1 Intelligent Information Systems on the Web and in the Aether Tim Finin University of Maryland Baltimore County March 21, 2003 finin/talks/swig03.ppt.

11

Intelligent Intelligent Information Systems Information Systems

on the Web on the Web and in the Aetherand in the Aether

Tim FininTim FininUniversity of Maryland University of Maryland

Baltimore CountyBaltimore CountyMarch 21, 2003

http://umbc.edu/~finin/talks/swig03.ppt

Joint work with Scott Cost, Benjamin Grosof (MIT), Anupam Joshi, Jim Mayfield (JHU), Charles Nicholas, Yun Peng, Yelena Yesha & many

students.

This work was partially supported by DARPA contract F30602-97-1-0215, NSF grants CCR007080 and IIS9875433 and grants from IBM, Fujitsu and HP.

tell

register

tell

register

UMBCUMBCan Honors University in an Honors University in

MarylandMaryland 22

(1) Opening thoughts(1) Opening thoughts(2) Some current SW (2) Some current SW

related research at related research at UMBCUMBC

(4) Parting thoughts(4) Parting thoughts


MarylandMaryland 33

““XML is Lisp's bastard nephew, with XML is Lisp's bastard nephew, with uglier syntax and no semantics. Yet uglier syntax and no semantics. Yet XML is poised to enable the creation XML is poised to enable the creation of a Web of data that dwarfs anything of a Web of data that dwarfs anything since the Library at Alexandria.”since the Library at Alexandria.”

-- Philip Wadler, Et tu XML? The fall of the relational empire, VLDB, Rome, September 2001.


MarylandMaryland 44

““The web has made people smarter. We The web has made people smarter. We need to understand how to use it to make need to understand how to use it to make machines smarter, too.”machines smarter, too.”

-- Michael I. Jordan, paraphrased from a talk at AAAI, July 2002 by Michael Jordan (UC Berkeley)


MarylandMaryland 55

““The Semantic Web will The Semantic Web will globalize globalize KRKR, just as the , just as the WWW globalize hypertextWWW globalize hypertext””

-- Tim Berners-Lee


MarylandMaryland 66

““The multi-agent systems The multi-agent systems paradigm and the web both paradigm and the web both emerged around 1990. One emerged around 1990. One has succeeded beyond has succeeded beyond imagination and the other imagination and the other has not yet made it out of has not yet made it out of the lab.”the lab.”

-- Anonymous, 2001


MarylandMaryland 77

IMHOIMHO The web is like a universal acid, eating The web is like a universal acid, eating

through and consuming everything it through and consuming everything it touches.touches. Web principles and technologies are equally good Web principles and technologies are equally good

for wireless/pervasive computing.for wireless/pervasive computing. The semantic web is our first serious attempt The semantic web is our first serious attempt

to provide semantics for XML sublanguages.to provide semantics for XML sublanguages. The semantic web will provide mechanisms The semantic web will provide mechanisms

for people and machines (agents, programs, for people and machines (agents, programs, CGI scripts) to come together.CGI scripts) to come together. Again: in all kinds of networked environments: Again: in all kinds of networked environments:

wired, wireless, ad hoc, wearable, etc.wired, wireless, ad hoc, wearable, etc. Solving the symbol grounding problem?Solving the symbol grounding problem?


MarylandMaryland 88

Some UMBC applicationsSome UMBC applications

(1)(1) Semantic web and agents Semantic web and agents (ITTalks) (ITTalks)

(2) (2) Ontology mappingOntology mapping

(3)(3) Cooperating personal agentsCooperating personal agents

(4)(4) Learning markup Learning markup

((5) 5) Information retrieval on the SWInformation retrieval on the SW

(6)(6) Modeling trust policiesModeling trust policies

(7)(7) SW meets pervasive computing SW meets pervasive computing

(8)(8) SW and trading agents SW and trading agents


MarylandMaryland 99

(1)(1) ITTALKS ITTALKS

• ITTALKS ITTALKS is a database driven webis a database driven website of IT related talks at UMBC andsite of IT related talks at UMBC andother institutions. The database other institutions. The database contains information oncontains information on– Seminar eventsSeminar events– People (speakers, hosts, users, …)People (speakers, hosts, users, …)– Places (rooms, institutions, …)Places (rooms, institutions, …)

• Web pages with DAML markup are generatedWeb pages with DAML markup are generated• The DAML markup supports agent-based The DAML markup supports agent-based

services relating to these talks.services relating to these talks. Users get talk announcements based on the Users get talk announcements based on the

interests, locations and schedules.interests, locations and schedules.

http://ittalks.org/


MarylandMaryland 1010



humanview



machineview



ITTALKS ArchitectureITTALKS Architecture

Web server + Java servlets

DAMLreasoning

engine

DAMLreasoning

engine

<daml></daml>

<daml></daml>

<daml></daml>

<daml></daml>

DAML files

Agents

Databases

People

RDBMSRDBMSDB

Email, HTML, SMS, WAP

FIPA ACL, KQML, DAML

SQLHTTP, KQML, DAML, Prolog

MapBlast, CiteSeer,Google, …HTTP

HTTP, WebScraping

WebServices

ApacheTomcat



ITTALKS OntologiesITTALKS Ontologies

We’ve defined and use the following We’ve defined and use the following ontologies, all at ontologies, all at http://daml.umbc.edu/ontologieshttp://daml.umbc.edu/ontologies// calendar-ont.damlcalendar-ont.daml – – calendar and schedule infocalendar and schedule info classification.damlclassification.daml – ACM CCS topics– ACM CCS topics person-ont.damlperson-ont.daml – people and their attributes– people and their attributes place-ont.damlplace-ont.daml – talk locations– talk locations profile-ont.damlprofile-ont.daml – – user modeling infouser modeling info talk-ont.damltalk-ont.daml – – talks infotalks info topic-ont.damltopic-ont.daml – – topics and intereststopics and interests



How does DAML Help?How does DAML Help?

ontologylanguage

usermodels

interoplanguage

agentcommunication

servicedescriptionlanguage

DAML+OIL provided a uniform language which metMany needs in developing a complex application.



Two Advanced CapabilitiesTwo Advanced Capabilities

I’ll briefly describe two advanced I’ll briefly describe two advanced capabilities facilitated by DAML:capabilities facilitated by DAML: Classifying talk topics and user Classifying talk topics and user

interests using DAML ontologiesinterests using DAML ontologies Using DAML as a communication Using DAML as a communication

language among software agentslanguage among software agents



(2)(2) Ontology mapping Ontology mapping

Techniques for mapping between Techniques for mapping between ontologies are thought to be ontologies are thought to be important for the semantic web.important for the semantic web.

Ontology mapping is a complex Ontology mapping is a complex problem problem

Even in our simple ITTALKS system Even in our simple ITTALKS system we’ve found a need to map between we’ve found a need to map between limited ontologies.limited ontologies.



What are talks about?What are talks about?

Topic hierarchies provide Topic hierarchies provide indexing termsindexing terms ACM CCS ACM CCS topic hierarchytopic hierarchy Open DirectoryOpen Directory

Encoded as DAML ontologiesEncoded as DAML ontologies These allow users to specify interests as well as These allow users to specify interests as well as

browse the database of talks by topicbrowse the database of talks by topic Automatic classification of talks (based on title Automatic classification of talks (based on title

and abstract) and abstract) and users (based on his web and users (based on his web pages, CV, papers, etc.)pages, CV, papers, etc.)

Discovery of mapping rules between CCS to OD Discovery of mapping rules between CCS to OD ontologies using IR techniquesontologies using IR techniques



Classifying TalksClassifying Talks

ACM CCS Ontology

Training corpus

CMU Bow

statisticaltext analysis

tools

CMU Bow


tools

ACM CCSclassifier

Now is the time for all good men to come to the aid of the country. Now is the time for

topics

e.g.: ACMCCS

e.g.:5K ACMabstracts

Topics Ontology

uses

uses



Mapping between topic Mapping between topic ontologiesontologies

Topic ontology T1

Training corpus T1

CMU Bow


tools

CMU Bow


tools

T1T2mapper

{(t2:bar, 0.8), (t2:qux, 0.7), …}

Topic ontology T2

Training corpus T2

T1

T2

t1:foo



Interactive topic ontology mapperInteractive topic ontology mapper

Users create maps Users create maps between ontologies between ontologies with URIs to text with URIs to text describing classes describing classes & properties.& properties.

Automates mapping Automates mapping process, taking into process, taking into account hierarchical account hierarchical relationships and relationships and user-specified user-specified landmark landmark mappings.mappings.

Text classification Text classification used to compute used to compute similarities similarities between topics.between topics.

A probabilistic A probabilistic approach used to approach used to combine combine hierarchical hierarchical information.information.

Used in XTalks to enable mappings betweenAlternative topic ontologies in DAML+OIL



General Ontology mappingGeneral Ontology mapping

Ontology mapping is a complex problem, and Ontology mapping is a complex problem, and external information is needed to simplify it.external information is needed to simplify it.

We are working to extend this approach as a We are working to extend this approach as a technique for semi-automatic generation of technique for semi-automatic generation of maps from one DAML+OIL ontology to maps from one DAML+OIL ontology to another.another. External information, e.g., text associated with External information, e.g., text associated with

properties and classes, suggests nodes to equate.properties and classes, suggests nodes to equate. Structural information produces contradictions or Structural information produces contradictions or

assumptions which must be made if the nodes are assumptions which must be made if the nodes are equated.equated.



(3)(3) DAML and Agents DAML and Agents

Much multi-agent systems work is grounded in Much multi-agent systems work is grounded in Agent Communication Languages (e.g., KQML, Agent Communication Languages (e.g., KQML, FIPA) and associated software infrastructure such FIPA) and associated software infrastructure such as the DARPA Gridas the DARPA Grid The paradigm has been peer-to-peer message oriented The paradigm has been peer-to-peer message oriented

communication mediated by brokers and facilitators.communication mediated by brokers and facilitators. The DAML program invites different paradigms The DAML program invites different paradigms

which will require some changes in ACLs and which will require some changes in ACLs and their associates software systems.their associates software systems. Agents “publish” beliefs, requests, and other “speech Agents “publish” beliefs, requests, and other “speech

acts” on web pages.acts” on web pages. Agents “discover” what peers have published on the Agents “discover” what peers have published on the

web.web. The software agent research community is very The software agent research community is very

interested in the semantic web and DAMLinterested in the semantic web and DAML


MarylandMaryland 2424MS Outlook

XSBDAML+OIL

Reasoner

ITTALKSagent

Travelagent

Calendaragent

Useragent

BrokerAgent

AgentNameServer

user’s daml profile

mapquest

MS Outlook

ITTALKS app

Common agent infrastructure

FIPA ACL

API

Communicationprotocol

1

8

7 6

5

3

2

4

9

10

11

12

13

17

16

15

14

18



XTalks Personal AgentXTalks Personal Agent

XPA is a configurable “personal agent” which XPA is a configurable “personal agent” which accepts FIPA messages from XTalks and other accepts FIPA messages from XTalks and other instances of XPAs as well as applications, e.g. MS instances of XPAs as well as applications, e.g. MS Outlook.Outlook.

JADE platformJADE platform

PersonalAgent

Infrastructure

PluginManager

UserInterface

UserModelCOM

Bridge

RuleEngine

Interface

XSB

Jess

yajxb

XtalksPlugin

MapquestPlugin

BuddyList

Plugin

ExternalPlugins

External World



Xtalks agentsXtalks agents

XtalksSystem

Xtalks Interface

Xtalks Agent

Mapquest Agent

Personal Agent (1)

FIPA RequestResponse Protocol

FIPA RequestResponse Protocol

Periodicquerying

Scenarios1,2

Personal Agent (2)

Personal Agent (3)

Scenarios3,4

1 – Xtalks Announcement2 – User Agent Solicitation3 – Buddy List4 – Travel Planning

XtalksSystem



(4)(4) Learning Markup Rules Learning Markup Rules Where will Semantic web markup come from?Where will Semantic web markup come from?

Created dynamically as web pages are generated Created dynamically as web pages are generated from databases (e.g., ITTalks).from databases (e.g., ITTalks).

Annotation editors (e.g., SMORE)Annotation editors (e.g., SMORE) NLP Information extraction apps (e.g., Aerotext)NLP Information extraction apps (e.g., Aerotext) Learning markup for specific domains (e.g., ITTalks)Learning markup for specific domains (e.g., ITTalks)

We’re experimenting with algorithms to learn We’re experimenting with algorithms to learn to add DAML markup to text documents from a to add DAML markup to text documents from a training corpus.training corpus.

This, coupled with state-of-the-practice This, coupled with state-of-the-practice Information Extraction techniques, should Information Extraction techniques, should support automatic markup of many documents.support automatic markup of many documents.



LTD LTD DAML Markup Learning DAML Markup Learning SystemSystem

We started with the Stalker algorithm developed at ISI for learning “web scraping” rules.

The ontology structure constrains where objects are sought (e.g., Location has Room, Building, etc.)

Rules are generated with a hill climbing algorithm.

Many rules are generated so co-training techniques can be used to mark up plain documents.

Results on learning to markup talk announcements are promising.



(5)(5) IR and the Semantic Web IR and the Semantic Web

Problem:Problem: How do we do information retrieval over How do we do information retrieval over documents and queries which combine free text documents and queries which combine free text and semantic web markup?and semantic web markup? IR systems and KB systems use different modelsIR systems and KB systems use different models

One Solution: One Solution: (1)(1) index both the text and index both the text and markup and then (2) use existing IR systems to markup and then (2) use existing IR systems to find documents that match queriesfind documents that match queries

Issues: Issues: (1) How do we index markup? (2) When (1) How do we index markup? (2) When and where do we do inferencing over the markup?and where do we do inferencing over the markup?

Applications:Applications: (1) Improved recall and precision (1) Improved recall and precision for IR systems, (2) Retrieving documents for for IR systems, (2) Retrieving documents for question answering.question answering.



Student Event ScenarioStudent Event Scenario

UMBC sends out descriptions of ~50 events a UMBC sends out descriptions of ~50 events a week to students.week to students.

Each student has a “standing query” used to Each student has a “standing query” used to route event messages.route event messages. A student only receives announcements of events A student only receives announcements of events

matching his interests and schedule.matching his interests and schedule. Use LMCO’s AeroText system to automatically Use LMCO’s AeroText system to automatically

add DAML+OIL markup to event descriptions.add DAML+OIL markup to event descriptions. Categorize text announcements into event typesCategorize text announcements into event types Identify key elements and add DAML markupIdentify key elements and add DAML markup

Use JESS to reason over the markup, drawing Use JESS to reason over the markup, drawing ontology supported inferencesontology supported inferences



Event OntologyEvent Ontology

A simple ontology A simple ontology for University for University eventsevents

Includes classes, Includes classes, subclasses, subclasses, properties, etc.properties, etc.

Can include Can include instance data, e.g., instance data, e.g., UMBC, NEC, UMBC, NEC, Fairleigh Fairleigh Dickenson, etcDickenson, etc

... ...

... ...

Event_Name

Event_DateStart_Time

End_TimePlace

Organizer

...

Property

KEY:

PropertyAssociation

Subclass Of

Instance Of

EVENT

DATETIME

TRIP SPORTMOVIESHOW

INDIVIDUAL

ATHLETICS CHESSBASKETBALLBASEBALL

CLASS

TEAM



IR EngineIR Engine We’re experimenting with two IR engines: JHU’s We’re experimenting with two IR engines: JHU’s

Haircut and UMBC’s SIRE, using this technique:Haircut and UMBC’s SIRE, using this technique: Convert DAML markup to RDF triplesConvert DAML markup to RDF triples Infer additional triples which follow from modelInfer additional triples which follow from model

(S,type,O) ^ (0,subclass,O2) => (S,type,O2)(S,type,O) ^ (0,subclass,O2) => (S,type,O2)

Use domain specific rules to infer additional Use domain specific rules to infer additional triplestriples ““for a movie, retrieve genre property from IMDB web site”for a movie, retrieve genre property from IMDB web site”

Generate 7 indexing terms from each (S,P,O) Generate 7 indexing terms from each (S,P,O) tripletripleSPO, SP*, S*O, *PO, S**, *P*, **OSPO, SP*, S*O, *PO, S**, *P*, **O

Index free text and resulting triple termsIndex free text and resulting triple terms



Indexing triples exampleIndexing triples example We process an event description which we We process an event description which we

recognize as a showing of a movie with title recognize as a showing of a movie with title “Spiderman”“Spiderman” _j:00255 a movie; title “Spiderman”._j:00255 a movie; title “Spiderman”.

Domain specific rules get additional information Domain specific rules get additional information from the web, such as the movie’s genrefrom the web, such as the movie’s genre _j:00255 genre “Action”._j:00255 genre “Action”.

That triple yields seven indexing terms:That triple yields seven indexing terms:_:j00255 http://owlir.umbc.edu/event/moviegenre "action“ ._:j00255 http://owlir.umbc.edu/event/moviegenre "action“ .

1.1. j00255.owlir.umbc.edu/event/moviegenre.actionj00255.owlir.umbc.edu/event/moviegenre.action2.2. *.owlir.umbc.edu/event/moviegenre.action*.owlir.umbc.edu/event/moviegenre.action3.3. j00255.*.actionj00255.*.action4.4. j00255.owlir.umbc.edu/event/moviegenre.*j00255.owlir.umbc.edu/event/moviegenre.*5.5. j00255.*.*j00255.*.*6.6. *.owlir.umbc.edu/event/moviegenre.**.owlir.umbc.edu/event/moviegenre.*7.7. **.action**.action



LMCO AeroText

+ Java

IR EngineQueryUser

Interface

Jess

JessText +triples Text +

triples

Event C

ategories

Movie

Sport

Talk

. . .

Trip

OWLIR Framework

Must

OK

Must not

Results User Interface

Final Results Inference on results

Text + triples

Jess

ClassificationInfo

Extraction

Extract triples & reason

Text+DAML

Agents

Text+DAML

Converttriples to

index terms

Text

Extract triples & reason

Converttriples to

index terms

Text

Text

Index

Expand EventDescription

EventDescriptions



DOCUMENTDOCUMENT<DOC><DOCNO>'http://gentoo.cs.umbc.edu/howlir/announcements/charity#charity_001</DOCNO><TEXT>'UMBC Blood Drive!!Office of Student Life launches its annual Blood Drive for the Red Cross on Mon, Nov 20 in the UC Ballroom from 10am - 4pm. </TEXT><TRIPLE>triple(charity_001)( 'http://gentoo.cs.umbc.edu/howlir/announcements/charity#charity_001_place', 'http://daml.umbc.edu/ontologies/event_ont#Building', 'University Center').triple(charity_001)( 'http://gentoo.cs.umbc.edu/howlir/announcements/charity#charity_001', 'http://daml.umbc.edu/ontologies/event_ont#Organizer', 'Office of Student Life').triple(charity_001)( 'http://gentoo.cs.umbc.edu/howlir/announcements/charity#charity_001_date', 'http://daml.umbc.edu/ontologies/event_ont#Day_of_week', 'Monday'). … </TRIPLE></DOC>



QUERYQUERY<Query><required> triple(query_001)( 'http://daml.umbc.edu/ontologies/query#query_001’, 'http://daml.umbc.edu/ontologies/event_ont#Movie_Name' 'Ocean’s Eleven').</required><allowed></allowed><disallowed>triple(query_001)('http://daml.umbc.edu/ontologies/query#query_001’, 'http://daml.umbc.edu/ontologies/event_ont#Organizer’ ‘SEB'). </disallowed></Query>



Results?Results?

Have done initial experiments measuring Have done initial experiments measuring recall and precision over a small recall and precision over a small collection of 1500 event announcements collection of 1500 event announcements and 12 queries.and 12 queries.

CompareCompare Only free textOnly free text Free text + base triples but no inferencingFree text + base triples but no inferencing Free text + triples + inferred triplesFree text + triples + inferred triples

Found improved precision and recall Found improved precision and recall Larger experiments neededLarger experiments needed



(4)(4) Delegation Based Model for Distributed Delegation Based Model for Distributed TrustTrust

We are developing a delegation based We are developing a delegation based model for distributed authorization and model for distributed authorization and trust for use in both wired and wireless trust for use in both wired and wireless scenarios.scenarios.

Trust depends on Trust depends on policies + credentials + delegation actions + policies + credentials + delegation actions +

proofs of permissions and obligations.proofs of permissions and obligations. Agents make speech acts about and Agents make speech acts about and

reason over these properties and relations reason over these properties and relations Grounded in an ontology represented in Grounded in an ontology represented in

DAMLDAML



Rei Policy LanguageRei Policy Language

Rei includes domain independent ontologies to Rei includes domain independent ontologies to represent basic concepts: agents, actions, represent basic concepts: agents, actions, permissions, obligations, prohibitions, permissions, obligations, prohibitions, delegations, etc.delegations, etc.

Users can specify domain dependent ontologies Users can specify domain dependent ontologies to describe classes, properties and roles of people to describe classes, properties and roles of people and agentsand agents

Includes a policy engine that interprets and Includes a policy engine that interprets and reasons over Rei policiesreasons over Rei policies

Has been developed in logic Has been developed in logic (SICStus Prolog)(SICStus Prolog) Very powerful reasoning capabilityVery powerful reasoning capability Relatively straightforward to translate to/from RDF-S Relatively straightforward to translate to/from RDF-S

(using triples)(using triples)



RDF OntologyRDF Ontology



RDF OntologyRDF Ontology

By extending the appropriate classes of By extending the appropriate classes of the ontologythe ontology Actions can be definedActions can be defined Domain dependent information can be addedDomain dependent information can be added Meta policies can be createdMeta policies can be created Policies in RDF can be developed Policies in RDF can be developed

Current ontologies…Current ontologies… Root SchemaRoot Schema Schema for policy objectsSchema for policy objects Schema for speech actsSchema for speech acts Schema for meta policiesSchema for meta policies



Resource ReiRoot

Policy

RulePolicyRule

MetaPolicy

Proposition

Condition

AndCondition

OrCondition

NotCondition

EntityAgent

Object

Action

DomainAction

Creatorrdf

PolicyConditionrdf

RulePolicyrdf

PolicyRuleActorrdf

PolicyRuleConditionrdf

PolicyRuleObjectrdf

ConditionValrdf

AndCondition1rdf

AndCondition2rdf

OrCondition1rdf

OrCondition2rdf

NotCondition1rdf

OwnerOfEntityrdf

Affiliationrdf Location

rdf

ActionNamerdf

OwnerOfActionrdf

Typerdf

TargetObjectrdf

PreConditionrdf

Effectrdf

PolicyObjectRule

Rule

Rule

Rule

Has

'has' used in logic programs, will be removed in rdfs section

Right

Obligation

Dispensation

Prohibition

PolicyObjectActionrdf

PolicyObjectConditionrdf

PolicyObjectCommitmentrdf Rule

From Jon Pastor’s ppt



Action SpeechAct

Delegation

Cancel

Request

Revoke

Senderrdf

Receiverrdf

Contentrdf

ReplyWithrdf

ReplyByrdf

ReplyTordf

Literal

Entity

Agent

PolicyObject

Speech ActSpeech Act




ModalityPrecedence

Overrides

PositiveModality

NegativeModality

PolicyOverrides

RuleOverrides

SubjectConditionrdf

MiscConditionrdf

ActionConditionrdf

Overridderrdf

Overriddenrdf

Policy

MetaPolicy

PolicyRule

Condition

Meta PolicyMeta Policy




We envision pervasive computingWe envision pervasive computingenvironments where devices can environments where devices can see, query and inform many data streams and see, query and inform many data streams and sources in their environment.sources in their environment.

Traditional mobile data management approaches Traditional mobile data management approaches are insufficient to address the new challenges are insufficient to address the new challenges posed by the data-intensive pervasive computing posed by the data-intensive pervasive computing environmentsenvironments

Our approach uses both agents and the semantic Our approach uses both agents and the semantic webweb Cross layer interactionCross layer interaction Proactive, (semi-) autonomous peer mobile devices in the Proactive, (semi-) autonomous peer mobile devices in the

networknetwork Profiles to describe users, devices and data objects using Profiles to describe users, devices and data objects using

BDI principles and a semantically rich language (e.g., BDI principles and a semantically rich language (e.g., DAML)DAML)

Negotiation for services resulting in contracts to model Negotiation for services resulting in contracts to model both transaction and QOS requirements and commitmentsboth transaction and QOS requirements and commitments

(5)(5) P2P Data Management in P2P Data Management in Pervasive Computing EnvironmentsPervasive Computing Environments



SW and Pervasive computingSW and Pervasive computing

One of our research foci is mobile One of our research foci is mobile computingcomputing

… … In particular, we are interested in In particular, we are interested in advanced pervasive and ubiquitous advanced pervasive and ubiquitous computing environmentscomputing environments

… … populated by devices, systems and populated by devices, systems and agents thatagents that are aware of their context, are aware of their context, try to understand what is going on, try to understand what is going on, anticipate our needs and anticipate our needs and actively work to serve us betteractively work to serve us better



Yesterday: Gadgets are EverythingYesterday: Gadgets are Everything

Cool toys…

Too bad they can’t talk to each other…



Today: Connection is EverythingToday: Connection is Everything

Sync. Download. Done.

Configuration? Too much

work…



Tomorrow: Service is EverythingTomorrow: Service is Everything

Thank God! Everything is done for me!



Context Aware PervasiveContext Aware PervasiveComputing EnvironmentComputing Environment

An intelligent context broker acquires context information from devices, An intelligent context broker acquires context information from devices, agents and sensors in its environment and fuses it into a coherent context agents and sensors in its environment and fuses it into a coherent context model, which is then shared with the devices and their agents.model, which is then shared with the devices and their agents.

Context Broker

Semantic Web &

Web Services

(RDF, DAML+OIL & OWL)

Database

(MySQL)

Information Servers

(Exchange Server, iCal,

YahooGroups , etc. )

Teamwork

Engine

Security &

Privacy Policy

Context

ReasoningLunch

Agent

Context -Aware Agents

Personal

Agent

Meeting

Agent

Presentation

Rec. Agent

Emergency

Alert

Agent

FIPA -ACL + RDF/OWL

Contexts in the Intelligent SpacesContexts in the Intelligent Spaces

Smart Tag Sensors

(Radio Frequency Identification)

Environment Sensors

(Xanboo & X10 technology)

Device & Gadget Sensors

(Java Ring, SmartCard etc.)

Contexts in External SourcesContexts in External Sources

Context -Aware Devices

SOAP + RDF/OWL

Etherne tEtherne t…

…

Context Broker

Semantic Web &

Web Services

(RDF, DAML+OIL & OWL)

Database

(MySQL)

Database

(MySQL)

Information Servers

(Exchange Server, iCal,

YahooGroups , etc. )

Teamwork

Engine

Security &

Privacy Policy

Context

ReasoningLunch

Agent

Context -Aware Agents

Personal

Agent

Meeting

Agent

Presentation

Rec. Agent

Emergency

Alert

Agent

FIPA -ACL + RDF/OWL

Contexts in the Intelligent SpacesContexts in the Intelligent Spaces

Smart Tag Sensors


Smart Tag Sensors


Environment Sensors

(Xanboo & X10 technology)

Device & Gadget Sensors

(Java Ring, SmartCard etc.)

Contexts in External SourcesContexts in External Sources

Context -Aware Devices

SOAP + RDF/OWL

Etherne tEtherne t…

…



An Intelligent Meeting RoomAn Intelligent Meeting Room

Alice enters a conference room

The broker detects Alice’s presence

B B

The broker buildsthe context model

Web

The broker knows Alice’s role and

intention

+

The broker informsthe subscribed agents

B A

The projector agentsetup the presentation



Why Semantic Web & OWLWhy Semantic Web & OWL

Everybody can say something about Everybody can say something about everythingeverything Information that describes resources on the Web Information that describes resources on the Web

is is distributeddistributed Information that describes people, places & Information that describes people, places &

objects in a pervasive context-aware system is objects in a pervasive context-aware system is distributed distributed

OWL enables ontology reasoning over OWL enables ontology reasoning over dynamic situational information.dynamic situational information.

OWL/XML is probably the most effective OWL/XML is probably the most effective representation representation for machines to process for machines to process ontology and share knowledge.ontology and share knowledge.



The CoBrA OntologyThe CoBrA Ontology

CoBrA = CoBrA = CoContext ntext BrBroker oker AArchitecturerchitecture It attempts to capture a set of common It attempts to capture a set of common

ontologies for describingontologies for describing People, places, devices, agents, services People, places, devices, agents, services

and non-computing objects in an and non-computing objects in an intelligent meeting environmentintelligent meeting environment

The properties and relationships between The properties and relationships between these entities and the environmentthese entities and the environment

It doesn’t attempt to be “the ontology”It doesn’t attempt to be “the ontology”



The CoBrA OntologyThe CoBrA Ontology



(8)(8) Trading Agents Trading Agents We’ve build an agent-based environment We’ve build an agent-based environment

inspired by TAC, the Trading agent Competitioninspired by TAC, the Trading agent Competition TAC is a forum promoting research into the TAC is a forum promoting research into the

trading agent problem with games run in 00, trading agent problem with games run in 00, 01, 02, 0301, 02, 03 TAC agents operate within a travel shopping scenario, buying TAC agents operate within a travel shopping scenario, buying

and selling goods to best serve their clients and are scored and selling goods to best serve their clients and are scored based on client's preferences for trips assembled, and based on client's preferences for trips assembled, and minimizing cost. minimizing cost.

TAC in organized around a central auction serverTAC in organized around a central auction server Our goal is to open up the system, allowing peer-to-Our goal is to open up the system, allowing peer-to-

peer communication among agents as well various peer communication among agents as well various kinds of mediator, auction, discovery, service provider kinds of mediator, auction, discovery, service provider agents.agents.



A Typical ScenarioA Typical Scenario

BulletinBoard

CA TA

AuctionService

Airline WS Hotel WS

1 2a

3

4

5

6

2b

Market Oversight Agent



TAGA Agents (1)TAGA Agents (1)

……..

Customer Agents

One CA joins the Game every 30 Sec.

Hotel Web Service

EntertainmentWeb Service

Airline Web Service

TA-1 (AAP)

TA-2 (AAP)

TA-4 (JADE)

Find travel arrangements Save $$

Organize travel Maximize profits

sell “goods” Maximize profits



TAGA Agents (2)TAGA Agents (2)

Market OversightAgent

Auction Service Agent

Helps CA find and engage

one or more TA

Operates the auctions markets: English, Dutch, Priceline and Hotwire.

Manage the financial records

Announces the winning TA

Bulletin Board Agent



TAGA in ActionTAGA in Action

TAGA Home Pagehttp://taga.umbc.edu

TAGA on Agentcities net(UMBCTac.agentcities.net)http://www.agentcities.net/

Download the latest TAGA pkg and docshttp://taga.umbc.edu/taga/download/

Create a TAGA game online

TAGA supports heterogeneous agent platform. A FIPA-JADE agent can

interact with a FIPA-AAP agent



(and more … new agent, user login, create game, game history)

Monitor Customer Agents

The TAGA Game ServerThe TAGA Game Server

http://taga.umbc.edu/taga/play/demo.htm

View TAGAgame statusView ACL message traffic

Monitor OpenMarket Auction



TAGA goal and featuresTAGA goal and features

• Goal:Goal: an open test bed for research an open test bed for research on agents, FIPA, and SW in an on agents, FIPA, and SW in an ecommerce environmentecommerce environment

• Features:Features:• Part of the Agentcities networkPart of the Agentcities network• Everything is a FIPA-compliant agentEverything is a FIPA-compliant agent• Supported by OWL ontologiesSupported by OWL ontologies• Agents use RDF and OWL as for their Agents use RDF and OWL as for their

content languagecontent language• DAML-S used for service description and DAML-S used for service description and

discoverydiscovery• New FIPA compliant protocols for various New FIPA compliant protocols for various

kinds of auctionskinds of auctions



Other UMBC SW workOther UMBC SW work

Service composition in pervasive Service composition in pervasive computing environmentscomputing environments

SW for Bluetooth SDP (service discovery SW for Bluetooth SDP (service discovery protocol)protocol)

Javaspaces with DAML+OIL descriptions Javaspaces with DAML+OIL descriptions instead of flat tuplesinstead of flat tuples

Intelligent opportunistic data caching in Intelligent opportunistic data caching in mobile computing environmentsmobile computing environments

Using DAML-S in FIPA’s directory facilitatorUsing DAML-S in FIPA’s directory facilitator



Conclusions and final thoughtsConclusions and final thoughts

SW might be a chance for us to get SW might be a chance for us to get some AI out of the labsome AI out of the lab Solving the symbol grounding Solving the symbol grounding

problemproblem Rethinking agent communicationRethinking agent communication

How do we get thereHow do we get there



The symbol grounding problemThe symbol grounding problem

An argument against human-like AI is An argument against human-like AI is that it’s impossible unless machinesthat it’s impossible unless machinesshare our perception of the world.share our perception of the world.

A solution to this “symbol groundingA solution to this “symbol groundingproblem” is to give robots with humanproblem” is to give robots with humaninspired senses.inspired senses.

But the world we experience is determined by our But the world we experience is determined by our senses, and human and machine bodies may lead senses, and human and machine bodies may lead to different conceptions of the world (e.g. Nagel’s to different conceptions of the world (e.g. Nagel’s What Is It Like To Be a Bat?What Is It Like To Be a Bat? ) )

Maybe the Semantic Web is a way out of this Maybe the Semantic Web is a way out of this problem?problem?

MIT’s Cog



Solving the symbol grounding Solving the symbol grounding problemproblem

The web may become a common world The web may become a common world that both humans and machines can that both humans and machines can understand.understand.

Confession: the web is more familiar and Confession: the web is more familiar and real to me than much of the real world.real to me than much of the real world.

Physical objects can be tagged with low Physical objects can be tagged with low cost (e.g., $0.05) transponders or RFIDs cost (e.g., $0.05) transponders or RFIDs encoding their URIsencoding their URIs See HP’s Cooltown projectSee HP’s Cooltown project

http://cooltown.com/http://cooltown.com/



Rethinking the agent communication Rethinking the agent communication paradigmparadigm

Much multi-agent systems work is grounded Much multi-agent systems work is grounded in Agent Communication Languages (e.g., in Agent Communication Languages (e.g., KQML, FIPA) and associated software KQML, FIPA) and associated software infrastructure.infrastructure. This paradigm was articulated ~1990, about the This paradigm was articulated ~1990, about the

same time as the WWW was developed.same time as the WWW was developed. Our MAS approach has not yet left the laboratory Our MAS approach has not yet left the laboratory

yet the Web has changed the world.yet the Web has changed the world. Maybe we should try something different?Maybe we should try something different?

The communication MAS paradigm has been peer-The communication MAS paradigm has been peer-to-peer message oriented communication to-peer message oriented communication mediated by brokers and facilitators -- an mediated by brokers and facilitators -- an approach inherited from client-server systems.approach inherited from client-server systems.



Rethinking the agent communication Rethinking the agent communication paradigmparadigm

A possible new paradigm?A possible new paradigm? Agents “publish” beliefs, requests, and Agents “publish” beliefs, requests, and

other “speech acts” on web pages.other “speech acts” on web pages. Brokers “search” for and “index” published Brokers “search” for and “index” published

contentcontent Agents “discover” what peers have Agents “discover” what peers have

published on the web and browse for more published on the web and browse for more detailsdetails

Agents “speak for” content on web pages byAgents “speak for” content on web pages by Answering queries about themAnswering queries about them Accepting comments and assertions about themAccepting comments and assertions about them



How do we get there from here?How do we get there from here?

This semantic web emphasizes ontologies This semantic web emphasizes ontologies – their development, use, mediation, – their development, use, mediation, evolution, etc.evolution, etc.

It will take some time to really deliver on It will take some time to really deliver on the agent paradigm, either on the Internet the agent paradigm, either on the Internet or in a pervasive computing environment.or in a pervasive computing environment.

The development of complex systems is The development of complex systems is basically an evolutionary process.basically an evolutionary process.

Random search carried out by tens of Random search carried out by tens of thousands of researchers, developers and thousands of researchers, developers and graduate students.graduate students.



Climbing Mount ImprobableClimbing Mount Improbable

““The sheer height of the The sheer height of the peak doesn't matter, so peak doesn't matter, so long as you don't try to long as you don't try to scale it in a single scale it in a single bound. Locate the mildly bound. Locate the mildly sloping path and, if you sloping path and, if you have unlimited time, the have unlimited time, the ascent is only as ascent is only as formidable as the next formidable as the next step.”step.” -- Richard -- Richard Dawkins, Climbing Dawkins, Climbing Mount Improbable, Mount Improbable, Penguin Books, 1996.Penguin Books, 1996.



The Evolution of Useful ThingsThe Evolution of Useful Things

The Evolution of Useful The Evolution of Useful ThingsThings, Henry Petroski, , Henry Petroski, 1994.1994.

Prior to the 1890’s, papers Prior to the 1890’s, papers were held together with were held together with straight pens.straight pens.

The development of “spring The development of “spring steel” allowed the invention steel” allowed the invention of the paper clip in 1899.of the paper clip in 1899.

It took about It took about 25 years (!)25 years (!) for for the evolution of the modern the evolution of the modern “gem paperclip”, “gem paperclip”, considered to be optimal for considered to be optimal for general use.general use.



So, we should …So, we should …

Start with the simple and move toward the Start with the simple and move toward the complexcomplex E.g., from vocabularies to FOL theoriesE.g., from vocabularies to FOL theories

Allow many ontologies to bloomAllow many ontologies to bloom Let natural evolutionary processes select the most Let natural evolutionary processes select the most

useful as common consensus ontologies.useful as common consensus ontologies. Support diversity in ontologiesSupport diversity in ontologies

Monocultures are unstableMonocultures are unstable There should be no THE ONTOLOGY FOR X.There should be no THE ONTOLOGY FOR X.

The evolution of powerful, machine readable The evolution of powerful, machine readable ontologies will take many years, maybe ontologies will take many years, maybe generationsgenerations Incremental benefits will more than pay for effortIncremental benefits will more than pay for effort



For more informationFor more information On our work at UMBCOn our work at UMBC

http://research.ebiquity.org/http://research.ebiquity.org/ ITTALKSITTALKS

http://ittalks.org/http://ittalks.org/ TAGATAGA

http://taga.umbc.edu/http://taga.umbc.edu/





(3)(3) Enhancing Bluetooth’s Enhancing Bluetooth’sService Discovery ProtocolService Discovery Protocol

Bluetooth’s SDP is very simpleBluetooth’s SDP is very simple Services and attributes represented Services and attributes represented

by UUIDs which are 128 bit numbers!by UUIDs which are 128 bit numbers! No registration, aggregation, multicasting, No registration, aggregation, multicasting,

event notificationevent notification Enhanced SDP uses DAML+OILEnhanced SDP uses DAML+OIL

We assume at least one resource rich device in We assume at least one resource rich device in the ad hoc network to serve as a matchmakerthe ad hoc network to serve as a matchmaker

Services and attributes described in DAML Services and attributes described in DAML using a “standard” ontologyusing a “standard” ontology

All available information from service and All available information from service and attribute descriptions used for matchingattribute descriptions used for matching

Reasons to obtain Reasons to obtain closestclosest match match

Service

AdHocNetworkService

rdfs:subClassof

OperatingSystem

ProvidedBy

ServiceCost

rdf:Property

rdf:P

rope

rty

ContactURI

ProviderName

NetworkTechnology

PrinterService

rdfs:subClassOf

PrintColorQuality

PrintFileType

PrintSpeed

PrintTechnology

PrintResolution

PrintOutputFormat

Printer

PrinterModel

Priority Value

rdf:Property

rdf:Property rdf:Property

rdfs:su

bC

lassO

f

Priority Value



Intrusion DetectionIntrusion Detection

Jeffrey UndercofferJeffrey Undercoffer Distributed Intrusion Detection ServicesDistributed Intrusion Detection Services

Data Mining over low level attributes at the following Data Mining over low level attributes at the following levels:levels:

ProcessProcess SystemSystem NetworkNetwork

Secure framework for inter process communicationsSecure framework for inter process communications Define a “Victim-Centric Ontology”Define a “Victim-Centric Ontology”

Formal theorem prover to reason over instances of the Formal theorem prover to reason over instances of the ontologyontology

Develop a statistical model, using the Kullback-Develop a statistical model, using the Kullback-Liebler dissimilarity measure, for analyzing streams Liebler dissimilarity measure, for analyzing streams of system callsof system calls



MoGATUMoGATU

Filip PerichFilip Perich Addressing the issues of data Addressing the issues of data

management in pervasive management in pervasive computing environmentscomputing environments

Use semantic language for Use semantic language for annotating data and providersannotating data and providers

Discovery of data and data Discovery of data and data sourcessources

Data dissemination modelsData dissemination models Collaboration between devices Collaboration between devices

for processing queries over for processing queries over multiple streams (network joins)multiple streams (network joins)

Transactional support for Transactional support for preserving (local) consistency preserving (local) consistency state state

Profiles for enabling devices to Profiles for enabling devices to proactively acquire and cache proactively acquire and cache data for future usedata for future use

Supported by NSF and DARPA Supported by NSF and DARPA awardsawards

InformationGenerator Y

InformationProvider X

InforMa

InformationProvider and

InstanceMetadata

RoutingInformation

Bluetooth.Interface

User.Interface

802.11b.Interface

http://mogatu.umbc.eduhttp://mogatu.umbc.edu



Service Discovery and Composition Service Discovery and Composition

Dipanjan ChakrabortyDipanjan Chakraborty Develop a peer-to-peer caching based Develop a peer-to-peer caching based

distributed service discovery mechanismdistributed service discovery mechanism Caching of “neighboring services”Caching of “neighboring services” Selective forwarding of requestsSelective forwarding of requests

Broker-based Service CompositionBroker-based Service Composition Dynamic Broker selection based mechanismDynamic Broker selection based mechanism Distributed Broker-based mechanismDistributed Broker-based mechanism Utilizes the peer-to-peer service discovery Utilizes the peer-to-peer service discovery

layerlayer Source-monitored fault-toleranceSource-monitored fault-tolerance

1 Intelligent Information Systems on the Web and in the Aether Tim Finin University of Maryland Baltimore County March 21, 2003 finin/talks/swig03.ppt.

Documents

daml slide

daml people

daml talkont

daml calendar

daml markup

daml topics

daml profileont

daml topicont