, Gary Doran, Christina E. Viviano, Serina
Diniega, Kiri L. Wagstaff, Samuel F. A. Cartwright, Katie
Hancock, Frank P. Seelos, Kimberley D. Seelos, and Jake
M. Widmer. 6th Planetary Data Workshop, Abstract #7051, June 2023.
We applied a convolutional neural network to 13 years of
CTX orbital images of Mars and found ~70 new fresh
impacts, plus a large number of additional candidates that
can help us reduce observational bias in the catalog of
known impacts.
New Craters on Mars: An Updated Catalog.
Ingrid J. Daubar, Colin M. Dundas, Alfred S. McEwen, Annabelle
Gao, Daniel Wexler, Sylvain Piqueux, G. S. Collins,
K. Miljkovic, T. Neidhart, J. Eschenfelder, G. D. Bart, Kiri
Wagstaff, Gary Doran, L. Posiolova, M. Malin, G. Speth,
D. Susko, and A. Werynski. Journal of Geophysical Research: Planets, DOI
10.1029/2021JE007145, 2022.
New Craters on Mars: Results from a Complete Catalog of
1,203 Recent Impacts.
Ingrid J. Daubar, Colin Dundas, Alfred S. McEwen,
Annabelle Gao, Daniel Wexler, Sylvain Piqueux,
G. S. Collins, K. Miljkovic, T. Neidhart, J. Eschenfelder,
G. D. Bart, Kiri Wagstaff, Gary Doran, L. Posiolova,
G. Speth, D. Susko, A. Werynski, and Michael Malin. 53rd Lunar and Planetary Science Conference, Abstract
#1590, March 2022.
Geomorphological Differences in Slope Streaks and RSL.
David Stillman, Katie Primm, Rachael Hoover, Brian Bue,
Tim Michaels, Hannah Kaplan, Kiri Wagstaff, Jake Lee,
Adnan Ansar, and Lori Fenton. Geological Society of America, October 2022.
LROCNet: Detecting Impact Ejecta and Older Craters on the Lunar Surface.
(poster,
content)
Emily Dunkel, Steven Lu, Kevin Grimes, James McAuley, and Kiri Wagstaff. Fall Meeting of the American Geophysical Union, December 2022.
Mars Image Content Classification: Three Years of NASA
Deployment and Recent Advances
(IAAI Deployed Application Award).
Kiri Wagstaff, Steven Lu, Emily Dunkel, Kevin Grimes,
Brandon Zhao, Jesse Cai, Shoshanna B. Cole, Gary Doran,
Raymond Francis, Jake Lee, and Lukas Mandrake. Proceedings of the Thirty-Third Annual Conference on
Innovative Applications of Artificial Intelligence,
p. 15204-15213, 2021.
We report on updates to Mars image classifiers deployed
at the NASA Planetary Data System, for orbital and rover images,
and lessons learned that inform ongoing activities to build
new classifiers for additional missions.
HiRISE data set (labeled Mars orbiter images, V3.2):
Designing a Machine Learning Local Data Dictionary.
(Poster)
Minh Le, Michael McAuley, and Kiri Wagstaff. 5th Planetary Data Workshop and Planetary Science Informatics and Data Analytics (PSIDA) Meeting, Abstract #7085, July 2021.
Machine learning finds / Fresh impacts on Mars that we
/ Missed before, what now?
Benchmarking Machine Learning on the Myriad X Processor
Onboard the ISS.
Emily Dunkel, José Luis Espinosa-Aranda, Juan Romero Cañas,
Léonie Buckley, Zaid Towfic, Faiz Mirza, Jason Swope, Damon
Russell, Joseph Sauvageau, Douglas Sheldon, Steve Chien, Mark
Fernandez, Carrie Knox, Kiri Wagstaff, Steven Lu, Michael
Denbina, Deegan Atha, Michael Swan, and Hiro Ono. International Space Station Research and Development
Conference, August 2021.
Novelty-Driven Onboard Targeting for Mars Rovers.
Kiri L. Wagstaff, Raymond Francis, Hannah Kerner, Steven Lu,
Favour Nerrise, James F. Bell III, Gary Doran, and Umaa Rebbapragada. Proceedings of the International Symposium on Artificial Intelligence, Robotics and Automation in Space, 2020.
How can we train classifiers when we don't know what all
possible classes are before we begin? To support
discovery in exploration settings, we employ active learning
to help with discovering classes and labeing data.
We apply this approach to a Mars rover data set.
We combine novelty detection with CNN image features
to achieve rapid discovery with interpretable explanations of
novel image content. We also report on a user study of
explanation utility.
We compare multiple novelty detection methods on
multispectral (6-channel) images collected by a Mars rover.
We find that some methods are more sensitive to spectral
(compositional) novelty, while others are more sensitive to
morphological (shape-based) novelty.
Data: Mars novelty detection Mastcam labeled dataset
COSMIC: Content-based Onboard Summarization to Monitor Infrequent Change.
Gary Doran, Steven Lu, Maria Liukis, Lukas Mandrake,
Umaa Rebbapragada, Kiri L. Wagstaff, Jimmie Young, Erik Langert,
Anneliese Braunegg, Paul Horton, Daniel Jeong, and Asher Trockman. Proceedings of the IEEE Aerospace Conference, 2020.
ARIEL: Autonomous Excavation Site Selection for Europa Lander Mission Concept.
Masahiro Ono, Gary Doran, Erik Langert, Kiri Wagstaff, David Inkyu Kim, Aaron Gaut, Tara Estlin, Abhi Jain, Marissa Cameron, David Muliere, Erik Roberts, Kristopher Kriechbaum, and Glenn Reeves. Proceedings of the IEEE Aerospace Conference, 2020.
Machine-Assisted Discovery Through Identification and
Explanation of Anomalies in Astronomical Surveys.
Kiri L. Wagstaff, Eric Huff, and Umaa Rebbapragada. Astronomical Data Analysis Software and Systems Conference, 2020.
Time-Series Analysis Methods for On-board Detection of Magnetic Field Boundaries by Europa Clipper.
Ameya Daigavane, Kiri L. Wagstaff, Corey J. Cochrane,
Caitriona M. Jackman, and Gary B. Doran, Jr. Second AI and Data Science Workshop, March 2020.
Content-based Classification of Mars Imagery for the PDS Image Atlas.
Steven Lu and Kiri L. Wagstaff. Second AI and Data Science Workshop, March 2020.
We make a case for the utility of onboard
analysis of data collected by Europa Clipper (future mission
to Europa) to enable prioritization of data for downlink.
We describe three onboard detection algorithms to address the
science goals detecting thermal anomalies (hot spots),
compositional anomalies, and active plumes. We also report
lessons learned from developing and evaluating these methods
in the context of an onboard computing environment and mission
acceptance.
We detect sub-pixel moving vessels in
ocean scenes as bright outliers in 14-bit observations.
Multi-angle MISR observations cover a temporal span sufficient
to also yield an estimate of vessel speed and direction of
motion from a single observation.
UKIRT Microlensing Survey as a Pathfinder for WFIRST.
Geoff Bryden, Yossi Shvartzvald, Savannah Jacklin, Selina Chu,
Kiri Wagstaff, Sebastiano Calchi Novati, C. Beichman, S. Gaudi,
C. Henderson, S. Johnson, D. Nataf and M. Penny. Extreme Solar Systems IV, August 2019.
We apply image analysis and adaptive thresholding methods to detect the target body's limb and determine whether there is evidence for an active plume. This technology could potentially be used to detect plumes in situ within images of Europa obtained by the Europa Imaging System (EIS) instrument on the upcoming Europa Clipper mission.
Data set (labeled images with and without plumes):
We describe experiments with different
novelty detection methods applied to rover images to aid in
directing attention to the images most likely to be of
interest to mission scientists and planners. These include
images of meteorites, rock veins, broken rocks, and other
surface features with unusual mineralogy. We found that a
convolutional autoencoder had the best overall precision@N
performance. However, individual classes showed different
behavior, with a standard SVD yielding the best AUC for
naturally occurring features (broken rocks, float rocks,
and meteorites) and a generative adversarial network (GAN)
yielding the best performance for novel features created
by the rover itself (drill holes, brush spots, and dump piles).
We report on several improvements to the classifiers we developed to assign content labels to NASA images collected by Mars rovers and orbiters. The improvements include data augmentation, classifier calibration, and updates to the fine-tuning methodology.
Responsive Onboard Science for the Europa Clipper Mission.
Kiri Wagstaff, Gary Doran, Ashley Davies, Saadat Anwar,
Srija Chakraborty, Diana Blaney, Marissa Cameron, Jonathan Bapst,
Steve Chien, Corey Cochrane, Ingrid Daubar, Serina Diniega,
Cynthia Phillips, and Sylvain Piqueux. JPL Data Science Showcase, Poster 29, April 2019.
Optimizing WFIRST Exoplanet Yield using Machine Learning
for Microlensing Event Detection.
Selina Chu, Kiri L. Wagstaff, Geoffrey Bryden, Yossi Shvartzvald,
and Savannah Jacklin. JPL Data Science Showcase, Poster 64, April 2019.
Deep Mars: A CNN Approach that Enables Image Content-Based
Search for PDS Image Atlas.
Steven Lu, Kiri Wagstaff, Kevin Grimes, Gary Doran, Lukas Mandrake,
Jake Lee, and Jesse Cai. JPL Data Science Showcase, Poster 35, April 2019.
Content-based Onboard Summarization to Monitor Infrequent
Change (COSMIC) - An Ultra-Low Bandwidth, Autonomous System
to Globally Monitor Mars for Dynamic Events.
Lukas Mandrake, Kiri Wagstaff, Gary Doran, Steven Lu,
Erik Langert, and Jimmie Young. JPL Data Science Showcase, Poster 14, April 2019.
An Anomaly Catalog for the Dark Energy Survey.
Umaa Rebbapragada, Eric Huff, and Kiri Wagstaff. JPL Data Science Showcase, Poster 68, April 2019.
We apply anomaly detection methods to NIMS observations of Europa and describe how they could be used onboard the upcoming Europa Clipper mission by the MISE instrument for in-situ detections. See also the poster.
We evaluated several machine learning classifiers in terms of their ability to predict which books are most likely to be weeded from a collection. We applied this method to a collection of more than 80,000 items from an academic library and found statistically significant agreement (p = 0.001) between classifier and librarian decisions.
Novelty Detection for Multispectral Planetary Images.
Hannah Rae Kerner, Danika F. Wellington, Kiri Wagstaff,
Samantha Jacob, James F. Bell III, and Heni Ben Amor. Fall Meeting of the American Geophysical Union,
Abstract #IN14A-01, December 2018.
Responsive Onboard Science for Europa Clipper.
Kiri L. Wagstaff, Ashley Davies, Gary Doran, Srija Chakraborty,
Saadat Anwar, Diana L. Blaney, Steve Chien, Philip R. Christensen,
and Serina Diniega. Outer Planets Assessment Group meeting, September 2018.
Onboard Detection of Thermal Anomalies for Europa Clipper.
Gary Doran, Ashley G. Davies, Kiri L. Wagstaff, Saadat Anwar, Diana L. Blaney, Steve Chien, Phil Christensen, and Serina Diniega. European Planetary Science Congress, Abstract #528, September 2018.
Scientist-Guided Autonomy for Self-Reliant Rovers.
Gary Doran, Umaa Rebbapragada, Eugenie Song, Kiri Wagstaff, Daniel Gaines, Robert Anderson, and Ashwin Vasavada. IJCAI 2017 Workshop on AI in the Oceans and Space, 2017.
Onboard Autonomy on the Intelligent Payload EXperiment CubeSat Mission.
Steve Chien, Joshua Doubleday, David R. Thompson, Kiri L. Wagstaff,
John Bellardo, Craig Francis, Eric Baumgarten, Austin Williams,
Edmund Yee, Eric Stanton, and Jordi Piug-Suari. Journal of Aerospace Information Systems, vol. 14, no. 6, p. 307-315, 2017.
We surveyed readers to find out about their attitudes toward marginalia, and whether and how often they indulged in it themselves. We also investigated whether marginalia translates into electronic books and which features are most desired by users of e-readers.
A Machine Learning Classifier for Fast Radio Burst Detection at the VLBA.
Kiri L. Wagstaff, Benyang Tang, David R. Thompson, Shakeh Khudikyan, Jane Wyngaard, Adam T. Deller, Divya Palaniswamy, Steven J. Tingay, and Randall B. Wayth. Publications of the Astronomical Society of the Pacific, 128:966(084503), 2016.
We describe the classifier used to filter VLBA fast radio burst detections
to reduce the human effort needed to review and classify them manually. We also describe
the web interface used for ongoing reviewing, which you can see for yourself at
the V-FASTR Data Portal. You can also view
the latest classifier results
(updated daily).
Real-Time Adaptive Event Detection in Astronomical Data Streams.
Randall B. Wayth, Kiri L. Wagstaff, Steven J. Tingay, Walid A. Majid, Divya Palaniswamy, Adam T. Deller, David R. Thompson, and Sarah Burke-Spolaor.
IEEE Intelligent Systems, 29(1):48-55, 2014.
Autonomous Onboard Surface Feature Detection for Flyby Missions.
Thomas J. Fuchs, Brian D. Bue, Julie Castillo-Rogez, Kiri Wagstaff, David R. Thompson. Proceedings of the International Symposium on Artificial
Intelligence, Robotics and Automation in Space, 2014.
This is our introduction to the special issue of MLJ
featuring papers that show how machine learning can benefit
science investigations and problems of interest to society at large.
We analyzed 11,750 ChemCam spectra of
soil and rock targets on Mars. Here we present some of the most
unusual ones, according to the DEMUD discovery algorithm.
2013
Papers
Online Classification for Time-Domain Astronomy.
Kitty K. Lo, Tara Murphy, Umaa Rebbapragada, and Kiri L. Wagstaff. Proceedings of the Astroinformatics workshop, IEEE International Conference on Data Mining, December 2013.
We've created a camera that both collects images
and analyzes them for interesting textures. The analysis results can
identify high-priority targets for follow-up study. Ultimately, our
goal is to help rovers and spacecraft to graduate from remote
instrument to field assistant.
Guiding Scientific Discovery with Explanations using DEMUD
(pdf, 7 pages, 291K).
Kiri L. Wagstaff, Nina L. Lanza, David R. Thompson,
Thomas G. Dietterich, and Martha S. Gilmore. Proceedings of the Twenty-Seventh AAAI Conference on
Artificial Intelligence (AAAI-13), p. 905-911, 2013.
DEMUD is an algorithm for quickly finding
interesting novelties or anomalies in large data sets. Uniquely,
it also provides explanations for why each item is considered
interesting. We report on experiments with hyperspectral data from Mars
orbit and the Martian surface, finding that DEMUD indeed discovers
items of scientific interest, and that the explanations are useful:
with them, a scientist's ability to classify spectra doubled.
VAST: An ASKAP Survey for Variables and Slow Transients.
Tara Murphy, Shami Chatterjee, David L. Kaplan, Jay Banyer, Martin E. Bell, Hayley E. Bignall, Geoffrey C. Bower, Robert Cameron, David M. Coward, James M. Cordes, Steve Croft, James R. Curran, S. G. Djorgovski, Sean A. Farrell, Dale A. Frail, B. M. Gaensler, Duncan K. Galloway, Bruce Gendre, Anne J. Green, Paul J. Hancock, Simon Johnston, Atish Kamble, Casey J. Law, T. Joseph W. Lazio, Kitty K. Lo, Jean-Pierre Macquart, Nanda Rea, Umaa Rebbapragada, Cormac Reynolds, Stuart D. Ryder, Brian Schmidt, Roberto Soria, Ingrid H. Stairs, Steven J. Tingay, Ulf Torkelsson, Kiri Wagstaff, Mark Walker, Randall B. Wayth, and Peter K. G. Williams. Publications of the Astronomical Society of Australia, 30, 2013.
An overview of the VAST project, which will search for new variables and slow transient events in data collected by ASKAP, the Australian SKA Precursor radio array.
Abstracts and Posters
TextureCam: A Smart Camera for Microscale, Mesoscale, and Deep Space.
David R. Thompson, William Abbey, Abigail Allwood, Dmitriy Bekker,
Benjamin Bornstein, Nathalie A. Cabrol, Rebecca Castano,
Steve A. Chien, Joshua Doubleday, Tara Estlin, Greydon Foil,
Thomas Fuchs, Daniel Howarth, Kevin Ortega, and Kiri L. Wagstaff. 44th Lunar and Planetary Science Conference, March 2013.
2012
Papers
Smart Cameras for Remote Science Survey (pdf, 8 pages, 12M).
David R. Thompson, William Abbey, Abigail Allwood,
Dmitriy Bekker, Benjamin Bornstein, Nathalie A. Cabrol,
Rebecca Castano, Tara Estlin, Thomas Fuchs, and Kiri L. Wagstaff. Proceedings of the International Symposium on Artificial
Intelligence, Robotics and Automation in Space, 2012.
Put a random forest classifier brain inside a camera,
and you can accomplish some impressive things!
Our commensal fast transient detection system (V-FASTR)
has acquired over 1300 hours of radio astronomy observations, allowing us to place
new limits on the (non-)occurrence of such transients.
Machine Learning that Matters (pdf, 6 pages, 234K)
[Slides (pptx, 2.1M)].
Kiri L. Wagstaff. Proceedings of the Twenty-Ninth International Conference on
Machine Learning (ICML), p. 529-536, 2012.
This position paper outlines some ways in which much of
current machine learning research has become disconnected from
real problems of significance to the world outside of machine learning.
Its goal is to instigate creative discussions about how to
remedy the situation.
This is an extended version of our CIDU 2011
paper, which offers two improvements: modeling local
context around known false alarms, and employing a sparse
PCA for better interpretability of the results.
Surface Sulfur Detection via Remote Sensing and Onboard Classification.
Lukas Mandrake, Umaa Rebbapragada, Kiri L. Wagstaff, David Thompson,
Steve Chien, Daniel Tran, Robert T. Pappalardo, Damhnait Gleeson,
and Rebecca Castano. ACM Transactions on Intelligent Systems and Technology, 2012.
A report on the use of SVMs and careful data
analysis to develop a reliable surface sulfur detector from
orbital data, with implications for the search for biosignatures
on Europa.
Salience-based methods for finding interesting and unusual surface features on Mars, and a discussion of several such features we discovered, including new dark slope streaks and seasonally exposed bedforms.
A report on the performance of different
strategies for classifying variable and transient sources
in radio astronomy data, in support of the VAST investigation
using the ASKAP array. We explored both offline (archival)
and online (realtime) classification methods.
Big Data Challenges for Large Radio Arrays (pdf, 6 pages, 1.2M).
Dayton L. Jones, Kiri Wagstaff, David R. Thompson, Larry D'Addario,
Robert Navarro, Chris Mattmann, Walid Majid, Joseph Lazio,
Robert Preston, and Umaa Rebbapragada. Proceedings of the 33rd IEEE Aerospace Conference, March 2012.
Data Clustering.
Kiri L. Wagstaff.
Chapter in Advances in Machine Learning and Data Mining for Astronomy, Chapman & Hall/CRC Press, 2012.
Basic survey of clustering methods and how they can be useful in the context of astronomical data analysis.
Exploring Mars via Autonomosly Networked Spacecraft.
E. Jay Wyatt, Scott C. Burleigh, Loren P. Clare, J. Leigh Torgerson, and Kiri L. Wagstaff. Concepts and Approaches for Mars Exploration, Abstract #4310, 2012.
TextureCam: Autonomous Image Analysis for Astrobiology Survey
(pdf, 2 pages, 1M).
David R. Thompson, Abigail Allwood, Dmitriy Bekker, Nathalie A. Cabrol,
Tara Estlin, Thomas Fuchs, and Kiri L. Wagstaff. 43rd Lunar and Planetary Science Conference, March 2012.
Describes the TextureCam project, which aims to build a
trained random forest classifier into an FPGA attached to a camera,
for use onboard future rovers. This "smart camera" will be able to
classify scenes as they are collected, to aid in autonomous targeting
of other instruments.
Why space missions need machine learning, and where we ML folk should be devoting our efforts.
V-FASTR: The VLBA Fast Radio Transients Experiment.
Randall B. Wayth, Walter F. Brisken, Adam T. Deller, Walid A. Majid,
David R. Thompson, Steven J. Tingay, and Kiri L. Wagstaff. The Astrophysical Journal, 735(2), doi: 10.1088/0004-637X/735/2/97, 2011.
Classification of ASKAP VAST Radio Light Curves.
Umaa Rebbapragada, Kitty Lo, Kiri L. Wagstaff, Colorado Reed,
Tara Murphy, and David R. Thompson. Proceedings of the International Astronomical Union,
New Horizons in Time-Domain Astronomy, p. 397-399, 2011.
Fast Transient Detection as a Prototypical "Big Data" Problem.
Dayton L. Jones, Kiri Wagstaff, David Thompson, Larry D'Addario,
Robert Navarro, Chris Mattmann, Walid Majid, Umaa Rebbapragada,
Joseph Lazio, and Robert Preston. Proceedings of the International Astronomical Union,
New Horizons in Time-Domain Astronomy, p. 340-341, 2011.
Using automated change detection and landmark analysis, we identified several transient surface features in the El Dorado dune field. The features were likely caused by the activity of dust devils, both removing dust from the surface and later depositing it. The changes are subtle enough that detecting them manually can be very difficult.
2010
Papers
Constrained Clustering.
Kiri L. Wagstaff. Encyclopedia of Machine Learning, p. 220-221, Springer, 2010.
The Commensal Real-Time ASKAP Fast-Transients (CRAFT) Survey.
Jean-Pierre Macquart, M. Bailes, N. D. R. Bhat, G.C. Bower, J.D. Bunton, S. Chatterjee, T. Colegate, J.M. Cordes, L. D'Addario, A. Deller, R. Dodson, R. Fender, K. Haines, P. Hall, C. Harris, A. Hotan, S. Johnston, D.L. Jones, M. Keith, J.Y. Koay, T.J.W. Lazio, W. Majid, T. Murphy, R. Navarro, C. Phillips, P. Quinn, R. A. Preston, B. Stansby, I. Stairs, B. Stappers, L. Staveley-Smith, S. Tingay, D. Thompson, W. van Straten, K. Wagstaff, M. Warren, R. Wayth, and L. Wen (the CRAFT Collaboration). Publications of the Astronomical Society of Australia, 27(3),
p. 272-282, doi:10.1071/AS09082, 2010.
This paper describes the goals of the CRAFT team, an international collaboration developing new ways to detect transient radio events in real-time with large radio array telescopes. It serves as a precursor to develop technology needed for the upcoming Square Kilometer Array (SKA).
Given an image of a rock that contains layered structures, is it possible to determine whether the layers were created by life (biogenic)? We evaluated several quantitative measures that capture the degree of complexity in visible structures, in terms of compressibility (to detect order) and the entropy (spread) of their intensity distributions. None of the techniques provided a consistent, statistically significant distinction between all biogenic and abiogenic samples, but the PNG compression ratio provided the strongest distinction and could inform future techniques.
How can the best classifier be learned, when data features may be missing both while training and when classifying new items? Which missing features should be acquired, and in what order? This paper presents a greedy approach to achieving good performance with low feature acquisition costs.
Progressive refinement starts with a rough classification result for all items and then iteratively refines the least-confidently-classified ones. This approach is ideal for applications that require "anytime" solutions that are always complete, such as onboard processing of camera images that informs spacecraft decisions.
Modeling and Learning Preferences over Sets (author's version pdf, 32 pages, 1.4M;
official journal page).
Kiri L. Wagstaff, Marie desJardins, and Eric Eaton. Journal of Experimental and Theoretical Artificial Intelligence, doi:10.1080/09528130903119336, vol. 22, issue 3, p. 237-268, 2010.
This work permits explicit control over the diversity and the depth of an automatically generated collection (e.g., music playlist, rover image download batch) through user preferences. The preferences can also be learned from examples of highly preferred sets assembled by the user.
Change Detection in Mars Orbital Images using Dynamic Landmarking (pdf, 2 pages, 596K).
Kiri L. Wagstaff, Julian Panetta, Adnan Ansar, Melissa Bunte, Ronald Greeley, Mary Pendleton Hoffer, and Norbert Schorghofer. 41st Lunar and Planetary Science Conference, March 2010.
We developed a content-based analysis of Mars orbital images to automatically identify and classify landmarks (impact craters, dust devil tracks, slope streaks, etc.) and then detect changes in the landmarks (both new and vanished).
This report describes a concept for how existing Mars
assets (orbiters and rovers) could use Delay-Tolerant Networking
to support automated, coordinated science investigations.
Simulating and Detecting Radiation-Induced Errors for Onboard Machine Learning (pdf, 7 pages, 243K).
Robert Granat, Kiri L. Wagstaff, Benjamin Bornstein, Benyang Tang, and Michael Turmon. Proceedings of the Third IEEE International Conference on Space Mission Challenges for Information Technology (SMC-IT), p. 125-131, July 2009.
An update to the ICML paper below, this work extends the analysis
of ML algorithm radiation sensitivity to SVM classifiers as well.
It also includes updated quantitative results for clustering, finding that
radiation levels would have to increase by 9 orders of magnitude above
the level observed in low-Earth orbit before any impact to performance was
observed.
The work described in this paper assesses how sensitive various
k-means clustering algorithms are when exposed to destructive
radiation during the clustering process. One key finding was that,
for the data sets tested, k-means can operate in low Earth orbit radiation
levels without requiring rad-hardened memory. Also, subsampling k-means was
more radiation resistant than kd-kmeans.
[Note: An error in units conversion caused the experiments in
this paper to over-estimate the impact of radiation. See "How
much Memory Radiation Protection do Onboard Machine Learning
Algorithms Require?" above for updated results.]
Constrained Clustering: Advances in Algorithms, Theory,
and Applications.
Edited by Sugato Basu, Ian Davidson, and Kiri L. Wagstaff, Chapman & Hall/CRC Press, 2008.
Papers
Multiple-Instance Regression with Structured Data
(pdf, 10 pages, 307K).
Kiri L. Wagstaff, Terran Lane, and Alex Roper. Proceedings of the 4th International Workshop on Mining
Complex Data, December 2008.
Automatic Code Generation for Instrument Flight Software
(pdf, 8 pages, 289K).
Kiri L. Wagstaff, Edward Benowitz, DJ Byrne, Ken Peters, and
Garth Watney. Proceedings of the 9th International Symposium on
Artificial Intelligence, Robotics, and Automation in Space,
February 2008.
Onboard Detection of Active Canadian Sulfur Springs
(pdf, 8 pages, 235K).
Rebecca Castano, Kiri Wagstaff, Damhnait Gleeson, Robert Pappalardo, Steve Chien, Daniel Tran, Lucas Scharenbroich, Benyang Tang, Brian Bue, and Thomas Doggett. Proceedings of the 9th International Symposium on
Artificial Intelligence, Robotics, and Automation in Space,
February 2008.
Automatic Landmark Identification in Mars Orbital Imagery.
Kiri L. Wagstaff, Julian Panetta, Ron Greeley, Norbert Schorghofer, Melissa Bunte, Mary Pendleton Hoffer, and Adnan Ansar. Eos Transactions of the American Geophysical Union, 89(53), Fall Meeting Supplement, Abstract #P53C-1469, December 2008.
On-board Analysis of Uncalibrated Data for a Spacecraft at Mars
(pdf, 9 pages, 849K).
Rebecca Castano, Kiri L. Wagstaff, Steve Chien, Timothy M. Stough, and Benyang Tang. Proceedings of the 13th International Conference on Knowledge Discovery and Data Mining (KDD), p. 922-930, August 2007.
Surface Change Detection from Mars Orbital Imagery.
Baback Moghaddam, Brian D. Bue, Rebecca Castano, and Kiri L. Wagstaff. Eos Transactions of the American Geophysical Union, 88(52), Fall Meeting Supplement, Abstract #P33A-1020, December 2007.
Measuring Constraint-Set Utility for Partitional Clustering Algorithms
(pdf, 12 pages, 276K).
Ian Davidson, Kiri L. Wagstaff, and Sugato Basu. Proceedings of the Tenth European Conference on Principles and Practice of Knowledge Discovery in Databases (PKDD), p. 115-126, September 2006.
Active Learning with Irrelevant Examples
(pdf, 8 pages, 395K).
Dominic Mazzoni, Kiri L. Wagstaff, and Michael Burl. Proceedings of the Seventeenth European Conference on Machine Learning (ECML), p. 695-702, September 2006.
Learning User Preferences for Sets of Objects
(pdf, 8 pages, 270K).
Marie desJardins, Eric Eaton, and Kiri L. Wagstaff. Proceedings of the Twenty-Third International Conference on
Machine Learning (ICML), p. 273-280, 2006.
Fast, Interactive Analysis of Remote Sensing Data with the HARVIST System
(html).
Michael J. Kocurek, Kiri L. Wagstaff, Dominic Mazzoni, Stephan R. Sain, Lucas Scharenbroich, and Timothy M. Stough. Eos Transactions of the American Geophysical Union, 87(52), Fall Meeting Supplement, Abstract #IN21A-1204, December 2006.
Automatic Plume Detection for Planetary Bodies (html).
Brian Bue, Kiri L. Wagstaff, Rebecca Castano, and Ashley Davies. Eos Transactions of the American Geophysical Union, 87(52), Fall Meeting Supplement, Abstract #IN52A-04, December 2006.
Automating the Detection of Enceladus-Style Plumes (html).
Kiri L. Wagstaff, Becky Castano, Ashley Davies, and Brian Bue.
Division for Planetary Sciences meeting #38,
Bulletin of the American Astronomical Society, Vol. 38, p. 522, October 2006.
When is Constrained Clustering Beneficial, and Why?
(pdf, 2 pages, 71K).
Kiri L. Wagstaff, Sugato Basu, and Ian Davidson.
AAAI Member Abstracts and Posters,
Proceedings of the Twenty-first National Conference on Artificial Intelligence (AAAI), July 2006.
Detecting Dust Storms and Water Ice Clouds Onboard THEMIS (html -- note incorrect citation; should be Spring 2006, not Fall 2007 meeting).
Kiri L. Wagstaff, Joshua L. Bandfield, Rebecca Castano, Steve Chien,
Michael D. Smith, and Timothy M. Stough. Spring AGU Joint Assembly, Revolutionary Space Exploration
Concepts using Onboard Computing, May 2006.
Evidence of Life or Not? Scale Sensitive Analysis of
Stromatolite Biogenicity.
Kiri L. Wagstaff and Frank A. Corsetti. Third Annual Southern California Geobiology Symposium, April 2006.
An Onboard Data Analysis Method to Track the Seasonal Polar Caps on Mars
(pdf, 8 pages, 2.8M).
Kiri L. Wagstaff, Rebecca Castano, Steve Chien, Anton B. Ivanov, and Timothy N. Titus. Proceedings of the International Symposium on Artificial Intelligence, Robotics, and Automation in Space, 2005.
Validating Rover Image Prioritizations
(pdf, 8 pages, 354K).
Rebecca Castano, Kiri Wagstaff, Lin Song, and Robert C. Anderson. The Interplanetary Network Progress Report, vol. 42-160, 2005.
Current Results from a Rover Science Data Analysis System
(pdf, 10 pages, 696K).
Rebecca Castano, Michele Judd, Tara Estlin, Robert C. Anderson, Daniel Gaines, Andres Castano, Ben Bornstein, Tim Stough, and Kiri Wagstaff. Proceedings of the 2005 IEEE Aerospace Conference, 2005.
Clustering with Missing Values: No Imputation Required
(pdf, 10 pages, 244K).
Kiri Wagstaff. Classification, Clustering, and Data Mining Applications
(Proceedings of the Meeting of the International Federation of
Classification Societies), p. 649-658, 2004.
Mining GPS Traces for Map Refinement
(pdf, 29 pages, 837K).
Stefan Schroedl, Kiri Wagstaff, Seth Rogers, Pat Langley, and
Christopher Wilson. Data Mining and Knowledge Discovery, vol. 9, issue 1, p. 59-87, 2004.
Generalized Clustering,
Supervised Learning, and Data Assignment (ps, 6 pages, 238K).
Annaka Kalton, Pat Langley, Kiri Wagstaff, and Jungsoon Yoo. Proceedings of the Sixth International Conference on Knowledge
Discovery and Data Mining (KDD), p. 299-304, 2001.
Multi-document Summarization via Information Extraction
(pdf, 7 pages, 49K).
Mike White, Tanya Korelsky, Claire Cardie, Vincent Ng,
David Pierce, and Kiri Wagstaff. Proceedings of the Human Language Technology (HLT) Conference,
2001.
Alpha Seeding for Support Vector Machines (ps, 5 pages, 205K).
Dennis DeCoste and Kiri Wagstaff. Proceedings of the Sixth International Conference on Knowledge
Discovery and Data Mining (KDD), p. 345-349, 2000.
Clustering with Instance-level Constraints
(pdf, 1 page, 11K).
Kiri Wagstaff and Claire Cardie.
AAAI-2000 Student Session;
Proceedings of the Seventeenth National Conference on Artificial Intelligence, p. 1097, 2000.
1999
Papers
Noun Phrase Coreference as Clustering (ps, 8 pages, 219K).
Claire Cardie and Kiri Wagstaff. Proceedings of the Joint SIGDAT Conference on Empirical Methods in
Natural Language Processing and Very Large Corpora (EMNLP), p. 82-89, 1999.
Extravehicular Activity Suit
Systems Design: How to Walk, Talk, and Breathe on Mars
(pdf, 22 pages, 485K).
George Barton, Akio Cox, Lauren DeFlores, Alison Diehl,
Ari Garber, Randall Goldsmith, Joel Haenlein, Alex Iglecia,
Kerri Kusza, Brett Lee, Saemi Mathews, Jonathan Mitchell,
Abigail Ross, Rachel Sanchez, Stephen Shannon,
Sri Priya Sundararajan, Mike Valdepenas, and Kiri Wagstaff.
HEDS-UP, 1999.