308kb PPT - SCEC

erminerebelΤεχνίτη Νοημοσύνη και Ρομποτική

15 Νοε 2013 (πριν από 3 χρόνια και 9 μήνες)

107 εμφανίσεις

Loom/PowerLoom Group

Semantic Markup:

Current Trends for Information
Management

Hans Chalupsky


with contributions from


Tom Russ and Yolanda Gil


USC Information Sciences Institute


Knowledge

Loom/PowerLoom Group

Information Management Problem


EarthScope will generate


Huge amounts of raw data


Large amounts of interpreted data


Many refined data products (for end users)


New computational components


Computational models


Simulation codes


Analysis codes


Tools



Information Management Problem


How can all this be organized and integrated


To maximize
scientific progress


To maximize
accessibility

from inside and outside the community


Without central control

(distributed, heterogeneous, collaborative)


With
minimal burden

on scientists

Loom/PowerLoom Group

Information Management Technology Soup

Globus

XML

RDF

RDF(S)

XSD

PICS

DAML

OIL

DAML+OIL

IDL

DAML
-
S

WSDL

RSS

KIF

MELD

GRID

SRB

XLink

XPath

?

?

?

?

?

?

?

XTM

XQUERY

SMIL

XSLT

CORBA

Fissures

SOAP

IMTS:

Loom/PowerLoom Group

Some Very Useful Ingredients


Distributed object frameworks


Web services (WSDL), CORBA, Java RMI,…


Make everything a service, optimize for flexibility



Grid technologies


Coordinated resource sharing


Data grids


Computational grids



Solve very important infrastructure problems


Publishing of service APIs, discovery and invocation of services


Uniform access to heterogeneous, federated data collections


Access to distributed computing resources


Large scale, high
-
performance, persistence, security, fault tolerance, …

Loom/PowerLoom Group

Is That All We Need for Good E
-
Science Soup?


Maybe


some quotes from previous talks:


“E
-
science is just a bunch of XML”


“all you need are lots of metadata acquisition interfaces”


“data management is solved”


“information management is basically solved”



How close are we to an information management solution?


Integrating

all these pieces into useful, distributed earth science
applications is still a
very difficult

task


Will earth scientists really be able to do this themselves?


Will they still have time to do science?



Only the future will tell, but…

Loom/PowerLoom Group

Need Another Ingredient:
Semantics


Previous technologies focus on
syntactic interoperability


Standardized protocols, ports (TCP/IP, FTP, HTTP, SOAP, etc.)


Standardized data formats (properly lined up bits, arrays,…)


Standardized description languages (XML, WSDL, …)


Standardized remote invocation mechanisms (RMI, CORBA, …)


Interoperation possible as long as exchanged messages/data are
syntactically correct

(something a compiler could check)



Claim: Full E
-
Science needs
semantic interoperability


Computer needs to
understand

what the bits
mean

(e.g., is this
number a
wavelength

or a
frequency

of what
wave

in what
context
)


Facilitates stronger semantic integrity or compatibility checks


Facilitates integration (automatic transformations and translations)


Facilitates more informed search engines


Loom/PowerLoom Group

Semantic Interoperability Anecdote

Helicopter Mission Planning

Moksaf

Route
Planner

TEAMCORE

Teamwork Agent

Ariadne

SAM Site
Finder

Quickset

Mission Editor

ModSAF

Battlefield
Simulator

CoABS Grid

Generates route in
9m segments

Helicopters won’t fly
“I’m already there”

Sends moves to helicopters
segment by segments

Loom/PowerLoom Group

Syntactic vs. Semantic Information

Move(object, toLocation)

ModSAF API operation:

MovableObject

Location

Syntactic information:

type

type

Semantic information:

Precondition:


If type(object) = Helicopter


then distance(location(object),toLocation) > 100m

Postcondition:


location(object) = toLocation

Loom/PowerLoom Group

Knowledge Representation & Reasoning


Subdiscipline of Artificial Intelligence


Studies formal theories and languages to explicitly
represent

semantic

information and relationships


Types of entities and their relationships


Time, space, action, causality


Rules, propositional attitudes, …



Reasoning

or inference makes implicit semantics explicit


Logical deduction, abduction, subsumption, classification


Probabilistic inference, planning, constraint satisfaction, etc.



KR&R systems implement particular representation &
reasoning formalisms


Loom, PowerLoom, Cyc, Ontolingua, Protégé, SNePS, Classic, …

Loom/PowerLoom Group

Current Trends: The “Semantic Web”


Why should you care about KR&R?


Discipline has been around for over 40 years


What does it have to do with earth science?



Hot new initiative called the “Semantic Web”


Tries to solve problems that are related to E
-
science


Does so by using
KR&R on the Web

Loom/PowerLoom Group

The “First” Web


Uniform access to huge number of distributed documents


Via standard protocols such as HTTP


Via tools such as Web browsers



HTML markup (Hyper
-
text Markup Language)


Markup specifies display properties, for example:

<strong> this <em> is </em> important </strong>



this
is
important


Hyperlinks to other documents



Limitations


Information primarily in natural language rendered for human consumption


Computers do not (yet) understand
meaning

of documents


Limited automated exploitation of information on the Web


Uninformed keyword
-
based search engines, etc.

Loom/PowerLoom Group

Earth Science Analogy


Uniform access to data (maybe)


standardized data formats (HDF, CDF, NetCDF, etc.)


data centers, data grids



Limited non
-
uniform markup with metadata



Limitations


Meaning of data sets primarily understandable by humans,
documented in README files


Difficult manual navigation, interpretation, integration, etc.


Uninformed keyword
-
based searches (but digital libraries can help)


Loom/PowerLoom Group

The New “Semantic Web”

W3C’s Tim Berners
-
Lee: “Weaving the Web”:


“I have a dream for the Web… and it has two parts:”



The
first Web

enables

communication between people



The Web shows how computers and networks enable the information space while
getting out of the way


The
new Web

will
bring computers into the

action


Step 1
--

Describe:

put data on the Web in
machine
-
understandable

form
--

a

Semantic Web


RDF (based on XML)


Master list of terms used in a document (RDF schema)


Each document mixes global standards and local agreed
-
upon terms (namespaces)


Step 2
--

Infer and reason:

apply logical inference


Operate on partial understanding


Answering why


Heuristics

Loom/PowerLoom Group

What The New “Semantic Web” Enables


Access to documents by
content

instead of keywords


Semantically informed search engines: “Find documents that suggest
fixes for problem P with the Linux X server”



Sophisticated Web service mechanisms


Discovery of services by content
: “find travel services that sell tickets
for airline X between cities Y and Z and accept credit card C”


Automatic
composition, selection and interoperation

of services
based on a high
-
level objective: “plan a trip to city X, rent a car and a
hotel for the duration of stay”


Automatic
service execution monitoring
: monitor status of a request
or adapt to changed situations



Many other things we just haven’t thought of yet


Loom/PowerLoom Group

Earth Science Analogy


Access to data sets by
content

instead of keywords


Semantically informed search engines: “Find time series data for
areas which generate aerosol sulphates”


Semantic search will also return data for volcanic sites, since
volcanoes produce sulphates”



Sophisticated access to earth science services and tools


Discovery of services by content
: “find codes that do anelastic wave
propagation based on model X”


Semi
-
automatic
composition, selection and interoperation

of data and
services: “use data set X as input for model Y and pipe its output into
model Z”, check for constraints, apply necessary transformations


“Plug and play” science, a science “collaboratory”



Many other things we just haven’t thought of yet


Loom/PowerLoom Group

The Layer Cake [TBL,XML2000]

Loom/PowerLoom Group

Semantic Markup Languages


Semantic Markup languages are KR languages for the Web


All of them are based on XML or have an XML version


Active field of investigation, many different choices



RDF, DAML, DAML+OIL


KR languages with different levels of expressivity and inference


DAML
-
S


Semantic service description language


Topic Maps, XTM


Multi
-
dimensional indicies, digitial libraries


Specialized science markup languages:


ESML, GML

Loom/PowerLoom Group

Some Earth Science Examples


Domain: Probabilistic Seismic Hazard Analysis (PSH)


Pathway 1 in SCEC
-
ITR collaboration



Examples based on Ned Field’s work


Java
-
applets for different ground motion attenuation models


Reimplementation of PSH code in Java


Proposed object
-
oriented framework


Flexible scheme for implementing model parameters




Loom/PowerLoom Group

Web
-
based Applets for Ground Motion Attenuation

Loom/PowerLoom Group

Similar Programs Use Different Inputs



Different names for


similar parameters


Rjb


Rrup



Different ways of


specifying site


parameters


S
-
wave velocity


Geology



Different sets of


options for fault types


Loom/PowerLoom Group

Parameter Differences Between Models

* Also has “basin depth” parameter

*

*

*

Model

Site Type

Fault Type

A&S 97

rock, deep soil

strike slip, reverse,
reverse/oblique,
reverse/hanging wall

BJF 97

S
-
wave velocity

strike slip, reverse slip,
unknown/other

Campbell 97

hard rock, soft rock,
firm soil

strike slip, normal, other

Sadigh 97

soft rock, deep soil

reverse/thrust, other

Steidl 2000

M, T, Q, Q1, Q2, Qy,
Qo, Orig

reverse/thrust, other

Compare Attentuation

Quarternary, Tertiary,
Mesozoic

strike slip, reverse

Loom/PowerLoom Group

Parameter Modeling


Ned Field’s Java Model for Parameters


Flexible representation


Specifies interface information


Needed for data interchange



KR&R Augmentation of Parameter Descriptions


Describe meaning


Add constraints to support reasoning


Add inference and functional definitions

Loom/PowerLoom Group

Parameter Modeling Example

Compression_Wave

disjoint

Model 1

Model 2

Model 3

Name

Vs30

Vs

Site Type

(Data)Type

Number

Integer

String

Value

310.0

31000

“Hard Rock”

Constraint

200 < x < 400

20000 < x < 40000

One of …

Units

“m/s”

“cm/s”



Java

model

Vs30


isa Mean_Value_Parameter


isa Velocity_Parameter


of Shear_Wave



with wavelength = 30 meters

KR augmentation

Loom/PowerLoom Group

Parameter Modeling Example

Constraints:


“Hard Rock”: 285 m/s ≤ mean Shear_Wave velocity ≤ 325

m/s


Inferences:


Vs30 of Model 1 is compatible with “Hard Rock” of Model 3


“Hard Rock” in Model 3 maps onto Vs30=305m/s in Model 1

Note: these numbers are made up!

Model 1

Model 3

Name

Vs30

Site Type

(Data)Type

Number

String

Value

310.0

“Hard Rock”

Loom/PowerLoom Group

Input Validation and Error Advice

Assume that this model
saturates at magnitudes ≥ 7.0

Warning: The magnitude of 8.2
exceeds the limits of this model’s
magnitude parameter (7.0).

For best results, choose a
magnitude less than or equal to 7.0

An attempt to enter a
magnitude of 8.2 might
produce this warning:

8.2

Warning: The magnitude of 8.2 exceeds the limits of this
model’s magnitude parameter (7.0).

Options:

(1) Accept possibly inaccurate results

(2) Choose a magnitude less than or equal to 7.0

(3)
Use a different model




A&S 97 with magnitude 8.2 and soil type = “rock”




Steidl 2000 with magnitude 8.2, site type = “Q”

With KR&R

Loom/PowerLoom Group

Ontologies


Ontologies provide
terminological scaffolding

for
representing knowledge


Formalizes terminology and conceptual abstractions for a
particular domain:


Terms


Fault, strike
-
slip
-
fault, velocity, frequency
,
acceleration, soil
-
type, rock



Definitions


cold

=
(10
°

to 39
°

F)


Relationships


Shear
-
wave

and
compression
-
wave

are disjoint



Rough approximation:


ONTOLOGY + FACTS = KNOWLEDGE BASE

Loom/PowerLoom Group

Conclusions


There is more to semantics than metadata



Semantic Web/semantic markup still very active research


Technology landscape still very much in flux


Currently testing some of these ideas within SCEC
-
ITR


Stay tuned for transferable results



Ontology/terminology building can start now


We are starting within SCEC
-
ITR


Doing this right will be a long
-
term effort


UGLS?


Loom/PowerLoom Group

Background


Loom/PowerLoom Group

Classification Reasoning in Loom & PowerLoom


Given:
a new concept and
its definition


Classifier determines:

where concept belongs in
hierarchy by reasoning with
its definition


Classifier performs reasoning services that are essential to
ontology
management
:


Concepts automatically positioned within taxonomy


Checks consistency of local and inherited properties on a concept


Finds superset/subset/equivalence relations between concepts

1.





2.

Loom/PowerLoom Group

A Simple Ontology

animal

mammal

dog

sick animal

rabies

disease

has

“A dog is

a mammal”

“A sick animal
has a disease”

“rabies is
a disease”

Loom/PowerLoom Group

Defining a “rabid dog”

animal

mammal

dog

sick animal

rabies

disease

has

rabid dog

has

Loom/PowerLoom Group

Classifier Concludes “sick animal”

animal

mammal

dog

sick animal

rabies

disease

has

has

rabid dog

Loom/PowerLoom Group

Defining “rabid animal”

animal

mammal

dog

sick animal

rabies

disease

has

has

rabid dog

rabid animal

has

Loom/PowerLoom Group

Classifier Places Concept at Correct Place

animal

mammal

dog

sick animal

rabies

disease

has

has

rabid dog

rabid animal

has

Loom/PowerLoom Group

The End