DataDriven Discovery through eScience Technologies

1
Data-Driven Discovery through e-Science
Technologies

• Kirk D. Borne

George Mason University and QSS Group Inc.,
NASA-Goddard Space Flight Center kborne_at_gmu.edu
or kirk.borne_at_gsfc.nasa.gov http//rings.gsfc.na
sa.gov/nvo_datamining.html
7/19/2006
2
Looking to NASAs Future -- Constellations of
Spacecraft Sensor Webs

• Working Definition of a Constellation or a Sensor
  Web (Sensor Network) Spatially distributed
  network of individual vehicles, or assets, acting
  collaboratively as a single collective unit,
  exhibiting a common system-wide capability.
• Constellation and Sensor Web operations include
  homogeneous sensor clusters, heterogeneous
  distributed sensors, multiple missions (space
  and ground).
• Constellation missions and Sensor Webs/Networks
  will provide opportunities for
• Application of on-board data processing
  functions.
• Application of parallel data processing
  techniques.
• Application of autonomous intelligent systems for
  mission operations.
• Application of client-server architecture for
  inter-sensor communications, for
  constellation/network operations, and for active
  mission participation in a Virtual Observatory.
• Application of interoperable systems techniques
  within the constellation, and with other sensors
  (space-based, ground-based, and/or virtual).
• Application of data reduction (filtering) and
  code-shipping techniques.
• Application of data mining for target of
  opportunity, event, anomaly detection.

3
(No Transcript)
4
Data Challenges for Space Mission Sciencecraft
of the Future

• Data and Information Explosion massive volumes
  from single missions and from spacecraft suites,
  constellations, and armadas sciencecraft
• Data Discovery within distributed data systems
  for science decision support
• Transparent Access to Data across heterogeneous
  mission environments
• Interoperability of systems, metadata, data,
  information, and knowledge
• New Information Technology Infusion across
  multiple distributed systems
• Data Fusion and Information Integration from
  multiple sources (spacecraft / missions /
  sciencecraft) across various scales and
  modalities
• Information and Knowledge Sharing among
  cooperating science nodes
• Intelligence within the sensor and measurement
  and data systems
• Knowledge Representation and Ontology
  reconciliation across discipline-specific
  instruments
• Semantic Knowledge Extraction and Retrieval
  from multiple mission science data streams
• Knowledge Management, Sharing, and Reuse
  lessons learned remembered!

5
The New Face of Science 1

• Big Data (usually distributed across systems)
• High-Energy Particle Physics
• Astronomy, Planetary, and Space Physics
• Earth Observing System (Remote Sensing)
• Human Genome and Bioinformatics
• Numerical Simulations of any kind
• Digital Libraries (electronic publication
  repositories)
• e-Science
• Built on Web Services (e-Gov, e-Biz) paradigm
• Distributed heterogeneous data are the norm
• Data integration across projects institutions
• One-stop shopping The right data, right now.

6
The New Face of Science 2

• Data and data systems are central to science.
• Databases enable scientific discovery
• Data Handling and Archiving (management of
  massive data resources)
• Data Discovery (finding data wherever they exist)
• Data Access (HTTP-Database interfaces)
• Data/Metadata Browsing (serendipity)
• Data Sharing and Reuse (within project teams and
  by other missions and programs scientific
  validation)
• Data Fusion (across multiple modalities
  domains)
• Data Integration (from multiple sources)
• Knowledge Sharing Reuse (through ontologies and
  semantic representation in knowledgebases)
• Data Mining (KDD Knowledge Discovery in
  Databases)

7
e-Science

• Key Technologies
• Data Mining
• KDD Knowledge Discovery in Databases
• Machine Learning
• Distributed Data Discovery Access
• Vx0s Virtual any science Observatories
• Web Services
• Grid Computing
• The Semantic Web (ontologies)
• Key Benefits
• Provides seamless uniform discovery, access,
  mining, and analysis of distributed heterogeneous
  data sources ...
• ... here, there, and beyond.
• Find the right data, right now
• Enables ... Data Information integration and
  fusion ...
• ... across multiple distributed heterogeneous
  data collections ...
• ... to enable scientific knowledge discovery ...
• ... and decision support ...
• ... (with minimal human assistance, or
  autonomously).
• Provides intelligence within the data system.

8
Modeling of a business process, using an
event-driven process chain.
Science data ordered
Intelligent Database event-driven
valid request
invalid request
check data availability
data not available
data available
Steps in this process chain are business-related,
but you can easily transform these steps into
scientific decision support within a space
mission data system.
Design and implement data processing plan
Collect new data from sensors
Science data shipped
generate data products
Reference EPML Event-driven Process chain
Markup Language
product finished
Any more Requests?
http//xml.coverpages.org/ni2003-11-21-a.html
Order completed
Happy User!
9
Data Mining is the killer app for e-Science

• Data Mining is Knowledge Discovery in Databases
  (KDD).
• Data Mining is defined as an information
  extraction activity whose goal is to discover
  hidden facts contained in (large) databases.
• Data Mining is the application of Machine
  Learning to large databases.
• Machine Learning is defined as the application
  of computer algorithms that improve automatically
  through experience.

10
Data Mining Technique Example Decision Tree
Rule Induction

• Decision Tree (DT) Construction is a form of data
  mining called Rule Induction.
• The DT algorithm learns the rules from the
  database of historical records.
• Picks predictors and their splitting values on
  the basis of an information gain metric
• The difference between the amount of information
  that is needed to make the correct prediction
  both before and after the split has been made.
• If the amount of information required is much
  lower after the split is made, then the split is
  said to have decreased the disorder (entropy) of
  the original data. This is good (i.e., the
  more ordered the data, then the more certain is
  our final classification.)
• Similar to the game 20 Questions good
  questions provide good DT splits. For example
  adult asks Is it alive?, but child asks Is it
  my daddy?.
• After the rules are learned (induction), they are
  then applied to new events (new records, or new
  data entries, ) to make predictions on unseen
  data. This can be applied in real-time in deep
  space missions.
• Some databases have automated rule-learning
  algorithms built-in. These are called Inductive
  Databases, and they are data-driven.
• Reference http//www.mli.gmu.edu/projects/idb.ht
  ml

11
Data Mining Technique Example Bayes Inference
Engine

• The application of Bayes Theorem is a form of
  Inference (the act or process of deriving a
  conclusion based solely on what one already
  knows).
• The Bayes Inference Engine (BIE) learns the
  likelihood of possible hypotheses (models, or
  classifications) being correct, for a given set
  of observational measurements, based upon the
  database of historical records.
• The historical knowledgebase for a space mission
  is used to populate the data history (including
  known outcomes from prior experience) with
  particular measurement values of the observed
  features that correspond to particular events.
• As new data arrive, the Bayes Inference Engine
  (BIE) estimates the most likely model (outcome,
  class, hypothesis) to describe what the
  spacecraft sees.
• Following the interpretation of the new events,
  which are then properly labeled, these events
  (class labels) and their corresponding measured
  data values (evidence, features) add to the
  knowledgebase for the next application of the
  BIE.
• Bayes Inference data-driven, learns as it goes,
  incremental learning, self-correcting, can test
  multiple hypotheses, unbiased by prior prejudice,
  model-independent inference, yields
  probabilistically ranked predictions, popular
  standard machine learning tool for optimal
  knowledge-based decision-making.
• References
• http//www.astro.cornell.edu/staff/loredo/bayes/
• http//aisrp.nasa.gov/projects/5665da55.html

H C hypothesis or class E F evidence or
feature P(H E) probability of
hypothesis being correct, given the evidence
E. P(C Fk) probability of classification
C, given the set of measured features Fk.
12
Existing Space Science Data Infrastructure

• The Recent Past many independent distributed
  heterogeneous data archives
• Today VxOs Virtual Observatories
• Web Services-enabled e-Science paradigm
  (middleware, standards, protocols)
• Provides seamless uniform access to distributed
  heterogenous data sources
• Find the right data, right now
• One-stop shopping for all of your data needs
• Emerging environment consists of many VxOs for
  example
• NVO National Virtual Observatory (precursor to
  VAO Virtual Astro Obs)
• VSO Virtual Solar Observatory
• VSPO Virtual Space Physics Observatory
• NVAO National Virtual Aeronomy Observatory
• VITMO Virtual Ionospheric, Thermospheric,
  Magnetospheric Observatory
• VHO Virtual Heliospheric Observatory
• VMO Virtual Magnetospheric Observatory
• Standards for data formats, data/metadata
  exchange, data models, registries, Web Services,
  VO queries, query results, semantics
• And of course The Grid, Web Services,
  Semantic Web, etc. ...

13
Sun-Earth Space Environment Rich Source of
Heliophysical Phenomena
14
Multi-point Observations and Models of Space
Plasmas Deliver a Deluge of Physical Measurements
15
Data-Driven Knowledge Discovery
16
Space Weather Example Early Warning System for
Astronauts in Space
CME Coronal Mass Ejection SEP Solar Energetic
Particle
17
Data Mining in Action

• Data Mining facilitates Intelligent Data
  Understanding (IDU).
• Data Mining enables Decision Support and Active
  Control Systems.
• IDU refers to the application of techniques for
  transforming data into scientific understanding.
• Web reference http//is.arc.nasa.gov/IDU/index.h
  tml
• IDU specifically refers to automating the
  following techniques for machine-assisted science
  data analysis
• Data Mining (e.g., http//is.arc.nasa.gov/IDU/tas
  ks/NVODDM.html)
• Knowledge Discovery
• Machine Learning

18
Case Study - Mars Rovers
19
e-Science Data Mining Applications on Mars Rover
(1)

• Rove around the surface of Mars and take samples
  of rocks (mass spectroscopy a data histogram)
• Supervised Learning (search for rocks with known
  compositions)
• Unsupervised Learning (discover what types of
  rocks are present, without preconceived biases)
• Association Mining (find unusual associations)
• Clustering (find the set of unique classes of
  rocks)
• Classification (assign rocks to known classes)
• Deviation/Outlier Detection (one-of-kind
  interesting?)

20
e-Science Data Mining Applications on Mars Rover
(2)

• On-board Intelligent Data Understanding
  Decision Support Systems (Fuzzy Logic Decision
  Trees Cased-Based Reasoning ) Science Goal
  Monitoring
• stay here and do more or else move on to
  another rock
• send results to Earth immediately or send
  results later
• Learn as it goes (Machine Learning Neural Nets)
• Relate the results to other factors, such as dust
  storms (XML Information Retrieval Information
  Fusion with other data from orbiting satellite
  mother ship)
• Predict where to go in order to find interesting
  rocks (Logistic Regression Case-Based Reasoning)

21
Mars Rover as an e-Science Data System

• Decisions are based on data mined, prior
  experience, new knowledge, and the set of learned
  rules.
• Rover acts autonomously, without human
  intervention, in Deep Space environment.
• Actions are driven by mining actionable data from
  all sensors.

http//www.samsi.info/200506/astro/presentations/t
ut1loredo-7.pdf
22
Autonomous Mineral Detectors for Mars Rovers and
Landers

• NASA / AISRP PI Martha Gilmore, Wesleyan
  University

Objective Design and develop software to
enable rovers to autonomously analyze spectral
data and identify data indicating geologically
important signatures. Motivation Both rover
and orbital missions can collect more data than
can be returned due to downlink restrictions.
Results Software is designed to allow onboard
processing of Vis/NIR spectra to identify and
select spectra that contain minerals of geologic
interest autonomously.
Non-carbonates
Carbonates
Credit M. Gilmore
23
A Neural Map View of Planetary Spectral Images
for Precision Data Mining and Rapid Resource
Identification

• NASA / AISRP PI Erzsébet Merényi, Rice
  University

Uses advanced variants of the self-organized
machine learning paradigm Self-Organizing
Map, applied to spectral imagery. They detected
orthopyroxene and clinopyroxene dominated mineral
subclasses within a rare undifferentiated mineral
type nicknamed "black rock" by geologists. SOM
by eye! (SOM self-organizing map)
Credit E.Merenyi
24
Application of Machine Learning Technology to
Martian Geology
NASA / AISRP PI Ruye Wang, Harvey Mudd College

• Machine Learning algorithms have been applied to
  the analysis of Themis (Thermal Emission Imaging
  System) image data of Mars, for the purpose of
  studying mountain ranges on Mars (the Thaumasia
  Highlands and Corprates rise).
• Specifically, various clustering and
  classification algorithms (e.g., K-means,
  competitive neural network, support vector
  machine, Independent Components Analysis) have
  been applied to the Themis image data covering
  certain areas in the Thaumasia highlands.
• Objectives
• Develop an intelligent system for robust
  detection and accurate classification in
  multispectral remote sensing image data
• Demonstrate system in context of Martian geology
  application

Credit R. Wang
25
Some e-Science Projects

• International Virtual Observatory Alliance
• The Thinking Telescope Project
• Open Science Grid (OSG)
• Biomedical Informatics Research Network (BIRN)
• Network for Earthquake Engineering Simulation
  (NEES)
• PlanetLab computer science testbed
• Earth System Grid (ESG)
• Department of Energy's Fusion Collaboratory
• UK myGrid project
• Enabling Grids for eScience in Europe (EGEE)
• e-Framework for Education and Research
• Global Earth Observation System of Systems (GEOSS)

• http//www.ivoa.net/
• http//www.thinkingtelescopes.lanl.gov/
• http//www.opensciencegrid.org/
• http//www.nbirn.net/
• http//www.nees.org/
• http//www.planet-lab.org/
• https//www.earthsystemgrid.org/
• http//www.fusiongrid.org/
• http//www.mygrid.org.uk/
• http//public.eu-egee.org/
• http//www.e-framework.org/
• http//www.epa.gov/geoss/

26
Astronomy Example The Thinking Telescope
Machine Learning Applications Automated Feature
Extraction Real-time identification of
artifacts and transients in direct and difference
images. Classifiers Automated classification of
celestial objects based on temporal and spectral
properties. Anomaly Detection Real-time
recognition of important deviations from normal
behavior for persistent sources.
Credit http//www.thinkingtelescopes.lanl.gov/
27
Sample e-Science Data Mining Use Cases

• Discover data stored in distributed heterogeneous
  systems.
• Search huge databases for trends and correlations
  in high-dimensional parameter spaces identify
  new properties or new classes of scientific
  objects.
• Discover new linkages associations among data
  parameters.
• Search for rare, one-of-a-kind, and exotic
  objects in huge databases.
• Identify repeating patterns of temporal
  variations from millions or billions of
  observations.
• Identify moving objects in huge image databases.
• Identify parameter glitches / anomalies /
  deviations either in static databases (e.g.,
  archives) or in dynamic data (e.g., science /
  instrumental / engineering data streams).
• Find clusters, nearest neighbors, outliers,
  and/or zones of avoidance in the distribution of
  objects or other observables in arbitrary
  parameter spaces.
• Serendipitously explore huge scientific databases
  through access to distributed, autonomous,
  federated, heterogeneous, multi-experiment,
  multi-mission science data systems.

28
Applications of e-Science Data Mining Techniques
to the Space Mission I.T. Data System Environment

• Archival research applications
• cross-links between archives active mission
  data will offer improved data analysis,
  calibration, anomaly detection, and scientific
  discovery with active missions.
• Decision support for selecting interesting
  targets for observation.
• Identification of interesting events for rapid
  followup observation planning.
• Real-time on-board decision-support functions,
  such as
• the rapid analysis of large volumes of
  time-series data (engineering, telemetry, and
  science streams) in order to make decisions
  (about operations, maneuvers, and science
  observations) in deep space without human
  intervention.

7/19/2006